• Faceoffs, defencemen and Corsi

    by  • July 28, 2009 • Uncategorized • 38 Comments

    There was a post at The Puck Stops Here that was excerpted in a thread at Hockey’s Future that I thought warranted a bit of comment, as it touched on mc79hockey fave Dan Heyda. I’m going to excerpt a rather large part of it but go read the whole thing:

    It is clear that when Jan Hejda is on the ice, his team outscores their opponents. It is also clear that his team is outshot significantly in the process. The logical conclusion is that Hejda forces opponents to take bad shots instead of creating good offensive opportunities. This is consistent with observation of his play.

    Hejda highlights a problem with Corsi. It does not distinguish between bad shots and good shots. A blocked low percentage shot counts the same as a goal. He is the most significant example of a player in the NHL who has successfully forced lots of bad shots in place of a lesser number of good shots as a defensive strategy. He is an example that shows that Corsi Numbers of defencemen with little offensive talent should be taken with a grain of salt – especially when they disagree with +/- ratings.

    Jan Hejda is a good defenceman. His +/- shows that he is effective in limiting scoring when he is on the ice. The fact that he does this by forcing low percentage shots does not hurt his playing level. His Corsi is misleading. He is the best example of that in the NHL. It is reasonable to be skeptical of the Corsi Ratings of other defensive players with low offensive value. In cases where their Corsi Rating and +/- disagree, the +/- is more likely to be the meaningful one.

    I enjoy The Puck Stops Here and thought that the author, whoever he might be, was doing the Lord’s work in the playoffs when he was trying to convince red clad heathens that Chris Osgood was not, in fact, some sort of a playoff force and Hall of Fame level talent. This post above strikes me as being just crazy though.

    To start with, I don’t know how he concludes that Hejda was forcing low percentage shots. Maybe it’s true, maybe it’s not, but the Jackets had a shooting percentage of 10.4% when Hejda was on the ice last year and shot 8.0% when he wasn’t on the ice. I am likely amongst the 250 or so biggest Hejda fans in the world and if we were all assembled in one place, I don’t think that any of us would be crediting him with nudging the Jackets’ shooting percentage by 2.4 percentage points. If the Jackets shoot 8% with him on the ice, he falls from +14 at 5v5 to even (all stats courtesy of timeonice), which seems to kind of gut TPSH’s position.

    Now, I’m a Hejda fan. So how can I overlook his Corsi? Well, out of the 186 defencemen who were on the ice for at least 500 draws, Hejda had the fifth toughest ratio of D/O zone draws. I put together a chart graphing D/O zone draw ratio against Corsi ratio:

    DOCorsi

    The correlation of Corsi to D/O faceoff ratio for defencemen who were on the ice for at least 500 draws is -0.63. Of the 42 guys with a ratio of D/O faceoffs of 1.2 or better (Hejda was at 1.3), 3 of them had a positive Corsi. Nobody with a D/O faceoff ratio as high as Hejda’s posted a positive Corsi.

    One of the axioms of the hockey staterati has been that context is everything. That applies to the new fangled stats too – you need to check the percentages and other factors for the guys at the extremes in order to get a real idea of why the results were what they were.

    About

    38 Responses to Faceoffs, defencemen and Corsi

    1. R O
      July 28, 2009 at

      I had to learn how to contextualize the hard way too. I think a big problem is that people look at Corsi (or shots +/-) and think along the lines of “oh, a shots metric, all well and good but how about shot quality???”. And they are totally missing the forest for the trees. None of us care about the shots, we just use it as a proxy for territorial advantage. And when you step back and look at it like that, then the next logical question is “what part of the ice does the coach make the player start in the most?”

    2. R O
      July 28, 2009 at

      The last sentence should read:

      And when you step back and look at it like that, then the next logical question is “what part of the ice does the coach make the player start in the most?” and then work from there to see how that player, through his efforts, influenced the tilt of the ice from the starting point that the coach dictated.

    3. July 28, 2009 at

      It’s pretty clear now that PuckStopsHere doesn’t really understand the implications of the numbers and the context behind them.

      He needs to spend some time just shuffling a bunch of advanced stats around to get a feel for them.

    4. July 28, 2009 at

      He’s had his blog for a long time, this (and the item CoachPB brings up) is symptomatic of his whole body of work. Very thoughtful guy, wants to contribute original thought (and sometimes does), but he quite obviously does not read enough other hockey blogs.

      I hate to bash a guy who’s writing about hockey on his own time for fun, so I won’t really, but his pieces would be a lot better if he was more engaged with the rest of the hockeysphere. He’s clearly smart enough, he just doesn’t read enough.

    5. Vic Ferrari
      July 28, 2009 at

      Wow, the picture tells the story. Cool stuff.

      Who are the guys with Def/Off zone faceoff starts at 2.5 and 2.9? Jebus, somebody really should send them a card.

      Pretty impressive corsi ratio considering the starting point, especially the 2.9 guy. Obviously they play for terrible teams, and are a bunch better than their defender teammates. They must have been on the ice for more EV defensive zone draws than DET had as a team all season, no?

      These are the kind of guys that you hope your GM will try and acquire. But as Oiler fans, we know damn well that these are exactly the kind of players that this management team wouldn’t touch with a ten foot pole.

      Sigh.

    6. July 28, 2009 at

      Vic, they have to be Nick Schultz and Kurt Sauer.

    7. July 28, 2009 at

      I hate to bash a guy who’s writing about hockey on his own time for fun, so I won’t really, but his pieces would be a lot better if he was more engaged with the rest of the hockeysphere. He’s clearly smart enough, he just doesn’t read enough.

      I don’t think he fiddles enough either. I had an intern :) script all of vic’s ZS, ZE and ZSh as well as the Corsi, Fenwick PDO stuff + Desjardins stuff to a DB and when I get bored I fiddle. I don’t try to do the advanced maths because I’m not capable, but I go hunting for causation, correlation and interesting tidbits.

    8. mc79hockey
      July 29, 2009 at

      Matt and Coach are probably right, although I want to also mention that I think he does do some interesting stuff that doesn’t require fiddling with the numbers, like the bit about the Russians that MacLean picked up on.

      Coach is right: Sauer and Schultz. Gretzky and Lemaire really ran their D tight. Sauer/Michaelek paid the price for Jovo, Yandle and Morris in PHX. Schutlz, Skoula and Johnsson did so for Bergeron and Zidlicky in Minnesota. Tough to argue with either of those.

      This stuff is great data, by the way Vic. Just fascinating stuff. Thanks for the work that you put into TOI. I think about where we were in terms of information when you, me and RiversQ were kicking this stuff around back in 2003 and it’s come an awful long way.

    9. July 29, 2009 at

      The issue that your post misses is how can Jan Hejda have a top +/- and a poor Corsi simultaneously. Its not a fluke. Essentially the same thing happened last year.

      His team allows a lot more shots than they take with him on the ice and scores more goals than they allow with him on the ice.

      How can that be? I think the only logical answer comes in shot quality. Hejda is forcing bad shots.

      Where faceoffs occur with Hejda on the ice explains the poor Corsi, but not the good +/- he has at the same time.

    10. July 29, 2009 at

      He addressed the plus-minus in the second paragraph. Hejda had an absurdly high on-ice SH% that, if you regress it to team average (which is perfectly fair given that you’d be hard pressed to argue Hejda is driving high percentage shots), makes up all of his excess goals-for at 5v5.

      His SV% is mildly above average at .922 vs a team average of .917, but if that’s evidence of his impact on shot quality then you’d be hoping to sign Klesla and Backman who put up on-ice SV% numbers in the mid-.930s. I have a hard time believing Klesla and Backman are superior to Hejda as defensemen, so to each his own.

    11. July 29, 2009 at

      Saves percentage is only part of the story. When Hejda was on the ice, his opponents had a lot of missed shots and he was one of the best shot blockers on the Blue Jackets. That is missed entirely in the analysis so far. Those both explain where the missing low quality shots may have gone.

      It is true that Hejda was on the ice for 7 short handed goals (and that led the Blue Jackets). That does account in part for +/- difference from his Corsi. Hpwever, even removing those goals, Hejda has the worst Corsi and best +/- on his team.

      There is something to that discrepancy. His team allowed a tremendous number of shots when he was on the ice and still outscored their opponents. The missed and blocked shots account for a lot of that. That is forcing your opponents to take worse shots.

    12. July 29, 2009 at

      Sorry I misspoke. Hejda has the second worst Corsi on the team (Mike Commodore is worse).

      And I am Greg Ballentine. I didn’t realize that I have my name listed differently on different computers. Sorry if that causes confusion

    13. mc79hockey
      July 29, 2009 at

      Greg – by my math, the ratios suggest that there were 36 additional blocked/missed shots with Hejda on the ice. If those shots have a normal chance of going in, we’re talking about 3 goals. If, as you’re suggesting, those were low percentage shots, then maybe it’s less.

      Not playing on the PP but playing tons SH is going to drive his +/- up. So is having a shooting percentage that’s 30% better than team average.

    14. July 29, 2009 at

      Greg – I don’t know if you’ve read the work I’ve listed here but if not…you should.

    15. July 29, 2009 at

      I get 92 extra missed and blocked shots when he is on the ice shots from the behind the net numbers (82GP*17.31/60mins*(1.7diffin misses+2.2diff in blocks)). And when Hejda is not on the ice (82GP*28.29/60min*(-2.2-1.6))=-147. The team goes from one that blocks a lot of shots and forces their opponents to miss the other way around.

      That is a big swing. It is that difference that makes for a bad Corsi and a simultaneous good +/-.

      I think there is something real there. It happened the same way last year. It happens to Jan Hejda more than to any other NHL player. Why? If you want to claim the difference is all usage, you have to claim nobody else gets used the way Hejda does in the NHL.

    16. mc79hockey
      July 29, 2009 at

      Greg – I’ll look at your point tonight. To be clear, I’m not attributing the difference to usage. This year, I’m saying it was shooting percentage and shorthanded goals.

    17. MattM
      July 29, 2009 at

      Hey Coach, what kind of DB is that and how big is it? Fiddling seems like fun and I don’t have an intern. Is it the sort of thing you could put up for download somewhere?

    18. July 29, 2009 at

      In 2007/08, Hejda had the best +/- on the team and the worst Corsi among the team’s defencemen. Whatever is going on is clearly repeatable.

    19. July 29, 2009 at

      Hejda, in 07-08, had an on-ice SH% of 8.4% which was the best on the team (among players above 40 GP) and 1.7 percentage points higher than the team average.

      What’s going on is that Jan Hejda is one of the greatest offensive defensemen of our generation. He’s like Scott Niedermeyer, only without a name I constantly mix up the ‘i-e’ on and a lot of announcers who wax poetic about how he’s a like a fourth forward or rover out there!

      He also had the second best on-ice SV% among eligible players, so that at least matches up with your ideas, unlike 08-09.

    20. July 29, 2009 at

      Quain.

      Your argument is that is essentially a fluke that happened twice in a row?

    21. Vic Ferrari
      July 29, 2009 at

      I won’t speak for Quain, but if that is his argument, and I suspect that it is … it is highly likely that it is the correct argument.

      The repeatability of on-ice EVsave% behind a player, relative to his teammates, is virtually nil. (That’s the stat that helped Hejda in 0708, though as MC79 rightly points out SH and epty net +s flatter defensive DMen in the newspaper +/- as well).

      The repeatability of on-ice EVshooting% ahead of a player, relative to his teammates, is quite small as well. And absolutely minute for defensemen. (That’s the stat that helped Hejda in 0809, and as MC79 rightly points out, anyone who thinks that Hejda was driving EVshooting% has surely been smoking the drapes. And I’m one of the guys in the Dan Hejda fan club, by the by).

      Some guys are “lucky” two years in a row because the universe requires that a certain number must be. The dice have no memory, after all.

    22. July 29, 2009 at

      My argument is that 08-09 isn’t even a shot quality issue. His on ice SV% was essentially team average; he may have blocked a bunch of shots, but I’m of the opinion that a blocked shot has no ‘quality’ to it, so I don’t think it belongs in a discussion on his ability to impact opposing shooting percentages. His entire plus/minus is explained by higher than average SH% and EN/SH goals (I can never remember if Vic’s app filters out EN or not).

      In 07-08 he had a well higher than team average on ice SV%, which at least doubles as evidence that he can impact shot quality. I don’t actually believe it, but it’s better than 08-09.

      In truth? I believe one of three things are occuring:

      1) Jan Hejda is an offensive dynamo who should be expected, from here on out, to have a well above team average Team SH% when out on the ice.

      2) When Jan Hejda is out on the ice, in the offensive zone, he is almost always paired with above average finishers, so you’d expect him to have a higher on ice SH%.

      3) It’s a fluke. Two good rolls of the dice. Magic. Withcraft. Whichever works.

      I’m probably voting #3, but I’m open to evidence of the other two, or even #4. But if #4 is ‘he impacts shot quality’ I need someone to explain to me why his SV% is so middle of the road.

    23. July 29, 2009 at

      Number one is a ridiculous strawman.

      Number two is pretty ridiculous too. If he plays in defensive situations etc. you dont pair him with your best scorers.

      Number three is possible

      Number four is what I argue. And impacting shot quality is being misinterpreted by you. When Hejda is on the ice there is a noticable increase in blocked and missed shots from when he is not. This is how he impacts shot quality. It is the “shots” Corsi counts that saves percentages do not where his effect can be seen.

    24. July 29, 2009 at

      We are seeing two repeatable trends. Jan Hejda bad Corsi and Jan Hejda good +/- simultaneously. He is the player in the league with the biggest spread between the two tis year (significantly) and quite possibly last year as well (though I havent played with the data as much to be confident in that).

      I am pretty confident THAT means something.

    25. Vic Ferrari
      July 29, 2009 at

      Greg

      For all the badness associated with starting the lion’s share of your shifts in your own end, facing a lot of odd man rushes isn’t one of them.

      That’s why, by and large, guys who are on the ice for an extra bit of blocked and missed shots … they should also see an ever so slightly higher EVsave% behind them. Though you’d have to look at the aggregate of a big group of defenders from this category, because it’s a small amount. We’re painting with a fiene brush here.

      I’ve never checked this btw, but I would bet money on it. I mean it’s never favourite to start in your own end, and the badness outweighs the positive effects by a mile, but at the very least everyone on your team is in proper position to start.

    26. Vic Ferrari
      July 29, 2009 at

      You know Greg, I have no trouble with people who don’t value territorial advantage in hockey. None whatsoever.

      If you google “corsi numbers” AND NHL OR hockey … anything like that. You’ll see that the concept has been missed, but that the stat itself really hasn’t fared that badly. I assumed it would have been bashed. Not that I care either way.

      The place it takes an absolute pounding is on Oiler fan message boards. And the Oiler related blogs are, by and large, the place it has been most widely embraced as a measure of territorial advantage that (just like every other stat in hockey) cannot be divorced from context.

      This stratification of the Oiler internet fanbase … I don’t know how it happened, but it did. And it is probably a healthy thing.

      The Hockey Sabermetricians outside of this corner of the web are embracing Desjardin’s shot quality work and largely disregarding his QualComp stuff. (As maybe they should, did you know QualComp correlates strongly to points/60? Damn, it must mean far less than zero, no?) And faceoff zone stats are considered pure black magic.

      And that’s fine too. I have no problem with everyone sticking to the stuff they believe in. IMHO, we should all carry on as we were.

    27. July 29, 2009 at

      Vic

      I am not arguing that Hejda doesn’t start a lot of shifts in his own zone or that doesn’t hurt his Corsi. It also should hurt his +/-. Clearly it doesn’t. He has led the team both of the last two years.

      I am saying that high +/- coupled repeatably with a low Corsi is meaningful.

    28. mc79hockey
      July 29, 2009 at

      Greg – I think that you need to clarify your terminology here. Are you talking about +/- or even strength +/-? It seems to me that the most relevant issue is ES +/-; regular +/- is polluted by shorties. Can you confirm for me what you’re talking about?

      Also – do you agree that empty netters should be pulled from the discussion?

    29. July 29, 2009 at

      It doesn’t matter which +/- I use. Hejda leads Columbus in any case. He is the top +/- with or without shorthanded goals included.

      It is most meaningful to talk about 5 on 5 situations for his +/- since his Corsi is in 5 on 5 situations.

      Yes ES and if you can remove empty net goals (I cannot easily do that – but I am assuming its not a huge factor) that makes the comparison “apples to apples”.

    30. MattM
      July 29, 2009 at

      Greg,

      To be clear. Your contention here is that Hejda’s strong +/- and poor Corsi suggest that he impacts the quality of shots taken by his opposition by causing them to miss or be blocked in a larger portion, leading to his strong +/-.

      I don’t have the resources at my fingertips for this, but when doing an analysis of ability to play good defense while your opponent has possession, it seems to me that dropping off the “+” half of the equation would be a logical first step. It really has nothing to do with the argument. Do the same with his Corsi. Divide and now we’ve basically just got the volume of goals against per shot attempted against him, or the other way and get roughly the number of attempted shots it takes to score when Hejda is on the ice. If he’s forcing weaker shots consistently, then it should take more attempted shots to score on him, so this ratio should be noticeably better for him than for other dmen, I would think. Or at least, a random sampling of dmen should show that this ratio is better for guys who we would qualitatively assess as better defensive defensemen?

      Does that make sense? Can we test that?

    31. July 29, 2009 at

      It seems like a reasonable test with several caveats, the biggest being that if you compare between different teams you are to some degree measuring the quality of their goaltending.

    32. PDO
      July 29, 2009 at

      The Hockey Sabermetricians outside of this corner of the web are embracing Desjardin’s shot quality work and largely disregarding his QualComp stuff. (As maybe they should, did you know QualComp correlates strongly to points/60? Damn, it must mean far less than zero, no?) And faceoff zone stats are considered pure black magic.

      Vic:

      I’d bet good, good money you would see a lot of goalies in that group of people.

    33. Quain
      July 29, 2009 at

      Matt’s test essentially just measures who had the best on-ice SV%. I ran it, total negative Corsi events divided by GA, and Backman/Klesla/Russel outpace Hejda by a fair margin entirely on the basis of Mason stopping more pucks when they’re on the ice.

      If you run it for 07-08 I’d wager that Hejda would be near the top on the basis of an excellent on-ice SV% in that season. His great plus-minus for the past two years is driven nearly entirely by offense.

    34. July 30, 2009 at

      This seems like a good BAMBi situation if I’ve ever seen one.

    35. MattM
      July 30, 2009 at

      Thanks Quain. That’s what I figured would happen. Based on Vic’s earlier work about players having little impact on the sv% behind them, I think this suggests quite clearly that the arguments you guys have put forward to explain Hejda’s stats (luck and context) are correct. Or at least, more correct than the idea that he’s drastically better at limiting shot quality than his peers.

    36. August 6, 2009 at

      If my memory is correct Greg Maddux had two seasons in which he lead all of MLB with a ERA below 2.00, in both seasons Maddux benefited from a lower than average BABIP (Batting Average on Balls i Play is largely a random variable). What does this mean?

      a) Greg Maddux was a great pitcher.
      b) Greg Maddux was great and also a little lucky.
      c) Greg Maddux was great and a little lucky two year consecutive years.

      I believe that c) is the correct answer.

      There is nothing wrong with saying that Hejda is a great defender who was also be a bit lucky two years in a row.

    37. August 22, 2009 at

      I always find the sportsnet scout reports to be very entertaining. Many Oilers fans have been hoping for another dman to fall of the Hejda tree again.

      Here’s the current sportsnet scout reports for Hejda which is pretty funny in not completely inaccurate:

      Assets: “Has big-league size and a wealth of experience. Moves the puck well and is aggressive in the defensive zone.”

      Weakness: “Has limited upside and is basically a fill-in player at this stage of his career. Won’t be much of a factor on the score sheet in the big league.”

      Career potential: “Reserve defenseman.”

    38. The Friar
      December 21, 2010 at

      The correlation of Corsi to D/O faceoff ratio for defencemen who were on the ice for at least 500 draws is -0.63.

      The coefficient of correlation is not terribly appropriately used here, because the relationship between D/O faceoff ratio and Corsi is nonlinear. Leave Corsi on its natural scale from negative to positive infinity, take the log transform of the D/O faceoff ratio to put that on a minus-infinity-to-infinity scale as well, and, with the relationship thus linearized, the correlation coefficient should be seen to be even higher. In truth, these results are actually kind of damning of the Corsi value as explanatory, since probably more than half the variation in them (appropriately scaled) can be explained by zone starts.

    Leave a Reply

    Your email address will not be published. Required fields are marked *