Cherry to Corsi: ‘Get off my lawn’

by Colby Cosh

http://www.youtube.com/watch?v=JCaWBFAJL44#t=4m58

It must be 2010; I’m watching Don Cherry talk about Corsi numbers on TV [fast-forward to 5:00 in the video above]. Granted, he’s denouncing them, but a) he has a couple of good points, and b) that’s what old guys do when they’re confronted with statistical innovations. Read your Kuhn.

Ron Maclean didn’t do a very good job of explaining the Corsi stat (yes, it was invented by Jim Corsi), and he picked a slightly inopportune occasion to bring it up with Grapes sitting next to him. As this Globe & Mail primer explains, Corsis are essentially a more powerful extension of the plus-minus stat you see in the newspaper; they count not only the goals for and against while the player is on the ice at even strength, but all shots directed at the net either way (goals, shots on goal, missed shots, and blocked shots).

Everybody knows plus-minus isn’t a very robust or accurate way of measuring a player’s contribution, and Corsi numbers mitigate some of the disadvantages of only counting goals. You’re counting a lot more events per game—scoring chances, loosely speaking–which gives you more statistical power and leaves luck and contextual factors with less relative influence on the stat. You’re also factoring out the quality of the goaltending behind (and in front of) a skater.

That doesn’t mean Corsis are a perfect means of understanding or isolating a player’s contribution. Shifts in hockey aren’t like a batting order, in which everyone must take his turn. Some players are out there with inferior teammates, some players are shielded from the toughest competition, and some players provide value just by chewing up a lot of minutes. Context is important, and in hockey we may never be able to correct advanced stats for context as well as we can for hitters in baseball. (That’s why stats in hockey aren’t very advanced. We’ve really only just gotten around to expressing events as rates in the simplest possible way. The guy who did this for baseball was born 186 years ago.)

The biggest easily-measurable influence on Corsi numbers—easily measurable thanks to the work of Gabe Desjardins—is where a player tends to start his shifts on faceoffs. A guy who is rolled out for a lot of defensive draws is going to have a worse Corsi rating through no fault of his own—indeed, he is penalized for being trusted by his coach. In that sense, Ryan Johnson was a bad choice for Maclean to pick on, and Don Cherry’s outburst of skepticism was entirely appropriate. Desjardins’ site also tracks “zone starts”, so we know that Johnson, who has the league’s worst Corsi rating, is one of the league’s most disadvantaged regular skaters zone-wise. Through the games of March 28 he’s been sent out for only 78 offensive-zone faceoffs but 165 in his own end. Which is why he’s near the very bottom of this list.

Like Desjardins himself, I am less impressed by the subtly different argument actually made by Cherry—that it’s unfair to penalize Johnson for blocked-shots-against that he himself has blocked. Insofar as Corsi numbers are measuring any ability, it’s the ability to not have to block shots in the first place—to help your team promote the puck out of your end and into the enemy’s defensive zone. The counting of blocked shots has a problem similar to the counting of double plays turned by a team in baseball; they correlate negatively, if at all, with the winning of games. An individual blocked shot might have a positive value—though even that’s certainly not true in every case—and you want players who are willing to block them. But it’s better not to give up lots of opportunities for blocked shots.

And, hell, it’s better still not to be a low-talent, high-grit player who has to block them to keep a job. Even Don Cherry knows that.

Cherry to Corsi: ‘Get off my lawn’

    • Cherry-friendliness isn't exactly a killer argument for "Fenwick numbers", though (they should probably have been called "Fenwick-ized Corsis"), and I believe it's been shown that taking out the blocks doesn't add information on the team level. It shouldn't; blocks are, broadly, a sign of a bad team. I wasn't really hip to the argument when you were having it but the blocks really ought to be in there. IF Johnson were actually better he'd have a lot fewer OF them (which is not true of the more serious zone-start problem).

      • I'm not sure total blocked shots really correlate all that much to how good or bad a team is. Pittsburgh is 8th in total blocked shots, while Anaheim has the 2nd fewest, and sit in 12th in the West…

        • I've checked this in the past, and as I recall you're not wrong–it's a negative correlation but it's feeble. I'm looking around now for team-level blocked-shots figures so I can do a scatter diagram (I've been teaching myself R, it'll be good exercise). If you get a minute send the URL for those to me at colbycosh@gmail.com.

          • Are those blocked shot totals, Colby? You should see a noticeable negative relationship with goal scoring at even strength and a positive relationship on the PK. Same goes for missed shots.

          • I can’t reproduce your plot.

            (1) Are you sure you’re not plotting total hits instead of total blocked shots? Your x axis goes up to 2400 or so but the total blocked shots tend to be around ~1000. Since it’s the next column over (on the NHL pages, anyway) I think it’s an off-by-one error in your column indexing.

            (2) I don’t seem to get the same plot as yours even when I plot hits (although there’s enough in common that I think I must be close.) For example, looks like there are only two teams (2007 COL and 2007 EDM) which had fewer than 1200 total hits (1168 and 1130). Your plot has one with less than 1000. So I’m still missing something. Could equally be a problem on my end, I can never correct anything without introducing new errors of my own.

            My version of your plot shows a stronger correlation. If I plot total blocked shots against win percentage or point total, I find that there’s a weak but noticeable trend: more blocked shots correlates with doing worse, although the very best teams seem to have close to the median number of blocked shots.

          • "Off-by-one error in column indexing" is the exact correct diagnosis. Well done, thank you! The new diagram is up at the same URL, let me know if it reproduces OK.

          • Yep, I can match the new version of the plot. (The other differences were just the choice of 2006 season vs 2009-to-date.)

      • I basically agree with the responses to Cherry's objections, but I still maintain that, big picture:
        - the idea is to have a number that represents scoring chances for/against better than any other
        - shots that are blocked, by you or your opponent, are a lot less obviously chances than saved shots or misses

        Semi off-topic, but the last time anyone counted, the best predictor of future EV outscoring was not where you started your shifts OR any variation of the Corsi metric, but rather where your shifts end. Somehow this is the only one not published by Desjardins, nor referenced on a regular basis in the hockeysphere.

        • I have to take back the "blocks are a sign of a bad team" thing anyway; they aren't much of one. To the degree that Corsi represents time-of-possession or zone-advancement rather than scoring chances per se, your big picture has a weakness.

          • I know there's some folks who think well-recorded time-of-possession stats would be an upgrade on Corsi, but I don't see it. Corsi (with blocks excluded IMO) basically does represent time-of-possession in the offensive zone, with an adjustment reflecting who is accomplishing more with that time. Why go sideways, or backwards?

          • OK, but full Corsi is known to be a better proxy for actual zone time than Fenwickized Corsi. You are going backwards, very slightly, when you take out the blocks.

        • It's harder to extract from the gamesheets.

          There's also the issue of the TOI charts sometimes having some pretty crazy errors. Rewinding one second from a faceoff doesn't guarantee that you'll get the players who ended their shifts in those faceoffs.

          Although the same can be said for any of the events in the PBP logs, there are occasionally the egregious error in which players were on the ice.

  1. A guy like Johnson, who can't move the puck out of his own end when he starts most of his shifts there, well, you can point to the coach as giving him tough minutes but he is almost surely doing some of it to himself through icings. Especially when the team he plays for has another guy in Kesler who is on another level as far as heavy lifting goes?

    • I agree, though taking twice as many draws in your end as in the O-zone is "heavy lifting" no matter how you slice it.

      • Yeah, as far as its effect on your counting numbers and underliers and overall "how you look", his shifts are killer.

        Cherry's argument is a bit asinine though, even if he did know about these numbers a tiny bit of common sense would tell him that no coach would really trust Johnson that much. He and his linemates force the coach's hand.

  2. You are all nerds.

  3. I feel like I'm reading matchsticksandgasoline not Macleans. Nice change though.

  4. Colby, I watched last night a movie called "State of Play" with Russell Crowe and he stole your look…

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