Sunday, October 31, 2010

Roy MacGregor Hates Corsi

Unfortunately, it looks true. Here's his article. Sigh...

Here's a response. Not great, but just what I can come up with right now.

May I first remind you that Buffalo Sabres goaltending coach Jim Corsi (or maybe even the stats keepers at the 1972 Summit Series) first started tracking the statistic. So apparently someone important though it was worth something.

But neither compares to the incessant baseball-ization of the game that has become even more absurd than fans dressed as hockey pucks bouncing off each other in a humiliating race around the rink.

So what if it's becoming like baseball? That means hockey will be wildly popular very soon. Numbers help people understand the game.

Perhaps the least helpful statistic, however, is shots, as they can vary as much as those coming from an AK-47 to a BB gun. Coaches tend to prefer counting scoring chances, but even this is a dubious measure as the NHL is loaded with losing coaches who will claim they “outchanced the opposition.”

Actually, that's probably correct. If you're trailing, you attack more and more, and thus generate more scoring chances. It's score effects, purely normal.

As with virtually all hockey stats, the scoring chance defies pure definition.

Which is why bloggers at MC79, En Attendant Les Nordiques, Flames Nation, and other sites try and record them themselves. The NHL is the one trying to objectify the definition and recording.

The definition of a scoring chance is pretty intuitive. If nothing else instead of quantities one can use ratios. I'm sure you can understand that. If a player has poor Corsi but you think he's good then he shouldn't be allowing too many chances. Right?

Scoring chances, however, are the prime evidence backing a story that flew around the Internet this past week: “Early stats show Kovalchuk not worth it.”

Scoring chance != Corsi. Kovalchuk is a Corsi black hole, but Gabriel Desjardins has looked at what makes goal scorers great, and found that taking lots of shots plus getting into good shooting position are the repeatable skills that great goal scorers have. Kovalchuk gets into great shooting positions, and he may have the best shot in the game today.

Now, it might well be argued that the New Jersey Devils – hockey's musk ox of defensive strategy – were the worst possible team for free agent Ilya Kovalchuk to sign with last summer, but surely it is too early to conclude the $100-million (U.S.) was a waste.

No one disagrees here. It is worth noting that the Devils are trying a more offensive system this year under John Maclean, though.

Using a measure called CORSI – oversimplified, a tabulation of all shots directed at both nets while a player is on the ice – the argument went that, a mere handful of games into the season, Kovalchuk was a disaster while two other Devils players – Travis Zajac and Dainius Zubrus (the pure definition of “journeyman”) – were far more valuable to the team.

Context, context, context. Firstly, all three get pretty good Zonestart. Zajac is an elite two-way forward, so there should be no surprise that he tops Kovalchuk. That being said, Kovalchuk's pure scoring ability is immensely valuable as well.

Now, Kovalchuk is a -10.0 Corsi Rel, only ahead of Jamie Langenbrunner pretty much. Yes, that's bad. Quite honestly, it's along the lines of what we should expect. Kovalchuk isn't exactly taking every shift with Zach Parise and Travis Zajac. He's been juggled around a ton, and you can't fault him much if his Corsi isn't that great thus far. Through a sample size of nine games.

Anyways, here are Kovalchuk's Corsis by game:
20007 Dallas -9
20017 Washington 0
20027 Pittsburgh 8
20036 Buffalo -2
20048 Colorado 6
20056 Boston -15
20086 Montreal -9
20100 Buffalo 0 (scratch)
20112 NYR 5
20128 San Jose -8
20142 Anaheim 8
20155 Los Angeles 13

That's -3 total.

And the oversimplified part? Well, it irons out anomalies. We see how closely goalies perform relative to one another--a difference in save percentage of only .03 is the difference between Ryan Miller and Antti Niemi. Hockey players perform within a narrow range of skills. It is a mistake not to adjust--we can't, yet--but that doesn't mean Corsi is absolutely worthless.

Another way of looking at this, of course, might be that the Devils were off to a horrendous start because its best players were stumbling out of the blocks, whereas journeymen were largely performing as journeymen are expected.

Isn't that how you should look at it? I'm confused. You reject Corsi and offer a perfectly viable explanation for why it is as it is.

No one who uses Corsi properly considers it a be-all-end-all. Context is important, as is sample size. A guy like Samuel Pahlsson will never have good Corsi, for example, because he's always used as a shutdown player.

It is, however, likely only a matter of time before the league adds this moot measure to its nightly game sheets, one more “baseball” stat to appease those who fail to comprehend that the best explanation for the game of hockey lies in a slight variation of the phrase “scat happens.”

See, in any one game or group of games, yes. But longer-term controlling the play is sustainable. It's been mathematically shown that a team's Corsi% with the score tied correlates pretty well with overall points at the end of the season, to the tune of 0.5 < r < 0.6 for overall Corsi%. It's higher for score-tied Corsi.

Think of it this way:

If you attack the other team's net more than they attack yours, you're in good shape for winning.

If you attack the other team's net more than they attack yours, you probably have more chances to shoot and take more shots.

If you attack the other team's net more than they attack yours, you probably have more chances to shoot and take more shots, and thus have good Corsi.

It's perfectly intuitive.

Anyways, if it's only a matter of time, then the metric has some value, no?

On the scoreboard.

Sounds like you're playing the "it's all about the wins" card. That tells me you don't really have much of an argument here.

Of course it is. But if you control the play--control Corsi--more than the other teams, you're going to win lots of games, most likely. There are few factors that are even remotely consistent game-to-game--off the top of my head I say goaltending and Corsi, and even those are wildly inconsistent. But over 40, 50 games, they give you a good picture of what might happen. Better than goal plus minus.

Like the Habs upsetting the Caps. Or the Hawks dominating.

Anyways, if the scoreboard was all you need to know, Jeff Schultz is the best defenseman in the NHL. Corsi said it was Keith. Which do you trust?

Don't be a sabermetric-phobe. Just keep in mind the context and limitations of Corsi and other advanced metrics and you'll only enrich your understanding of hockey. People often miss the all-important context part. Only then will you enrich your hockey knowledge to the extent of the rest of us.

Or you can stay in the 20th century. Whatever you want.

No comments:

Post a Comment