Monday, August 11, 2008

Zone Time, Corsi and Correlation to Winning

The title and the graph pretty much say it all. This is just another representation of the data for the 2001/2002 season that was used in the post below.

Even though Zone Time and Corsi tell a bit of a different tale for different teams, as shown in the table in the previous post, both are clearly related to each other. And as you start adding more and more games to the pile, they more strongly relate to team goal differential (goals-for minus goals-against).

If we had faceoff zones for this season, it would follow along almost exactly the same in the chart below. And all three would be closely related, and it would require someone more clever than me with math to prise them apart.

The left axis is correlation squared, or coefficient of determination, as they tell me at Wiki. Since these things are largely independent of shooting% and save%, that's probably not a bad first conservative guess at the contribution to goal differential that possession gives us.


And we haven't really looked at finding ways to build a better possession metric than Corsi by rationally combining in faceoff zones in some way. Offside zones would be gold, because they are obviously very independent of Corsi or faceoff zones, so the math would be right in my wheelhouse (addition).

This number wouldn't go down with additional information, or a means of combining the stuff we do have, it would go up. And looking at the graph, for this season at least, it doesn't look like it's topped out yet. i.e. I suspect that if the season had been 40 games longer we'd be looking at stronger ties between goal diff and Corsi and Zone Time.

Mike Babcock's point is seeming less overstated. Hell just difference in goaltending has to take a big bite out of whatever is left of this apple.

Note * I've shaved off the last four games of the season for everybody here, because the missing data wasn't spread around evenly amongst the teams. I'm just going as far as 78 GP for everyone in the graph above.

9 Comments:

Blogger choppystride said...

And we haven't really looked at finding ways to build a better possession metric than Corsi by rationally combining in faceoff zones in some way.


Perhaps the goal shouldn't be trying to build a metric that mimics possession. Frankly, I never thought that Corsi was purely about measuring possession anyway. If you think about it, it could be a proxy for any one of the following:

1) possession
2) zone time
3) scoring chance

Perhaps the best way to conceptualize it is to simply accept the fact that it's a fusion of those three ideas. Personally, I've always thought of it as some sort of "pressure" index.

If you accept this premise, then the next logical step is to ask: "what other events indicate that pressure is being asserted?". The obvious answer would be: "whenever a faceoff is earned/yielded".

I would simply add these data points to Corsi. By this I mean that the pluses/minuses should be accrued to the skaters who were on the ice the instant before a play stoppage. Of course, redundant events would need to be filtered out (i.e. only one event should be counted when a faceoff is caused by a glove save on a shot on goal). There are probably other issues to be ironed out but this is generally how I would tackle it.

8/12/2008 1:25 pm  
Blogger Vic Ferrari said...

Ya, maybe. Just summing faceoffs & corsi doesn't work worth a damn. I'd probably have to watch what Ron Wilson does in a game for a bit to rationalize why his teams have too much poissession without the corsi to show for it. I mean I know how he runs a bench and generally how he runs the ice, but proper game tape and a lot of free time would probably be needed.

Mostly I just think that Corsi is a great representation of meaningful possession in it's own right. I can appreciate how using ratios, as you always like, that will make the math clean. But it's not the way it happens in a game to my eye, it's the difference that drives it. I mean we could hypothesize about "if I had a team with 12 Kovalenkos and you had a team with 12 Dvoraks ...", but it's never going to happen for practical reasons.

And Corsi is the bomb in terms of repeatability, and that leads to predictive value.

So if you say corsi-schmorsi, load o' crap! You'd take the ass-kicking of a lifetime if you dared to bet against it last year, saying corsi didn't predict a change in fortunes of a teams overall +/- (evens or otherwise) from one half of the season to the next. (24W-6L last year).

No other metric can touch that, that's why I'm loathe to fuck with it too much.

Having said all of that, I take your point.

8/12/2008 3:49 pm  
Blogger PerformanceOil said...

saying corsi didn't predict a change in fortunes of a teams overall +/- (evens or otherwise) from one half of the season t

What was your methodology for determining this?

8/12/2008 5:37 pm  
Blogger choppystride said...

I take your points too.

But please don't get me wrong. I'm not one of those Corsi-bashers. In fact, I even suggested looking at this shots-directed-at-net stuff before knowing that it was supposed to be called "corsi":

http://hfboards.com/showthread.php?t=193200#12 (post #12)

And the reason why I requested a ratio check is that I thought you were merely comparing it to plain old "possession". But having re-read your post and seeing this response, I guess all along you've meant "meaningful" possession, which is obviously a more elevated concept similar to my "fusion" viewpoint. But it was I who misinterpreted your meaning, so my bad.

And in all sincerity, my point was: why not try to elevate it to an even more meaningful level? That's the reason I suggested incorporating faceoffs. But, as usual, my suggestion was not based on any work on my part. And obviously, you've crunched your numbers. So if you say you're reluctant to do so, I respect that.

8/13/2008 11:15 am  
Blogger Vic Ferrari said...

choppystride:

I wasn't taking a shot at you. For what it's worth, I think that you're right on all counts. I just don't know how to do what you're suggesting, barring hiring people to review game tape in a specific way and tally events ... and that's not going to happen.

One thing that is spooky though, and I'm sure you know this, but if you count scoring chances for a game, esp if you allow for 10 cent and 25 cent chances, then even though a big chunk of those weren't recorded as a shot, or even a shot directed at net in any way (puck comes off of a dman's leg and skips just over Torres' stick in front ... a 25 cent chance there to anyone's eye), it's wild how that just seesm to mesh with the shot clock at the end of the game, and better yet with the 'shots directed at net' thing.

Which is obviously why I'm trying to divorce the shots metrics from the actual shots, and tie them to the possession. Hopefully I've done something to accomplish that here.

Alos, that's why I think that the offsides thing could separate the two even in the mind of the most literal and minutae oriented thinker. Not everyone's head works like yours, RiversQ's and slipper's, after all.

8/13/2008 1:47 pm  
Blogger Vic Ferrari said...

Glad to see you here, Performance Oil, a mathy type is good for all of us I think.

Methodology is simple: Pretend that zone/possession matters a bunch and that it's always there, a measure of the quality of your team. Pretend also that the results on any night is a consequence of mostly just luck, the bounces.

With any two samples, team quality is going to shine through, and to the right extent.

It can be better than this though, I've found a split that's cool, and I'm going to ask the Oilogosphere to come up with a reasonable prediction based on reasonable thinking. Has to be nath as an answer though, or more correctly, home truths converted to numbers as well as they can be.

Stick around.

8/13/2008 2:35 pm  
Blogger PerformanceOil said...

Thanks for the insight Vic, but I was more referring to the nuts and bolts of how you did your analysis. What variables were you using them and how? That sort of stuff.

8/13/2008 2:42 pm  
Blogger Vic Ferrari said...

Why don't we both just sit back and wait for slipper to give us instructions.

I like hockey and reason, you like math and arguing, and we both like a bit of each of the other's hobbies here ... it's not like either of us are giving anything up. It's all good.

Whether or not he really knows where both of us live and work, that doesn't matter, we're here now. Christ, I've told him twenty times that I'm a pianist, he still thinks that I work in a DNA lab when he's off the hook. Side note: Don't let him drink Tequila around ya' :D

8/13/2008 3:55 pm  
Blogger Slipper said...

http://www.behindthenet.ca/2007/5_on_5_shots.php?sort=9&mingp=&mintoi=10&team=DAL&pos=F

Could one incorporate shot distance with Corsi to express "meaningful possession"?

Despite poor goal total and a poor +/-, Brad Richards has decent underlying numbers, and is an own zone draw guy, but at the same time, he's firing the puck from the effin boonies.

I look at those tables and feel that there's something there. Thorton, Jagr and company seem to get the puck in close. If you think in your mind of those players who hit career years in goals or career spikes in shooting percentage, you'll find alot of them at the top of these lists aswell.

Is positioning luck? In many circumstances a player surely benefits from a string of fortunate bounces and rebounds. I'd wager my money on the player with a 150 or more shots from an average of 33 feet out over the player with 60 or less shots from 25 feet, atleast for repeatability. Yet, alteast on the surface, a 25 footer has more value than a 42 footer.

8/14/2008 1:40 pm  

Post a Comment

<< Home