Tuesday, April 14, 2009

New Puck Prospectus Article: Stanley Cup Scenarios

The key to predicting the performance of hockey teams is figuring out their "true talent," which is easier said than done. In baseball, numerous people make reasonable projections of playing time for each player on every team, along with predicting offensive and defensive contributions. With these projections, an expected level of performance for each team can be estimated. We're not there in hockey - all we have to go on is a team's actual performance; in particular, its expected winning percentage based on goals scored and allowed during regulation play. We can make adjustments to this winning percentage if the team faced a tough schedule or if it made significant trade deadline acquisitions or had star players returning from or lost to injuries. However, for the vast majority of teams - including the field this season - estimated "true talent" and actual performance are within a few percentage points of each other...


I have to say I disagree with most of this article. This statement:

"The Bruins are as likely to win the Cup as San Jose, Detroit and Chicago combined."

is an absolute joke imo. Tell you what: give me even money - you take Boston, and I'll take those other three teams - and you can name your price. Would you make that bet? :)

In a nutshell, using season-long raw GF and GA while making no adjustment for possible variance/randomness in shooting and save percentages, no adjustment for conference/schedule, and no adjustment for how teams have changed personnel/coaching/strengths/weaknesses throughout the season (i.e. expected performance today very different than November results indicate) is just plain silly.

This type of analysis is my issue with nearly all of the articles on Puck Prospectus. That one dude, week after week, keeps putting Boston at the top of his "Power Rankings". Um, what? Has he not noticed that Boston was running at a 10+ S% clip for a few months at the beginning of the season, and that for the past 30 or so games they've been a slightly better than even-goal-differential team?
I think you miss the point of what I'm writing for P.P. I'm not putting out proprietary methods - everything is out on the table for your inspection. And there is no way that I can describe a system that takes into account all of the things you ask for in the amount of space that I have.

My goal is to engage people with limited exposure to statistical hockey analysis, not to generate the world's best power rankings. If a more complex method correlates better with reality than a simple one, so be it, and I'm happy to hear the discussion.

Some people like a first-order approximation; some people like a different kind of system. So be it.
I hear ya, Gabe. Writing with space limitations and having to cater to an audience of varied statistical background is definitely a challenge. My comment was only intended to spark discussion on what types of analyses people think are useful for playoff handicapping.

Obviously I think the work you do here on BTN is invaluable. Keep up the great work.
heh, hopefully I get a second commenter over here!

What site do you think has the best playoff handicapping method?
hm, well i don't think anyone has put together one all-encompassing model that far exceeds others. but model-wise i'd probably say that i like Javageek's Expected Goals/shot quality model stuff at Hockeynumbers best.

general prediction-wise, i tend to agree most with JLikens at Objective NHL (you can read some of his recent posts to check out the discussion we've had over there) because we tend to approach handicapping in a similar way, which in my mind is using ALL these great resources available to make educated guesses about things.

hockey stat analysis is still at such infancy that i think it's important to be skeptical and open-minded about different correlation/causality theories, etc. but a lot of people, including you, are doing great work, and we're all making big strides imo.

to me, the key things i like to look at, at the team level, are:

-shot numbers (SF/60, SA/60, corsi, etc) at all three situations (ES, PP, PK)

-percentages (S% and SV%) at all three situations

-some sort of metric that factors in shot quality (i.e. - Expected Goals)

A few things I'd love to see you add here at BTN are:

1) That corsi rating (where you factored out own team influence) that you referenced in your previous Puck Prospectus article.

2) Some way of letting the user filter all your stats by date, so that we aren't just limited to looking at season total stats. (This way we could see trends.) Graphs would be ideal!

3) Eventually, some sort of shot quality model. (I realize that's probably more than just a couple lines of code lol.)
Post a Comment

Links to this post:

Create a Link

<< Home

This page is powered by Blogger. Isn't yours?

Subscribe to Posts [Atom]