### Friday, February 27, 2009

## A week of Puck Prospectus

I was quite interested to see the launch of Puck Prospectus, Baseball Prospectus' entry into hockey analysis. While there are lots of people doing hockey analysis, both inside the NHL and on their own, conventional hockey reporting has yet to absorb any of it, not even to the point where hockey columnists need to dismiss its practitioners as "a bunch of guys sitting in their mom's basements."

At any rate, I thought I'd give a brief run-down of Puck Prospectus' first week of articles:

Robert Vollman: Projecting Future Scoring. Similarity scores are a great way to predict a player's future performance, particularly in baseball, where a multitude of statistics and positions can give you a sense of what kind of player someone is. There are things you can suss out about a player's likely aging curve (eg: Ben Grieve, young players with old players' skills) that you would miss with a generalized aging curve. But using just hockey boxcar stats? I think there's not enough information there - I'm guessing that a simple "projection" system like Tangotiger's Marcels would do just as well. Hopefully Rob can give us more details and show us a better system.

Iain Fyffe: The Beginning. I like this approach: since so many readers are essentially uninitiated in hockey analysis, explain the basics. The most important thing you look for in a future NHL star is high point totals at a young age. And of course if you're comparing players to one another, you need to make an allowance for offensive variations over the years.

Timo Seppa: Plus/Minus - A Nonsense Stat?. This article compares Plus/Minus to RBIs, and, after showing the context-sensitive nature of +/-, dismisses it as a nonsense stat, like RBIs. Unfortunately, there's a big difference between RBIs and +/-: because run-scoring in baseball can be separated into context-independent components (singles, doubles, triples, home runs, walks, stolen bases, etc...), RBIs never come into play in advanced statistics. But a player's value in hockey is, in essence, captured by the number of goals he contributes to, minus the number he allows. There are obviously adjustments to be made for linemates, opponents, defensive zone faceoffs and the like, but at the most fundamental level, a player's plus/minus is his value to his team. In other words, it's very much not nonsense.

Tom Awad: Bayesian Power Rankings. This is a great way of looking at the NHL. You can see, for example, that the Leafs somehow have a 9% chance of making the playoffs! Since this is all done at a team level, one thing that's hard to assess is a team's real winning percentage - you have to assume that they are exactly as good as they have been. Compare that to baseball, where individual projections and lineup predictions can be used to estimate the inherent quality of a team and possibly get a better sense of how they'll do going forward, particularly at the beginning of the season. It probably takes one person's spare time to do this for the NHL, so maybe that's what Puck Prospectus has in store for us in the long run.

Andrew Rothstein: Second Half of February Trades. I'll admit I'm a little confused by the discussion in this piece. Andrew refers to several things that I'm not familiar with: Even-Strength Shooting Percentage Differential, a 1-point-per-game barrier for rookies, Special Teams Pythagorean winning percentage. BP has most of their jargon linked and defined - PP might be well-served to do the same, and to spend a number of articles outlining their methodology. BP had the advantage of developing their stats publicly, at least initially, so that analysts were aware of them. Hockey's a bit more obscure...

At any rate, I thought I'd give a brief run-down of Puck Prospectus' first week of articles:

Robert Vollman: Projecting Future Scoring. Similarity scores are a great way to predict a player's future performance, particularly in baseball, where a multitude of statistics and positions can give you a sense of what kind of player someone is. There are things you can suss out about a player's likely aging curve (eg: Ben Grieve, young players with old players' skills) that you would miss with a generalized aging curve. But using just hockey boxcar stats? I think there's not enough information there - I'm guessing that a simple "projection" system like Tangotiger's Marcels would do just as well. Hopefully Rob can give us more details and show us a better system.

Iain Fyffe: The Beginning. I like this approach: since so many readers are essentially uninitiated in hockey analysis, explain the basics. The most important thing you look for in a future NHL star is high point totals at a young age. And of course if you're comparing players to one another, you need to make an allowance for offensive variations over the years.

Timo Seppa: Plus/Minus - A Nonsense Stat?. This article compares Plus/Minus to RBIs, and, after showing the context-sensitive nature of +/-, dismisses it as a nonsense stat, like RBIs. Unfortunately, there's a big difference between RBIs and +/-: because run-scoring in baseball can be separated into context-independent components (singles, doubles, triples, home runs, walks, stolen bases, etc...), RBIs never come into play in advanced statistics. But a player's value in hockey is, in essence, captured by the number of goals he contributes to, minus the number he allows. There are obviously adjustments to be made for linemates, opponents, defensive zone faceoffs and the like, but at the most fundamental level, a player's plus/minus is his value to his team. In other words, it's very much not nonsense.

Tom Awad: Bayesian Power Rankings. This is a great way of looking at the NHL. You can see, for example, that the Leafs somehow have a 9% chance of making the playoffs! Since this is all done at a team level, one thing that's hard to assess is a team's real winning percentage - you have to assume that they are exactly as good as they have been. Compare that to baseball, where individual projections and lineup predictions can be used to estimate the inherent quality of a team and possibly get a better sense of how they'll do going forward, particularly at the beginning of the season. It probably takes one person's spare time to do this for the NHL, so maybe that's what Puck Prospectus has in store for us in the long run.

Andrew Rothstein: Second Half of February Trades. I'll admit I'm a little confused by the discussion in this piece. Andrew refers to several things that I'm not familiar with: Even-Strength Shooting Percentage Differential, a 1-point-per-game barrier for rookies, Special Teams Pythagorean winning percentage. BP has most of their jargon linked and defined - PP might be well-served to do the same, and to spend a number of articles outlining their methodology. BP had the advantage of developing their stats publicly, at least initially, so that analysts were aware of them. Hockey's a bit more obscure...

Labels: Puck Prospectus

Comments:

Agreed. It's funny how some folks will dismiss "traditional" +/- out of hand, but place more stock in things like GF/GA ON/60, which is a slightly more refined way of expressing the same damn thing. The considerations you mention -- QualTeam, QualComp, faceoff location -- are all important and provide context to raw +/- figures while hardly rendering them irrelevant.

Post a Comment

Links to this post:

<< Home

Even Strength Shot Differential = Getting outshot at even strength,

ST Pythagorean Theorem = Bill James win expecation theorem (same exponents), 1.0 PPG is a threshold for players in their first draft eligible year (Got this from Rob and Tom). It was a mistake to use rookie in place of first draft eligible year.

ST Pythagorean Theorem = Bill James win expecation theorem (same exponents), 1.0 PPG is a threshold for players in their first draft eligible year (Got this from Rob and Tom). It was a mistake to use rookie in place of first draft eligible year.

*There are obviously adjustments to be made for linemates, opponents, defensive zone faceoffs and the like, but at the most fundamental level, a player's plus/minus is his value to his team. In other words, it's very much not nonsense.*

Agreed. It's funny how some folks will dismiss "traditional" +/- out of hand, but place more stock in things like GF/GA ON/60, which is a slightly more refined way of expressing the same damn thing. The considerations you mention -- QualTeam, QualComp, faceoff location -- are all important and provide context to raw +/- figures while hardly rendering them irrelevant.

Links to this post:

<< Home

Subscribe to Posts [Atom]