Department of Hockey Analytics (Toronto Star)

[Bear of Bad News]

I'm not familiar with these guys (individually or collectively), and the article doesn't go into a lot of depth:

http://www.thestar.com/sports/hocke...alysis_project_like_moneyball_for_hockey.html

Click to read more...

 

[Bear of Bad News]

Oddly enough, I just tried a Google search on the group, and this article from today came up:

http://www.thespec.com/sports-story...t-about-stat-padders-not-all-goals-are-equal/

Seems similar of a concept to Victory-Important RBI (if you remember that from Bill James' Baseball Abstracts).

 

[TieClark]

Oddly enough, I just tried a Google search on the group, and this article from today came up:

http://www.thespec.com/sports-story...t-about-stat-padders-not-all-goals-are-equal/

Seems similar of a concept to Victory-Important RBI (if you remember that from Bill James' Baseball Abstracts).

This is definitely intriguing

I think they'd have to correlate the teams they're playing more often though as well. In the case of Ovie for example the Caps played in such a terrible division that literally ever division game he played in counted as a non-playoff team and you play your division rivals more often than others.

 

[Bear of Bad News]

Someone else seems to have taken an interest, and I have to admit that I enjoyed their first piece:

http://www.pensionplanpuppets.com/2014/2/7/5388600/department-of-hoakey-analytics-usa-goalies

 

[Hockey Outsider]

Someone else seems to have taken an interest, and I have to admit that I enjoyed their first piece:

http://www.pensionplanpuppets.com/2014/2/7/5388600/department-of-hoakey-analytics-usa-goalies

An interesting article, but I disagree with much of it. The author's premise is, essentially, hockey teams in a short tournament like the Olympics want consistent goalies. My problem with this is two-fold.

First, the crux of their analysis (a weighted-average probability table) is based on arbitrary assumptions; small tweaks to the assumptions could change their conclusions. It would have been more interesting and informative if they actually calculated the probabilities (empirically, a goalie performance with a save percentage in each third (or quarter, or decile, etc) corresponds to an X% probability of the team winning).

Second (this is an observation by the great basketball analyst Dean Oliver, whose observations can sometimes be linked to hockey), variance is generally unfavourable when you're already the favourite. Variance is very much a good thing for the underdog.

For example, on average, the 1977 Montreal Canadiens are likely to steamroll their opponents - they just want a steady, consistent goalie who won't blow the game for them. If Dryden stops 92% of the shots he faces, with a low level of variance, the team's high-powered offense and smothering defense will allow the Canadiens to win frequently. If Dryden was inconsistent (and alternated between Hasek-like performances and terrible performances), the variance would increase and the Habs would be less successful. In other words: it doesn't matter much if Dryden stopped 92% or 96% or 100% of the shots he faced - his team gave him so much offensive support and he saw so few shots that the incremental benefit of those extra saves is small. The benefit from his Hasek-like performances would be minimal, but clearly it would cost his team games (even a team as strong as the dynasty Canadiens) if he started allowing 5+ goals on a regular basis.

Similarly, a bad team (underdog) wants an inconsistent goalie. The team is likely to lose; a consistent goalie should result in consistent loses. An inconsistent goalie could turn some of those loses into wins (even if the team struggles to score, if the goalie sometimes records a shutout, he assures his team at least one point). Sure, he'll have awful performances too, but the team was likely to lose in the first place. Losing a blowout is no worse in the standings than losing a by two goals (those perhaps there's long-term psychological impact?) In other words: instead of consistently losing by 1-2 goals, some of those losses will turn into narrow wins (benefit in the standings), and some of those will turn into blowout losses (no worse than they were before) - and the team's in a better position overall.

My subjective perception is that Luongo is an inconsistent goalie - when he's on, he's almost unbeatable, but he's had a lot of terrible games as well. Assuming Team Canada is the favourite heading into the Olympics, perhaps Price is a better option (almost the same save percentage over the past three years and, subjectively, I feel he's more consistent). An underdog team would want an inconsistent goalie (with a high level of play obviously) and hope they go on a "JS Giguere in 2003" type of run.

====

Note 1: I haven't verified any of this empirically in the context of hockey. Oliver proved this in the context of basketball, and I think the same reasoning applies. I may look into this in the future (though if sometimes wants to look at it first, go for it).

Note 2: I'm not saying that Dryden was merely a consistent goalie; the most important things is that he was an excellent goalie. But being consistently excellent made him a more valuable fit for the Canadiens than a goalie with the same save percentage (same level of excellence), with less consistency.

Note 3: not interested in getting into a heated debate about Luongo vs Price. Those are subjective observations and if someone can show me that Luongo is the more consistent of the two, empirically, then I'll change my mind. Of course, more than just save percentage and consistency need to be taken into account, but they are two important factors.

 

[supsens]

A consistent goalie, is good most of the time and great some of the time, it's a better choice because his bad games are still good.
inconsistent is when he sucks he suckssssss hard

 

[Stringhama]

I would much rather read Neil Greenberg than the dept. I would put the dept. near the bottom of my advanced stats reading list.

 

[Viktri]

Oddly enough, I just tried a Google search on the group, and this article from today came up:

http://www.thespec.com/sports-story...t-about-stat-padders-not-all-goals-are-equal/

Seems similar of a concept to Victory-Important RBI (if you remember that from Bill James' Baseball Abstracts).

I like this, and agree - I came to similar (controversial) conclusions (as the article) from the earlier thread where you mentioned victory-important RBI (and data was provided).

Someone else seems to have taken an interest, and I have to admit that I enjoyed their first piece:

http://www.pensionplanpuppets.com/2014/2/7/5388600/department-of-hoakey-analytics-usa-goalies

I think their stance but disagree with their methodology.

They say:

To further this, look at Steve Mason, a goalie who everyone agrees is highly inconsistent. Mason has a higher percentage of games over a .940 sv% (38.1%) than Howard (32.81%) or Miller (32.84%), but also a higher percentage of games under a .900 sv% (38.1%) than either Howard (32.81%) or Miler (29.85%).

But I don't like this line of reasoning - basically they have their conclusion (it's not even a hypothesis) and look for data to support it.

Their categorization method also could be improved (theoretical one anyway). They divide goalie impact into: 1) goalie players terribly, 2) goalie plays average and 3) goalie plays well, none are a driver of goalie play so it really makes little sense to measure such things when you're trying to be predictive. A ******** the bed game could easily have the goalie player better than an average game if the goalie faced significantly fewer high quality chance shots in the average game.

A better theory-craft methodology (and significantly more labourous to test but much better if you are arguing about poker, which is analogous at best) would be a system that measures: 1) high quality chance saves/not saves, 2) routine saves/not saves and 3) other unusual or unintended situations saves/not saves, and build from there.

After this initial analysis, which only measures goalie reliability in totality (rather than on a game by game basis), you could then look at data by game sample - does sv% on high-quality saves change significantly after the first goal? We would expect sv% on high-quality saves to fall after the first goal (the goalie could possibly be facing a good team, and such teams often have better than average shooters) so we would need to compare it to other goalies.

I don't think their stance is wrong but I believe that you need a correct line of reasoning before really diving in and making conclusions, because if your reasoning isn't right in the first place, nothing can follow from it.