Hockey Analytics: What Does It Mean For Cornell?

[adamw]

Really nit-picking here, but the purpose of the sacrifice bunt is to give up trying to get on base to move base runners over.

Yeah, that's really nitpicking, because you know what I meant  That, and what css228 said.

[Rosey]

I don't think I've ever wanted an "ignore thread" option as much as I do now. :-)

[KeithK]

Here is the main difference between OPS and Corsi.  OPS is a direct building block to scoring runs.  You would never tell a baseball player "get on base less" or "please hit home runs only to this spot, or at this time".  Whereas with shots ... you can't say that the solution to improving your game is simply shooting the puck more.  That may or may not be the case, but you can't know that.

Really nit-picking here, but the purpose of the sacrifice bunt is to give up trying to get on base to move base runners over.

And the sacrifice bunt is a stupid play which lowers run expectancy. Years of studies have consistently shown that you can expect to score more runs per inning with no outs and a runner on first than 1 out and a runner on 2nd. There may be some situations with particular hitters when a sacrifice bunt might increase run expectancy (e.g. pitchers), but overall its dumb and shouldn't be done.

Here in a nutshell is what is wrong with sports analytics.  (I'm picking on your post and sorry for that!)  There are studies that show that bunting leads to a lower run expectancy and you take that study and jump to "it's dumb and shouldn't be done". There's a tendency for people who are advocates of analytics to put too much faith in the results of the studies.  This sometimes means you miss the nuances of how to apply what you' learned from the studies.  There's an even greater tendency to take on an attitude of "whoever doesn't follow analytics is dumb", which is especially prevalent when simultaneously missing the nuances.

Using your bunting example, I generally agree that bunting is a poor decision. But there are times where it makes sense: bad hitter, bad hitter-pitcher matchup, ground ball pitcher with slow batter meaning higher probability of double play, late inning situation where one run is vastly more important than multiple runs.  I suspect that a sufficiently detailed study would capture some of these effects when you can get enough sample size. But I routinely see people using the general purpose run expectancy of a bunt as reason to toss insults at any manager who employs it (or fans who support the idea).

Analytics are a useful tool but need to be used in the proper context. Since hockey is a sport that is more difficult to analyze and anlytics are their infancy we shoud be especially careful abo drawing strong conclusions.

[adamw]

And the sacrifice bunt is a stupid play which lowers run expectancy. Years of studies have consistently shown that you can expect to score more runs per inning with no outs and a runner on first than 1 out and a runner on 2nd. There may be some situations with particular hitters when a sacrifice bunt might increase run expectancy (e.g. pitchers), but overall its dumb and shouldn't be done.

I played in an adult baseball league a couple years ago ... and every time we got 1st and 2nd nobody out, some dude who played D-I baseball in college would yell and scream that we should be bunting - because one time, we grounded into a DP.  I thought we were going to get into blows because I kept telling him he didn't know what he was talking about.  I considered sending him all the math on the topic, but eventually said screw it.

But there's a good example where you can take a baseball event, and directly measure its effect on the thing that's most important -- scoring runs. Hockey is not there yet. But I appreciate everything you've said, and appreciate you acknowledging some of what I'm saying.

[adamw]

Analytics are a useful tool but need to be used in the proper context. Since hockey is a sport that is more difficult to analyze and anlytics are their infancy we shoud be especially careful abo drawing strong conclusions.

Right on Keith - that's exactly what I've been saying. Stats are stats - it's how you use them. I don't doubt that Fenwick generally demonstrates what it's intending to demonstrate. But what is it really telling us about the *ability* to score a goal, or ability of a certain player, or informing us of what strategy to use. We don't fully know yet.

Another example is carrying the puck over the line vs. dumping. Metrics tell you that carrying the puck into the zone leads to more shots. So - does that mean everyone should carry the puck in, or carry it all the time? Heck no. There's too many factors.  Yet, with baseball metrics, it's often pretty obvious what the decision should be, based upon the situation. It's more solveable.

[Dafatone]

And the sacrifice bunt is a stupid play which lowers run expectancy. Years of studies have consistently shown that you can expect to score more runs per inning with no outs and a runner on first than 1 out and a runner on 2nd. There may be some situations with particular hitters when a sacrifice bunt might increase run expectancy (e.g. pitchers), but overall its dumb and shouldn't be done.

I played in an adult baseball league a couple years ago ... and every time we got 1st and 2nd nobody out, some dude who played D-I baseball in college would yell and scream that we should be bunting - because one time, we grounded into a DP.  I thought we were going to get into blows because I kept telling him he didn't know what he was talking about.  I considered sending him all the math on the topic, but eventually said screw it.

But there's a good example where you can take a baseball event, and directly measure its effect on the thing that's most important -- scoring runs. Hockey is not there yet. But I appreciate everything you've said, and appreciate you acknowledging some of what I'm saying.

Two more reasons why bunting is lame.

1) Often, it's done after a pitcher's gotten himself into trouble with zero outs.  Why give him an out?  Most of the "math" doesn't factor in that if a pitcher's let two straight guys on, he might be off his game at that particular moment.

2) The math also usually assumes bunts have a 100% success rate.  It's 1 out, guy on 2nd versus 0 out, guy on 1st.  There's no consideration of the hitter maybe failing to get a bunt down.

This isn't to say never sac bunt.  Just, do it less.

[Josh '99]

Even WAR is not a be all and end all of value, which is why there are 2 types of WAR for non-pitchers and they're still most useful only when comparing players who play the same position.

Assuming you're talking about fWAR and rWAR, there are two types because two different entities set out to create a metric that "attempts to encapsulate a player's total value to their team in one stat" (in other words, it IS supposed to be a be all and end all, it's just imperfect at this point) and went about it somewhat differently.

Yeah I guess that one is, but it still can't really be compared across positions, even with the positional adjustments. Not to mention Pitcher WAR and regular WAR are calculated so differently that you can't really compare the value of a pitcher to an everyday player.

That's true, but I think that goes to the fact that the metric just isn't good enough to make those kinds of comparisons yet.

It's a fairly meaningless hair we're splitting here though.  I think you're right in general that almost anything you'd look at is still just a piece of the puzzle.

[RichH]

Another example is carrying the puck over the line vs. dumping. Metrics tell you that carrying the puck into the zone leads to more shots. So - does that mean everyone should carry the puck in, or carry it all the time? Heck no. There's too many factors.  Yet, with baseball metrics, it's often pretty obvious what the decision should be, based upon the situation. It's more solveable.

Well that's the heart of it, right? Situational analysis is much easier in baseball, simply because there are many more easily-definable situations. With no clock, one can take stock and neatly define the situation before each pitch and use it as a data point. Hockey, basketball, soccer, et al. are nearly always in motion so there are fewer opportunities to define a pre-play situation, as well as fewer measurable and reducable variables.

[ugarte]

Don't feel like responding to a particular post but ... tell me where I'm wrong:

From what I've read here, Corsi sounds like a stat that is descriptive, and useful for describing the past (and evaluating player/team quality), but not something that can be used to make specific improvements. I suppose you could juke the stat by SHOOOOOOOOOOOOOOOTing more, but what kind of chaos-motivated person would go about trying to juke the stat at the expense of winning? (Even in an arbitration year, you'll end up a healthy scratch with a high corsi because your coach thinks you are an imbecile.) So, since you want to take good shots, you take the shots that are there and the shots act as a proxy for possession, letting the world know who had the most opportunities. It isn't useful for improvement because the stat doesn't tell you how to get more shots, only that - if your corsi sucks - what you are doing isn't working and you should probably watch more tape.

So while the discussion shouldn't begin and end with corsi/fenwick it is a much better starting point than where hockey was before it was developed, no?

[css228]

Don't feel like responding to a particular post but ... tell me where I'm wrong:

From what I've read here, Corsi sounds like a stat that is descriptive, and useful for describing the past (and evaluating player/team quality), but not something that can be used to make specific improvements. I suppose you could juke the stat by SHOOOOOOOOOOOOOOOTing more, but what kind of chaos-motivated person would go about trying to juke the stat at the expense of winning? (Even in an arbitration year, you'll end up a healthy scratch with a high corsi because your coach thinks you are an imbecile.) So, since you want to take good shots, you take the shots that are there and the shots act as a proxy for possession, letting the world know who had the most opportunities. It isn't useful for improvement because the stat doesn't tell you how to get more shots, only that - if your corsi sucks - what you are doing isn't working and you should probably watch more tape.

So while the discussion shouldn't begin and end with corsi/fenwick it is a much better starting point than where hockey was before it was developed, no?

It is in my opinion a very good starting point, and I believe it does have some predictive value (especially fenwick) in that it indicates a process that is conducive to winning. But some here may think that is too strong a conclusion to draw.

[css228]

Maple Leafs Update. Not really looking good recently. Can't shoot 10%+ forever.

[Rita]

From Grantland, a general article on puck possession and zone entry. No equations or other hard math .

[billhoward]

The gist of the story in http://www.grantland.com/story/_/id/10083164/zone-entry-nhl

If you're familiar with advanced stats, you know how much emphasis they place on possession. In the case of dump-in vs. carry, you might suspect that that gives away the answer, and you'd be right. It turns out that crossing the blue line with possession is worth more than a dump-in ... How much more? Quite a bit, according to this paper presented at the 2013 MIT Sloan Sports Analytics Conference. After tracking more than 300 games from the 2011-12 season, the authors found that, even when accounting for the higher failure rate of carry attempts, that approach still generated roughly twice as many shots, scoring chances, and goals as dumping in the puck and trying to retrieve it.

... So a clean zone entry requires some skill, and skill isn't a quality that's distributed equally among NHL players. It may make perfect sense to have less-skilled guys who make up the league's third  and fourth lines keeping shooting it in, especially when a team is defending a lead.

... Are any NHL teams changing their strategy based on this kind of thinking? Yes, as it turns out. One prominent example: the Minnesota Wild, a classic dump-and-chase team last year that made the playoffs with that style of play. But after the Wild realized they were going to need to beat elite divisional rivals like the Blackhawks and Blues to get to the next level, the Minnesota brain trust decided to shift strategies. As outlined in this recent post by Elliotte Friedman, Minnesota has decided to focus on gaining the zone with the puck.

[profudge]

Interesting article on movement to detailed video analytics in ice hockey:  FiveThirtyEight article "The People Pushing the NHL Into the Advanced Stats Era"

...
Tracking by Schmidt and others has helped explain that a team's entry into the offensive zone has a big impact on its shot differential. Carrying the puck into the offensive zone leads to more than twice as many shots and goals as a dump-and-chase play does, even after removing plays like odd-man rushes and dump-ins that are made just to buy time for a line change. These results have even made an impact on strategy.

Schmidt's data from this season allows us to evaluate individual defensive contributions for the first time, which we couldn't do with traditional box scores.  ...

[Trotsky]

Interesting article on movement to detailed video analytics in ice hockey:  FiveThirtyEight article "The People Pushing the NHL Into the Advanced Stats Era"

...
Tracking by Schmidt and others has helped explain that a team's entry into the offensive zone has a big impact on its shot differential. Carrying the puck into the offensive zone leads to more than twice as many shots and goals as a dump-and-chase play does, even after removing plays like odd-man rushes and dump-ins that are made just to buy time for a line change. These results have even made an impact on strategy.

Schmidt's data from this season allows us to evaluate individual defensive contributions for the first time, which we couldn't do with traditional box scores.  ...

Dumb question: why have people tracking and recording visually?  Why not just stick an RFID on every player, stream the whole thing to a hard drive, and use software to analyze it?