All Purpose Analytics and Extended Stats Discussion

[RandyHolt]

Has anyone made a simple good play bad play log sheet?

Good play +1, bad play -1

I know posters/mods here did something similar. But I want shift by shift, making tallys. Have Blaine do it.

When a coach sees any or too many bad plays, review their recent play. When a player has a ton of great plays, considering increasing their role.

Simplicity Stats.

 

[Hivemind]

That's incredibly subjective, which defeats the point of using statistics.

 

[SimplySensational]

Has anyone made a simple good play bad play log sheet?

Good play +1, bad play -1

I know posters/mods here did something similar. But I want shift by shift, making tallys. Have Blaine do it.

When a coach sees any or too many bad plays, review their recent play. When a player has a ton of great plays, considering increasing their role.

Simplicity Stats.

Thats probably what Oates was doing.

For example with Erskine: Good play=make a pass. Bad play=Goal against. So Erskine was like a +3 everygame.

 

[Ajax1995]

That's incredibly subjective, which defeats the point of using statistics.

You don't find shots at least a little subjective?

How many times does a pass get deflected on goal and thus become a shot in a game? A decent amount IMO. How many times is it difficult to determine if a guy was shooting or passing? A decent amount IMO. How many times does a flurry in front of the net result in either a decent amount fewer or more shots than are actually given by the official scorer? A decent amount IMO. And I'm sure we would all come up with quite a few other scenarios that are similar.

All that adds up IMO to make shot attemps a very inexact statistic and in many cases a subjective one.

 

[txpd]

all stats are subjective if you want to break it down that way.

 

[Hivemind]

You don't find shots at least a little subjective?

How many times does a pass get deflected on goal and thus become a shot in a game? A decent amount IMO. How many times is it difficult to determine if a guy was shooting or passing? A decent amount IMO. How many times does a flurry in front of the net result in either a decent amount fewer or more shots than are actually given by the official scorer? A decent amount IMO. And I'm sure we would all come up with quite a few other scenarios that are similar.

All that adds up IMO to make shot attemps a very inexact statistic and in many cases a subjective one.

If you don't see the difference between shots and "good play"/"bad play," I'm not really sure what to tell you.

Sure, there are fringe cases of whether or not something should count as a shot, and scorekeeper biasing is a significant concern. But those cases don't really change the intent of measuring the stat (the puck still made it on goal). It's not 100% accurate, but it's the best we have to work with, for now. SportVU will help take care of some of those concerns, anyway.

 

[ChibiPooky]

Never mind that shot-based metrics don't actually use shots, but rather shot attempts.

 

[NobodyBeatsTheWiz]

All "advanced metrics" are incredibly subjective in how they weigh on-ice activities.

Let's not pretend they're the peak of objectivity.

 

[Hivemind]

All "advanced metrics" are incredibly subjective in how they weigh on-ice activities.

Let's not pretend they're the peak of objectivity.

Things like GVT or HART, absolutely. That's why you don't really see many outside of their creators citing them. Things like Corsi or Fenwick don't weight anything, though.

 

[g00n]

Things like GVT or HART, absolutely. That's why you don't really see many outside of their creators citing them. Things like Corsi or Fenwick don't weight anything, though.

Sure they do. The very premise that "shot attempt = possession = good play with set value" is the foundation for those stats and it's not objective to assume those things.

 

[RandyHolt]

Let me just take this moment to remind everyone that Semin's hit stats were understated because the stat logger (read: kid intern) was never trained and maybe even needed glasses, couldn't tell his and Troy's numbers apart.

#MustbeBrouwer

 

[Hivemind]

Sure they do. The very premise that "shot attempt = possession = good play with set value" is the foundation for those stats and it's not objective to assume those things.

Correlation is not subjective. It has been repeatedly demonstrated that shot attempts correlate with winning hockey games (as well as zone time and other measures of possession). That's objective.

 

[g00n]

Correlation is not subjective. It has been repeatedly demonstrated that shot attempts correlate with winning hockey games (as well as zone time and other measures of possession). That's objective.

No it hasn't. Not in the way these stats are commonly used. I posted several links showing this.

http://hfboards.mandatory.com/showthread.php?p=87969889&highlight=corsi#post87969889

http://hfboards.mandatory.com/showpost.php?p=86397811&postcount=48


Looking back on a game one can argue that better teams USUALLY possess the puck more and shoot more but to say that shooting = winning is simplistic and not causal. Neither is analysis of players shot attempts and assuming that all else is equal.



And as you know, correlation is not causation. As has been explained over and over again a stat may be an indication of many things other than what it's assumed to represent.

 

[ChibiPooky]

Who's arguing that "possession" metrics are 100% predictive?

And who cares what a stat actually indicates if it correlates highly with the thing you're trying to predict? It doesn't need to be causative to have high predictive value.

 

[g00n]

Who's arguing that "possession" metrics are 100% predictive?

And who cares what a stat actually indicates if it correlates highly with the thing you're trying to predict? It doesn't need to be causative to have high predictive value.

The question was whether they're objective. They're only objective in terms of clearly defined shots and shot attempts being used in formulas. They are not objective in measuring good and bad plays, or even possession. They are subjective in that regard, and subject to context even as they correlate to winning.

 

[Hivemind]

Had a long post typed out with links, but I lost it because I got timed out.

Will try and rewrite it later. But the moral of the story is that possession stats have both empirical correlation with zone time (possession), scoring chances, and points in a season (winning hockey games). They also have more predictive value than goal-based metrics. They are also theoretically sound when you look at the game in a probabilstic fashion. Those are objective reasons to value them.

 

[g00n]

Had a long post typed out with links, but I lost it because I got timed out.

Will try and rewrite it later. But the moral of the story is that possession stats have both empirical correlation with zone time (possession), scoring chances, and points in a season (winning hockey games). They also have more predictive value than goal-based metrics. They are also theoretically sound when you look at the game in a probabilstic fashion. Those are objective reasons to value them.

Value of some kind, yes. Rely on entirely and cite without context, no.

Again, correlation is of limited value without context. Since I'm playing Madden in prep for Madden Day tomorrow, looking at football...a quarterback in the NFL can rack up massive stats late in games and if you look just at that you may think he's elite or clutch. But it just may be that his team stinks and is playing against soft/prevent coverage late in games, or against subs. Same goes for a RB that doesn't get a lot of touches because his team trails all the time and the coach abandons the running game. Then there's the effect of systems, roles, etc.

I've gone over this many times so not rehashing it here in an acquisition thread. But I will say so-called possession metrics are more useful in a team context than an individual one. Teams that are in the o-zone and shoot more as a whole are probably going to win more. That's a no-brainer. But individual players are much more subject to the other factors I mentioned. Calling shots and shot attempts "possession" and then ignoring player roles and usage is not objective and you can't run a team based solely on it.

 

[Ajax1995]

so-called possession metrics are more useful in a team context than an individual one.

100% agree.

 

[malyk]

That's incredibly subjective, which defeats the point of using statistics.

Not exactly. The problem with not using stats is that your memory of events is skewed by the outliers. Jeff Schultz getting burned in the playoffs against the rangers is high on ever one's list of how bad he is/was. But his contributions were much more subtle and not memorable. Note I'm not saying Schultz was a great player here.

But, if you had the same person, or even better a group of people doing it individually, breaking down every play a player was involved in and aggregating the results, you could end up with something much more valuable and less subjective than you imply.

Of course, it would be super time consuming and you'd have to have an agreed upon standard for good vs. bad, but you could end up with a very useful result.

 

[Hivemind]

Value of some kind, yes. Rely on entirely and cite without context, no.

Which nobody does. Once again, going back to the Orpik contract that has stemmed so many of these debates this summer, entire articles were dedicated to providing context for the stats used.

Not exactly. The problem with not using stats is that your memory of events is skewed by the outliers. Jeff Schultz getting burned in the playoffs against the rangers is high on ever one's list of how bad he is/was. But his contributions were much more subtle and not memorable. Note I'm not saying Schultz was a great player here.

But, if you had the same person, or even better a group of people doing it individually, breaking down every play a player was involved in and aggregating the results, you could end up with something much more valuable and less subjective than you imply.

Of course, it would be super time consuming and you'd have to have an agreed upon standard for good vs. bad, but you could end up with a very useful result.

That's only one of the many cognitive biases that impacts the "eye test." The closest to a concrete answer of a "good play," "bad play" that you're going to get is the coaching staff, since they know the system and what the players' responsibilities on each play were. But even then it's not going to be immune from biasing, or even objective.
For instance:

Bowman — then a 29-year-old with a famous last name and a business background — actually got his foot in the door with the Hawks in 2001 by bringing some statistical analysis to a franchise that hardly had any. He started simply, with coaches rating players on a 1-5 scale each game, charting the ratings over the course of 10-game spans. But Bowman quickly learned that the ratings were flawed — subject to a coach’s affection for a player, or to his lowered expectations of some players and raised expectations of others. So Bowman started blending the old-school analysis with the new-school analytics.

http://www.suntimes.com/27179980-419/advanced-analytics-a

If you take it one step removed, to the scouts, you end up with something resembling your pro scouting report. If you fully crowd source it, you end up with the same garbled mess of fan opinion you already have.

 

[RandyHolt]

Maybe we can do my good play bad play for one game, or one period, and see what it says.

A guy busting his hump on a backcheck prevents an almost sure goal. RH gives him a +1. Where does that show up in the stat sheet otherwise?

A guy laying down to block a slap shot to preserve the win late, vs Sarge cringing as a wrister innocently hits him in the first period, simply in the way / screening the G. Stat sheet shows the same thing.

Mine wouldn't.

 

[g00n]

Which nobody does. Once again, going back to the Orpik contract that has stemmed so many of these debates this summer, entire articles were dedicated to providing context for the stats used.

Nobody does? Plenty of people worship at the altar of Corsi and Fenwick. Some are right here on this board. The arguments are tiresome. Those articles sprung up BECAUSE so many people just look at those stats.

 

[g00n]

Maybe we can do my good play bad play for one game, or one period, and see what it says.

A guy busting his hump on a backcheck prevents an almost sure goal. RH gives him a +1. Where does that show up in the stat sheet otherwise?

A guy laying down to block a slap shot to preserve the win late, vs Sarge cringing as a wrister innocently hits him in the first period, simply in the way / screening the G. Stat sheet shows the same thing.

Mine wouldn't.

It's not that good plays are too subjective to be useful. Coaches make careers off of recognizing good and bad plays. However I don't know that it would be practical to try and quantify good vs bad plays into a universal stat. You could probably do it internally per team, if you had a good team of people on it all year.

Why is everyone in such a rush to do away with the human element of hockey coaching and analysis? Is it because any guy with a spreadsheet can sit around and crunch numbers? Again, why not just have the stat guys run the teams?

 

[Hivemind]

Nobody does? Plenty of people worship at the altar of Corsi and Fenwick. Some are right here on this board. The arguments are tiresome. Those articles sprung up BECAUSE so many people just look at those stats.

Sometimes people use them as a conversational shorthand, instead of writing an entire essay on the player's utilization and deployment, sure. I think the vast majority of those who cite them understand their limitations and how to build proper context, though.

 

[Ajax1995]

http://www.suntimes.com/m/27179980-...he-blackhawks-secret-formula-for-success.html

From above:

They’re tallying every shot — on goal, blocked, or simply missed. They’re tracking where each one was shot, where it was aimed, how it was shot, against whom it was shot.

IMO when you include the bold part with the first part it all becomes a whole lot more valuable. It gives it actual context.

But as far as I can tell the fancystats crowd disagrees with this and says a shot is a shot is a shot and I have serious issues with that.