All Purpose Analytics and Extended Stats Discussion

[NobodyBeatsTheWiz]

Obviously I'm not talking about the two games worth of data we have.

Which data are you talking about?

In general, the Corsi numbers only give a general indication of what happened in terms of shots when a player or team is on the ice (just like +/- with goals).

Other similarities to +/-:

-No apportionment of blame or responsibilities for on-ice occurrences
-Generally biased against certain types of players (namely defensive defensemen)
-Being the 'loser' in either stat generally correlates to poor performance.

The way I see it, the only possible advantages to Corsi over +/- are that Corsi has been further developed to try to normalize (with utilization of zone starts, QoC, etc.), and Corsi eliminates goaltender performance. But more advanced +/- stats developed to normalize save percentage (extremely easy adjustment), and to take into account zone starts (another extremely easy adjustment), would arguably be more useful than Corsi.

 

[Burakovsky95*]

Which data are you talking about?

In general, the Corsi numbers only give a general indication of what happened in terms of shots when a player or team is on the ice (just like +/- with goals).

Other similarities to +/-:

-No apportionment of blame or responsibilities for on-ice occurrences
-Generally biased against certain types of players (namely defensive defensemen)
-Being the 'loser' in either stat generally correlates to poor performance.

The way I see it, the only possible advantages to Corsi over +/- are that Corsi has been further developed to try to normalize (with utilization of zone starts, QoC, etc.), and Corsi eliminates goaltender performance. But more advanced +/- stats developed to normalize save percentage (extremely easy adjustment), and to take into account zone starts (another extremely easy adjustment), would arguably be more useful than Corsi.

You cant just say +/- = corsi but let me **** with +/- so that it becomes an actual useful stat. That isn't +/-.

I'm not sure what you're trying to say with your first point.

Your 2nd point is just wrong as some of the leaders at D include Vlasic, Seabrook, Boychuk, Mitchell, Braun, Bouwmeester, Staal, Fayne, Rosival, Hjalmarsson, M. Greene, Methot, Jackman, Volchenkov, Hamhuis, etc.

Being the loser in any stat generally correlates to poor performance...

 

[Burakovsky95*]

Just wanna put it out there that I don't think these stats are to be worshipped or believed 100%. If it's of your personal belief that they are worthless that is completely fine.

 

[Hivemind]

When you look at the type of analysis conducted by the amateur bloggers (rather than the message board warriors), it can be some really impressive and thorough stuff, Ferrari's old blog included. I'm encouraged by the possibility of what they can do if it's their full time job. Especially as they gain access to the SportVu data (presumably before the public does), and scouts/data mining interns/game tape/etc in the interim.

As an aside, I find it amusing that the Flyers hired a ex-Capitals sales executive to head their stat department, as opposed to the blogger/Flyers fan who literally wrote the book on zone entries. Well, the paper anyway. To be fair, Tulsky was hired as a part-time consultant by an unnamed team. And I'm not so sure he'd give up his job in nanotech to take a hockey job.

 

[ChibiPooky]

So there's been some work done on zone entries and some work done on passes that lead directly to shot attempts. Since the stat community has largely determined that the best way to generate shot attempts is to gain the offensive zone with possession, I'd be quite interested in looking at passes that generate a controlled zone entry. I think there could be something meaningful there.

 

[Burakovsky95*]

So there's been some work done on zone entries and some work done on passes that lead directly to shot attempts. Since the stat community has largely determined that the best way to generate shot attempts is to gain the offensive zone with possession, I'd be quite interested in looking at passes that generate a controlled zone entry. I think there could be something meaningful there.

Corey Sznajder or @ShutdownLine is heading up that zone entries/exits project. It looks really interesting. Link to what exactly he's doing:

http://shutdownline.com/hurricanes/all-three-zones-project-aka-how-i-spent-my-summer-vacation.html

 

[BobRouse]

Absolutely love the hire. He's basically the father of these stats.

Great news. Caps killed it on this one.

 

[NobodyBeatsTheWiz]

You cant just say +/- = corsi but let me **** with +/- so that it becomes an actual useful stat. That isn't +/-.

Basic Corsi isn't an actual useful stat. It's the modified Corsis that become (a bit) more useful, given a large enough sample size. Comparing basic +/- to modified/developed Corsi isn't the right comparison to make.

I'm not sure what you're trying to say with your first point.

I'm saying that the stats don't take individual player actions into account, yet some want to use those stats to evaluate individual players.

Your 2nd point is just wrong as some of the leaders at D include Vlasic, Seabrook, Boychuk, Mitchell, Braun, Bouwmeester, Staal, Fayne, Rosival, Hjalmarsson, M. Greene, Methot, Jackman, Volchenkov, Hamhuis, etc.

It's not wrong. A shot attempt based stat is going to be biased against players that don't generate shot attempts, or are deployed in roles that don't encourage shot attempts.

Being the loser in any stat generally correlates to poor performance...

Hits, faceoff percentage, blocked shots, giveaways, takeaways, et al. don't really correlate to performance at all.

I'm not sure why you're trying so passionately to argue for certain statistics when you've got a general misunderstanding of the utility of statistics.

 

[Burakovsky95*]

Basic Corsi isn't an actual useful stat. It's the modified Corsis that become (a bit) more useful, given a large enough sample size. Comparing basic +/- to modified/developed Corsi isn't the right comparison to make.

I'm saying that the stats don't take individual player actions into account, yet some want to use those stats to evaluate individual players.


It's not wrong. A shot attempt based stat is going to be biased against players that don't generate shot attempts, or are deployed in roles that don't encourage shot attempts.
Hits, faceoff percentage, blocked shots, giveaways, takeaways, et al. don't really correlate to performance at all.

I'm not sure why you're trying so passionately to argue for certain statistics when you've got a general misunderstanding of the utility of statistics.

I'm not sure why you think I believe Corsi is the end all of the stats world. You can find many more intelligent people on the internet defend it as well. The same people getting hired by NHL organizations...

 

[SimplySensational]

It's not wrong. A shot attempt based stat is going to be biased against players that don't generate shot attempts, or are deployed in roles that don't encourage shot attempts.

But don't worry, anyone thing that doesn't correlate to corsi equates to corsi being stupid, but your argument is right even though the dman that considered the best all are good possession players.

Heck even Giordano was 53.9% on Calgary last year.

 

[NobodyBeatsTheWiz]

I'm not sure why you think I believe Corsi is the end all of the stats world. You can find many more intelligent people on the internet defend it as well. The same people getting hired by NHL organizations...

You sure post enough about Corsi to indicate such.

Feel free to find a single quote by people "getting hired by NHL organizations" that indicate Corsi (or any #fancystat) is a useful tool for evaluating players (or teams) in isolation or in small sample sizes.

The freaking creator of the stat says that he'd trust the eye test of hockey people over the stat.

 

[Dream Big]

I'm wading through a lot of info on Corsi and Fenwick. As far as I've been able to deduce is that the math is simple enough to calculate based on the data provided.

Some interesting things I've read:

a lot of giveaways is often good

http://grantland.com/the-triangle/the-fakers-guide-to-advanced-stats-in-the-nhl/


What I'm learning is how the original data upon which all the statistical manipulations are worked out could be complete junk.

The second obstacle is the accuracy of the data. Because statistics are compiled by human scorers at each rink, human error and bias enter into the picture, making it difficult to trust their accuracy. For instance, Colorado Avalanche scorekeepers tend to record significantly more shots on goal than league average, while the New Jersey Devils scorekeepers significantly undercount shots. Madison Square Garden scorekeepers consistently get shot location wrong, sometimes by as much as 20 feet, and NHL scorekeepers as a whole apparently can’t even decide where the faceoff dots are.

http://blogs.thescore.com/nhl/2012/06/26/the-next-step-in-hockey-analytics/

One of the problems with analytics is that the people doing them are taking their information from the event summaries provided by the NHL. Those summaries are done by human beings, all sorts of them, who might have a different interpretation of what just happened. Dallas Stars GM Jim Nill said it occurs often. He said the league will sometimes send out video of a disputed goal and ask 10 GMs if it should have counted. Five will come back saying it should count, the other five saying it should be disallowed. A giveaway or takeaway in one arena might not be one in another. Robert Svehla used to lead the league in hits when he played for the Florida Panthers, largely because he could count on being credited with double digits in hits every time he played a home game.

http://www.thehockeynews.com/blog/are-nhl-teams-warming-up-to-advanced-stats-like-corsi-and-fenwick/

Who are these people who are scorekeepers?
What is their training?
How do you become a scorekeeper?
Do they need to pass a test?
How many of them are there for each team?
Who is paying for their work?
Do they have their eye sight checked?

Seriously if all this time and effort is being spent on analytics the data in must be clean.

 

[Dream Big]

In such cases, this might be in play: http://en.wikipedia.org/wiki/Dunning–Kruger_effect

Thanks for the link. Just learned something new. Love learning. Keep it coming.

 

[g00n]

You sure post enough about Corsi to indicate such.

Feel free to find a single quote by people "getting hired by NHL organizations" that indicate Corsi (or any #fancystat) is a useful tool for evaluating players (or teams) in isolation or in small sample sizes.

The freaking creator of the stat says that he'd trust the eye test of hockey people over the stat.

I'm starting to feel like this is some kind of put-on.

 

[NobodyBeatsTheWiz]

I'm wading through a lot of info on Corsi and Fenwick. As far as I've been able to deduce is that the math is simple enough to calculate based on the data provided.

Some interesting things I've read:

http://grantland.com/the-triangle/the-fakers-guide-to-advanced-stats-in-the-nhl/


What I'm learning is how the original data upon which all the statistical manipulations are worked out could be complete junk.

http://blogs.thescore.com/nhl/2012/06/26/the-next-step-in-hockey-analytics/

http://www.thehockeynews.com/blog/are-nhl-teams-warming-up-to-advanced-stats-like-corsi-and-fenwick/

Who are these people who are scorekeepers?
What is their training?
How do you become a scorekeeper?
Do they need to pass a test?
How many of them are there for each team?
Who is paying for their work?
Do they have their eye sight checked?

Seriously if all this time and effort is being spent on analytics the data in must be clean.

The best way to overcome the inconsistencies in scoring/stat recording is to have huge sample sizes. That tends to eliminate the noise. Hence why using small sample sizes is completely useless.

 

[Dream Big]

The best way to overcome the inconsistencies in scoring/stat recording is to huge huge sample sizes. That tends to eliminate the noise. Hence why using small sample sizes is completely useless.

Thanks.

Any info on my questions about the scorekeepers qualifications etc.?

This analytics could become a full time job. I may just be a dabbler.

http://hfboards.mandatory.com/newreply.php?do=newreply&p=79203043

 

[usiel]

I'm wading through a lot of info on Corsi and Fenwick. As far as I've been able to deduce is that the math is simple enough to calculate based on the data provided.

Some interesting things I've read:

http://grantland.com/the-triangle/the-fakers-guide-to-advanced-stats-in-the-nhl/


What I'm learning is how the original data upon which all the statistical manipulations are worked out could be complete junk.

http://blogs.thescore.com/nhl/2012/06/26/the-next-step-in-hockey-analytics/

http://www.thehockeynews.com/blog/are-nhl-teams-warming-up-to-advanced-stats-like-corsi-and-fenwick/

Who are these people who are scorekeepers?
What is their training?
How do you become a scorekeeper?
Do they need to pass a test?
How many of them are there for each team?
Who is paying for their work?
Do they have their eye sight checked?

Seriously if all this time and effort is being spent on analytics the data in must be clean.

Definitely measuring the quality of the input data is where one should start before moving on to the methodology. Considering I have been providing technical support for an advanced spatial analytics software that analysis just data as well as spatial data for the past 13 years I will always tell users garbage in is garbage out. One has to weigh that. That said some garbage is better than no garbage at all. Just have to keep things in context in the big picture.

Honestly I keep looking at the potential of having RFID chips in players gear (and puck) being the ultimately goldmine data wise. So much awesomeness can come out of that data-wise.

 

[ChibiPooky]

Yes, the dirtiness and unreliability of the data is one reason everyone is so eagerly anticipating SportVU. RFID chips are another idea that makes a lot of sense.

 

[g00n]

Yes, the dirtiness and unreliability of the data is one reason everyone is so eagerly anticipating SportVU. RFID chips are another idea that makes a lot of sense.

That's all well and good but I think more needs to be done with context, too. Certain stats mean more for certain players in other sports, and it should be the same in hockey. Without getting too fancy...middle relievers in baseball are valued for holds. Closer, saves. And so forth. If you're looking at wins and other basic stats and not those specific to the position AND the role then you're missing important details.

Top line time should not be viewed the same as 4th, or even 2nd or 3rd. We can try to compensate for this with QOC and maybe zone starts but it still doesn't paint the whole picture based on the circumstance, what the coach wants from that player within the line, the score in the game at the time and if that affected strategy, and so on.

There is also the difference in repeatability. Compared to a sport like baseball that's perfect for large sample sizes and repeatable situations, hockey is chaos. There's constant ebb and flow and split second changes in matchups and advantages that make TRUE recurring situations much less frequent.

I don't think fancy stats will ever replace perception and the basic stats that win trophies for 99% of fans out there. The fancier stats are going to be more for those looking to trade or pay these guys because they need comparables. Any coach that waives a heart and soul player who's the glue of his line because his CF% is a few points off shouldn't be a coach for long.

 

[ChibiPooky]

That's what Barnes says is going to be the next big revolution. Give it some time. This is a field that is still very new to hockey.

 

[RandyHolt]

Like Flanders and his white bread with water for dippin', plus minus still rules the day for some.

Burakovsky +2 leads the team

 

[ChibiPooky]

Like Flanders and his white bread with water for dippin', plus minus still rules the day for some.

Burakovsky +2 leads the team

I think +/- is the worst stat ever, but at the end of every game I play I know what mine was. It's weird and frustrating.

 

[usiel]

Yes, the dirtiness and unreliability of the data is one reason everyone is so eagerly anticipating SportVU. RFID chips are another idea that makes a lot of sense.

SportVU from checking their website uses optical recognition to identify players through their jersey number. That seems needlessly difficult with a game like hockey with constant line changes. Substituting that with RFID chips unique to a player seems way more common sense tech-wise.

 

[Dream Big]

My concern with SportsVU is with the detail of analysis provided may make Oates' micromanagement look tame. Hockey may lose it's creativity and become robotic.

http://www.sportvu.com/PDFs/ESPN_62011.pdf

Another element execs love is that the cameras track the spacing between players, down to the inches. Teams combine that number with the result of the attempt -- made or missed -- to quantify how much room players ideally need to create, or defend, a shot.

 

[RandyHolt]

I think +/- is the worst stat ever, but at the end of every game I play I know what mine was. It's weird and frustrating.

LOL it is an tough stat to ignore. You finish a plus, you feel good.