Proposal: Statistical Analysis

91Kadri91* · Dec 17, 2015

The first, and most pressing, issue is the lack of 'Discussion' prefix; why isn't there a 'Discussion' tag? :laugh:

I was going to use the 'GDT' prefix, but I elected to use the 'Proposal' prefix, because that's (sort of) what this is. I'm proposing that this thread becomes the all-purpose statistics thread, where analysis for players across the league, new models, and questions regarding statistics can all 'take place'. I recognize (and respect) that not everyone wants to discuss statistics, so I figured this would be a good place to have arguments/discussion involving new (or old) models.

So what spurred this desire? Well, The Passing Project is releasing copious amounts of data tomorrow. TPP was started by Ryan Stimson as a way to measure a volitional, calculated skill:

That’s the first question I had before I even started tracking anything way back in October of 2013. What could I hope to learn from tracking passing plays? It started around the idea that as so much of hockey is random, passing represents a skillful play to link between players and generate shot attempts. By isolating that aspect of the game, you’re essentially isolating skilled, intentional plays made by the players. The other thing was a quote from Gavin Fleig in Soccernomics: "Teams that complete a higher number of passes in the final third consistently finish in the top four of the league."

So, that’s where it started. Could I find a passing metric that correlated well with teams winning games? Could isolating skilled plays teach us anything new about the game? The answers to both questions was a resounding "Yes."

http://www.allaboutthejersey.com/2015/5/20/8627957/2015-passing-project-data-release-volume-i

So how are (and which) passes tracked?

The issue with goals is that, while they're valid (goal differential is, at least), there's a ton of variance (unexplained variance that is often attributed to luck). Burtch did a study showing the reliability and correlation (to Pt% and Win%) a variety of different statistics had, and found that GF% had a YoY (year-over-year) correlation (R^2 based on 6 years of data) of 0.175:

Now look carefully at some of the other "important" yearly statistics. 5v5 PDO and the GF60 and GA60 stats. Notice how low their reliability scores are? That's because there's a large amount of variation in how much teams score or how many goals they allow year over year. This is because SH% and SV% are NOT repeatable, reliable statistics at the team level. Yes one player might be consistently good or consistently bad, but the randomness of all of his team-mates (the guy on the hot streak - the guy in a funk) has a way of balancing all of this out over the course of a season.

Corsi, meanwhile, is far more reliable, while offering some validity. The issue some have with Corsi is that it's not completely valid, nor is it entirely reliable. Consider this (very basic) explanation of what a statistic (or research) should 'have' to be considered sufficient:

While reliability is necessary, it alone is not sufficient. For a test to be reliable, it also needs to be valid. For example, if your scale is off by 5 lbs, it reads your weight every day with an excess of 5lbs. The scale is reliable because it consistently reports the same weight every day, but it is not valid because it adds 5lbs to your true weight. It is not a valid measure of your weight.

https://www.uni.edu/chfasoa/reliabilityandvalidity.htm

Another issue with goals is that they don't predict future goals due to the limited sample (understandable considering the lack of events is one of the reasons goal metrics are so unreliable, and why they're not a reasonable indication of player or team ability). Corsi is a vast improvement over goal metrics in this regard, but they're not great. A solution to this (a way to increase the validity of corsi and the reliability of goal metrics) is a Poisson model (expected GF-GF=xGF/EGF is the basic idea), where expected goals is calculated using a weighted goals and shots (shots from rebounds, shot angle, rush shots, volume etc) approach. This model shows a significant improvement over both Corsi and GF%:

That's pretty impressive, but it can still be better. How do passing metrics compare? It's hard to say right now, since the data set isn't particularly large for specific passing events, but the aggregate result (total passes) is predictive:

Here are some graphs showing the reliability of specific passing metrics (again, small sample):

Okay, so at the very least we can say that passing is reliable (and therefore indicative of a player's skill) even if we don't know how reliable specific types of passing are (although they appear to be reliable, even if the samples are still too small to make any definitive conclusions). How valid are they? Well... it appears that passing efficiency could be very valid (the sample is a little too small to state with absolute certainty):

Last season, I began tracking passes that generate shot attempts in an undertaking to explain how offense was being created. This led to various metrics to measure the efficiency of an individual player as well as a team when attempting passes. I name I ascribed to this metric was SAGE, Shot Attempt Generation Efficiency, and it was quite simple.

If a player completes two passes and the recipient attempts two shots on net, forcing a save and missing the net, the player making the pass had generated one shot on two attempts for a SAGE of 50%. Simple, yes?

I began keeping track of how often a team won various categories of efficiency and how often they also won games decided in regulation. The results through 82 games last season and 134 games this season are that it efficiency matters a great deal. Today, I wanted to offer an early answer on the relationship, if any, between passing (shot generation) efficiency and how many goals a team scores.

This is based on the tracking of five teams (Panthers, Devils, Rangers, Islanders and Blackhawks). Two of the teams didn't have all their games tracked. Here is the results using only the teams with all of their games tracked:

That's... almost perfect correlation. Again, the sample is small, but here's what the author has to say:

Well, isn’t that something? Through the first quarter of the season for the Devils, Panthers, and Blackhawks, there is nearly a perfect correlation between how efficient they are passing the puck and how many goals they score during 5v5 situations. Why could this be?

It’s my logical belief that a goalie has less time to get set and diagnose a situation when players make crisp, effective passes. We see this on a nightly basis when watching games. I don’t expect this correlation to remain at this level, but I also don’t expect to drop off significantly.

http://www.hockeyprospectus.com/the-pass-tracking-project-passing-and-goals/

So passing metrics may be more reliable than Corsi statistics, and just as valid as Goal metrics. That's incredibly exciting.

How, exactly, does passing data correlate to Corsi?

There's some relation, but it's far from perfect. Is it possible that amalgamating passing metrics and shot metrics (corsi, fenwick, shot attempts) could result in a complete picture (with both better reliability and validity)? Yes, many of the specific pass metrics are measured 'successful' when they result in a specific type of shot, but the shot is not accounted for. Both are a 'skill' (significantly less unexplained variances than goal metrics have), and accounting for both should improve the overall validity and reliability. As a result, there's a 'Passing Corsi' metric. Of course, these metrics can also show us who plays well together, and why:

It’s one thing for a player to shoot the puck. It’s another thing for them to generate offense from a pass. These are both valuable and, should be, essential means of player evaluation. If you’re only looking at a player’s individual shot attempts, you’re only getting half the picture. If you want a total picture, you need to look at both means of production.

Along that same line, certain players may combine with certain line mates better than others. So, if we can quantify how often Player A passes to Player B and vice versa, it allows to move closer to answering questions like, "Does Player A do something to improve Player B’s production? How does he do that? Is there a ‘chemistry’ between the two players? Is one simply easier or harder to play with?" Lots of questions like this come to mind when we look at players together and apart. That’s why I think tools like David Johnson’s WOWY and Super WOWYare of the utmost importance. How players play together and apart is incredible information. The passing linkups below simply take it a step closer to the ice level of what is actually occurring.

http://www.allaboutthejersey.com/2015/6/3/8694363/2015-passing-project-data-release-volume-ii

Okay, so that all sounds great, but where can you actually view these results? Well, you can view forward and defensemen data from last season. "But what about this season?" Well, you may recall that I referenced passing data being released tomorrow. While you're waiting, here are a few charts they've released that elucidate the capabilities of Leafs players (and the team) based on a 13 game sample:

View Forwards at the link: http://hockey-graphs.com/2015/12/16/toronto-maple-leafs-passing-lane-corsi/

View more at the link: http://hockey-graphs.com/2015/12/15/toronto-maple-leafs-passing-and-linkup-network/

91Kadri91* · Dec 17, 2015

Reimer's been very good, when he's played:

champs* · Dec 17, 2015

91Kadri91* · Dec 17, 2015

Taylor Hall is one of the best players in the NHL:

http://www.sportsnet.ca/hockey/nhl/why-taylor-hall-is-one-of-the-best-players-in-the-nhl/

91Kadri91* · Dec 17, 2015

'Dump and Chase' is not a good strategy:

https://jenlc13.wordpress.com/2015/09/30/back-to-basics-offensive-zone-entries/

Bomber0104 · Dec 17, 2015

Leave all this crap out of all the other threads and keep it all in here, and yeah I'm personally all for it.

Too many discussions get muddied with people throwing around meaningless junk data and then getting into debates about semantics.

91Kadri91* · Dec 17, 2015

Bomber0104 said:
Leave all this crap out of all the other threads and keep it all in here, and yeah I'm personally all for it.

Too many discussions get muddied with people throwing around meaningless junk data and then getting into debates about semantics.

Keep that crap in one thread.

I'm surprised you're not excited about the potential of passing data.

Wasn't your issue with Corsi that it lacked the validity of Goal metrics? The early results of passing metrics suggest that it will almost match the validity of Goal metrics, while being (potentially significantly) more reliable than Corsi (as an independent measurement), and significantly more reliable than Goal metrics.

Bomber0104 · Dec 18, 2015

91Kadri91 said:
I'm surprised you're not excited about the potential of passing data.

Wasn't your issue with Corsi that it lacked the validity of Goal metrics? The early results of passing metrics suggest that it will almost match the validity of Goal metrics, while being (potentially significantly) more reliable than Corsi (as an independent measurement), and significantly more reliable than Goal metrics.

No.

My issue is the inferences being made based on shot-based analysis.

There's a whole sub-forum dedicated to this junk.

Rather than littering this forum with this amateur analysis, perhaps you should be taking it there.

http://hfboards.mandatory.com/forumdisplay.php?f=241

I'm sure it will be a hit.

Bomber0104 · Dec 18, 2015

Babcock on Corsi/Fenwick/shot-based metrics:

“Don’t get me wrong,” Babcock said. “There’s some parts that I think are real valuable.”

“I think if you want to base signing players on shots, I think you’re nuts,” Babcock said. “And I think it doesn’t tell the story, so to me, it’s like all stats, stats can tell any story you want.

“You’ve got to be able to be smart enough to understand information from it. I’ve been very impressed by the kid we brought in to work with our team (Andrew Brewer) because he’s found us some stuff we didn’t think about. But it wasn’t based on shot clock or stats that don’t even tell the truth.”

http://windsorstar.com/sports/hockey/fancy-this-wings-wont-analyze-that

91Kadri91* · Dec 18, 2015

Bomber0104 said:
No.

My issue is the inferences being made based on shot-based analysis.

Right, the inference that it correlates to Win% (or that it's a winning strategy), which is validity (since, in this case, the statistic must correlate to Win% or, at the very least, goals).

There's a whole sub-forum dedicated to this junk.

Maybe take it there.

http://hfboards.mandatory.com/forumdisplay.php?f=241

I'm sure it will be a hit.

That kind of defeats (part of) the purpose. I'm hoping to be able to talk about statistics in-game (Marlies tracking, Leafs tracking, London/Marner tracking), if the Mods are okay with it. I know that not everyone is interested in this stuff, so I thought it would be less contentious if I made a separate thread to post data collection in.

LaPlante94 · Dec 18, 2015

91Kadri91 said:
'Dump and Chase' is not a good strategy:

https://jenlc13.wordpress.com/2015/09/30/back-to-basics-offensive-zone-entries/

I disagree/agree with this. Dumping it in to do a line change is a waste and very dumb as Don Cherry has said in the past, but it is a good idea to do it if teams are gonna stack the blue line. Carry it in every time if you have the chance but it isn't the worst idea to dump it in and chase.

Also, is their any books explaining these kind of statistics to get a better understanding for them? Wouldn't mind reading something like that. Not a big fan of reading long articles on computers.

Bomber0104 · Dec 18, 2015

Measuring passing will necessarily be a subjective, qualitative measure which will therefore require a set of eyes analyzing the play to make a measurement.

Welcome to the "eye-test" 91Kadri91, the very thing the blogging community so vehemently rejects..

Bomber0104 · Dec 18, 2015

91Kadri91 said:
Right, the inference that it correlates to Win% (or that it's a winning strategy), which is validity (since, in this case, the statistic must correlate to Win% or, at the very least, goals).

Goal differentials correlate to winning.

Goals correlate to scoring chances.

Scoring chances correlate to shot selection and quality passing.

Shot selection and quality passing correlate with puck possession.

Everyone knows this.

But you attempt to jump from one extreme to the other with literally nothing but a very weak correlation to fall back on.

That kind of defeats (part of) the purpose. I'm hoping to be able to talk about statistics in-game (Marlies tracking, Leafs tracking, London/Marner tracking), if the Mods are okay with it. I know that not everyone is interested in this stuff, so I thought it would be less contentious if I made a separate thread to post data collection in.

Make a Toronto Maple Leafs study in that forum then....

No one wants to see your hobby project on the front page here which demonstrates nothing useful.

91Kadri91* · Dec 18, 2015

Bomber0104 said:
Babcock on Corsi/Fenwick/shot-based metrics:

http://windsorstar.com/sports/hockey/fancy-this-wings-wont-analyze-that

Corsi's technically (supposed to be) a possession proxy:

I see your point, though, which is why I'm excited about statistics that are better (and separate) measurements (in terms of their relation to Pt%), such as zone entries:

What's particularly exciting is that these correlation improve when you 'combine' Corsi and Zone data:

The more data that becomes available, the better the statistics will become at determining a 'winning strategy'.

Daisy Jane · Dec 18, 2015

91Kadri - you are v. smart.
and you have a way of making me feel v. dumb. LOL

i am going to have to re-read this really, really, really slowly.

Bomber0104 · Dec 18, 2015

91Kadri91 said:
Corsi's technically (supposed to be) a possession proxy:

The more data that becomes available, the better the statistics will become at determining a 'winning strategy'.

All of these demonstrate weak-to-moderate correlations between your variables, with some severe outliers hanging around (which are conveniently skewing correlation in a favorable way....I wonder why? lol) ...

Why do you post graphs and waste your time when the information demonstrates nothing?

91Kadri91* · Dec 18, 2015

Bomber0104 said:
Goal differentials correlate to winning.

But goals aren't a skill. They're not indicative of any sort of player or team ability on their own; goals don't beget goals.

Goals correlate to scoring chances.

Yes, moderately.

Scoring chances correlate to shot selection and quality passing.

Yes.

Shot selection and quality passing correlate with puck possession.

As do LPR's, zone entries/exits/defense etc.

Everyone knows this.

You have yet to get to a point.

But you attempt to jump from one extreme to the other with literally nothing but a very weak correlation to fall back on.

Passing efficiency has (on the low-end) an R^2 (with GF%) of 0.83; that's very strong.

Make a Toronto Maple Leafs study in that forum then....

No one wants to see your hobby project on the front page here which demonstrates nothing useful.

Other than objective, substantiated discussion? Yeah, who wants that? I mean, you don't, but that's because (when you're in this thread) you'd have to post a thought provoking response/concept.

indigobuffalo · Dec 18, 2015

LaPlante94 said:
I disagree/agree with this. Dumping it in to do a line change is a waste and very dumb as Don Cherry has said in the past, but it is a good idea to do it if teams are gonna stack the blue line. Carry it in every time if you have the chance but it isn't the worst idea to dump it in and chase.

Also, is their any books explaining these kind of statistics to get a better understanding for them? Wouldn't mind reading something like that. Not a big fan of reading long articles on computers.

It's like passing vs. rushing plays in football.

Passing is much better at getting downfield, but if it's all you do, the opposing team will just pull defenders back to increase coverage downfield.

But it leaves them vulnerable to a rushing play, so you need a balance of both to be successful.

Kinda wonder to myself why they don't have helmets mic'ed up and have an assistant coach watching from above directing players whether to dump and chase or use a controlled zone entry based on the defensive structure in the neutral zone.

Bomber0104 · Dec 18, 2015

91Kadri91 said:
Other than objective, substantiated discussion? Yeah, who wants that? I mean, you don't, but that's because (when you're in this thread) you'd have to post a thought provoking response/concept.

Substantiated with what?

Junk data, with junk correlations?

Glad you consider that objective....

Someone who values good statistics and good information won't.

Kurtz · Dec 18, 2015

91Kadri91 said:
This is based on the tracking of five teams (Panthers, Devils, Rangers, Islanders and Blackhawks). Two of the teams didn't have all their games tracked. Here is the results using only the teams with all of their games tracked:

That's... almost perfect correlation. Again, the sample is small, but here's what the author has to say:

I've gotta say, a correlation coefficient of 0.998 is pretty insane. So insane in fact, that there might be an issue with how the variables were calculated. I'm not sure if it's a sample size issue if he sampled 5 teams at 20 games each...that's a pretty reasonable sample size. But I mean...I just can't think of two variables in ANY walk of life where the coefficient of correlation is that high.

4thline · Dec 18, 2015

indigobuffalo said:
It's like passing vs. rushing plays in football.

Passing is much better at getting downfield, but if it's all you do, the opposing team will just pull defenders back to increase coverage downfield.

But it leaves them vulnerable to a rushing play, so you need a balance of both to be successful.

Kinda wonder to myself why they don't have helmets mic'ed up and have an assistant coach watching from above directing players whether to dump and chase or use a controlled zone entry based on the defensive structure in the neutral zone.

One of my favourite aspects of hockey and things I'm most proud of between sports is the cerebral ability it takes to play. To keep track of 10 players with unrestricted motion and act as a unit to counteract them trying to counteract you. It's beautiful. The ability to make these reads is a skill, I want no part of during the play instruction and micromanaging.

Mitchy · Dec 18, 2015

Bomber0104 said:
Goal differentials correlate to winning.

Goals correlate to scoring chances.

Scoring chances correlate to shot selection and quality passing.

Shot selection and quality passing correlate with puck possession.

Everyone knows this.

But you attempt to jump from one extreme to the other with literally nothing but a very weak correlation to fall back on.

Make a Toronto Maple Leafs study in that forum then....

No one wants to see your hobby project on the front page here which demonstrates nothing useful.

Speak for yourself. I, for one, very much invite this type of content. Keep going 91Kadri91 :handclap:

.

Bomber0104 · Dec 18, 2015

91Kadri91 said:
But goals aren't a skill. They're not indicative of any sort of player or team ability on their own; goals don't beget goals.

Scoring goals is not a skill.

Excellent evaluation of the game of hockey!

Yes, moderately.

What else would correlate better than scoring chances to goal production?

The amount of popcorn sold in the stands?

Really, what variable on the ice would correlate better with goal-scoring than scoring chances?

Let me guess...

Corsi's!!!

Yes.

Yes?

Ok then..

As do LPR's, zone entries/exits/defense etc.

Yes.

You have yet to get to a point.

Passing efficiency has (on the low-end) an R^2 (with GF%) of 0.83; that's very strong.

So I see you just ignored what I said entirely and just regurgitated a random tidbit of information...

Once again...

Shot-based proxy possession numbers don't correlate with winning.

Goal differential does. That's how games are won. The team who outscores the other wins.

How do goals get scored? Through scoring chances. The team with more quality scoring chances score goals.

This is common sense but instead guys like you prefer to chalk hockey up to a game of keep-away and, for whatever reason (I'm chalking it up to a lack of statistical understanding) like using weak correlations as some sort of "proof" that playing keep away is conducive to winning.

Perhaps you need to go back to the drawing board.

91Kadri91* · Dec 18, 2015

Bomber0104 said:
Measuring passing will necessarily be a subjective, qualitative measure which will therefore require a set of eyes analyzing the play to make a measurement.

Welcome to the "eye-test" 91Kadri91, the very thing the blogging community so vehemently rejects..

It's the eye-test with a defined objective, and conscious understanding of what is being analyzed, not a bunch of disgruntled Men mashing their keyboards in frustration. You're watching the game with a complete lack of understanding, and with the thought that you're capable of processing the minutiae (and total events) of every game. You look at a bad pass and claim that the player sucks, even if said player made 10 sublime plays prior to the gaff.

There's a difference between 'tracking' and 'watching'.

Bomber0104 · Dec 18, 2015

Kurtz said:
I've gotta say, a correlation coefficient of 0.998 is pretty insane. So insane in fact, that there might be an issue with how the variables were calculated. I'm not sure if it's a sample size issue if he sampled 5 teams at 20 games each...that's a pretty reasonable sample size. But I mean...I just can't think of two variables in ANY walk of life where the coefficient of correlation is that high.

A "perfect" correlation containing three data points.

:laugh:

Get this...for the past three days I made toast with whole-wheat bread and my toaster didn't short-circuit.

I guess you can say that toasters are more reliable when you use whole-wheat bread, right?

Proposal: Statistical Analysis

91Kadri91*

Guest

91Kadri91*

Guest

champs*

Guest

91Kadri91*

Guest

91Kadri91*

Guest

Registered User

91Kadri91*

Guest

Registered User

Registered User

91Kadri91*

Guest

Registered User

Registered User

Registered User

91Kadri91*

Guest

everything is gonna be okay!

Registered User

91Kadri91*

Guest

Portage and Main

Registered User

Registered User

Registered User

#HFOutcasts

Registered User

91Kadri91*

Guest

Registered User

Ad

Ad

Ad