News Article: "Fun With Numbers" - Advanced Stats Talk Here

[dumbdick]

Kind of slow in here today.

If anyone has some free time and wants to read about what a critical approach to analytics looks like, come visit www.integratinghockeyanalysis.com

Comments and feedback are always welcome.

Hey man, sorry to read about your daughter. Hope things are going well. I've been learning some data science techniques and was thinking of setting up a blog similar to your own this year, basically as a playground to test out the techniques and maybe as a personal career portfolio piece for myself in the future. If I get it off the ground, maybe we can collaborate at some point in the future. A Sensalytics alliance, of sorts.

On a side note, check out these cool computer vision applications to hockey http://www.cs.ubc.ca/~shervmt/. This isn't my work, but this is the kind of stuff I'm hoping to get to eventually.

 

[StefanW]

Hey man, sorry to read about your daughter. Hope things are going well. I've been learning some data science techniques and was thinking of setting up a blog similar to your own this year, basically as a playground to test out the techniques and maybe as a personal career portfolio piece for myself in the future. If I get it off the ground, maybe we can collaborate at some point in the future. A Sensalytics alliance, of sorts.

On a side note, check out these cool computer vision applications to hockey http://www.cs.ubc.ca/~shervmt/. This isn't my work, but this is the kind of stuff I'm hoping to get to eventually.

Sure thing, I think a collaboration down the road would be fantastic. Let me know when you are getting your stuff off the ground.

I think you'll appreciate this link.

 

[StefanW]

Hockey analytics conference coming up at Carleton U on Feb 7th.

http://statsportsconsulting.com/ottanalytics/

 

[starling]

Analytics conference in Ottawa Feb 7

If you are interested in advanced statistics there will be a great hockey analytics event at Carleton on Feb 7.
Should be pretty interesting for stats nerds. And it's free.
http://statsportsconsulting.com/ottanalytics/

 

[Vesa Awesaka]

If you are interested in advanced statistics there will be a great hockey analytics event at Carleton on Feb 7.
Should be pretty interesting for stats nerds. And it's free.
http://statsportsconsulting.com/ottanalytics/

"Corsi is the end all of advanced stats and should be accepted unanimously"

"You fool Fenwick is the far superior stat. Blocked shots are everything"

followed by a mob of purists outside denouncing knowledge as the devils works

Would be interesting to see a crime happen at the conference.

" Ten people saw that this man was shot by this man and the gun was found lying at his feet"

"We're going to have investigate this further. The Marshall stat gives us a 50 percent chance based on the angle of the gun on the floor that is was dropped by the man and when you include the quality of competition of the bullet and zone deployment of the alleged shooter he clearly didnt do it."

 

[OgieO]

Pretty cool site doing some work with advanced stats. I was attracted to it because it got into shot quality http://war-on-ice.com/hexteams.html

No time to offer much input now - gist of what's linked:
1. Ottawa allows higher rate of shots from the outside relative to other teams.
2. Ottawa has a terrible shooting % on outside shots relative to other teams.
3. On the flip side, opponents have a significantly higher shooting % on outside shots versus Ottawa relative to the same shots against other teams.

Two (and a half) conclusions I get from this:
1. Our shots against are overstated as many come from low % areas.
2. We need to do better making life difficult on goalies from outside shots.
21/2. Perhaps we've been a bit unlucky this year.

 

[starling]

Pretty cool site doing some work with advanced stats. I was attracted to it because it got into shot quality http://war-on-ice.com/hexteams.html

No time to offer much input now - gist of what's linked:
1. Ottawa allows higher rate of shots from the outside relative to other teams.
2. Ottawa has a terrible shooting % on outside shots relative to other teams.
3. On the flip side, opponents have a significantly higher shooting % on outside shots versus Ottawa relative to the same shots against other teams.

Two (and a half) conclusions I get from this:
1. Our shots against are overstated as many come from low % areas.
2. We need to do better making life difficult on goalies from outside shots.
21/2. Perhaps we've been a bit unlucky this year.

Yeah it's the best site for advanced stats right now. Its authors will be speaking at that conference in Ottawa on Feb 7.

 

[StefanW]

Integrating Hockey Analysis Wrapping Up Soon

I started my blog, Integrating Hockey Analysis, 11 months ago with a specific goal in mind, which is to come up with a variety of ways in which "old school" understanding of hockey and contemporary analytics can be integrated. I figured that each side had some good points, and picking one side or the other in the debate seemed limiting.

At this point I am coming to the end of the project, and I am not planning to renew the web site. I just put a post called "NHL Teams as Complex Systems," and my last post will be used to basically pull all of the strands of my argument back together into one story. Once that is done I will putter away at the Tank-Alytics page I have been working on until the blog goes dark in the last part of March (look at the top left of the web page and you'll see the tab).

Just for the record, I have tried to be respectful of the discussions here, and I have only posted links to 5 out of 30+ posts. I figured that if a post does not fit into the discussion, then why try to wedge it in here? The only exception I made to that rule was when I started a thread devoted to a post about taking my daughter to a Sens game. That seemed like the right thing to do, and the support I got here was amazing.

If you are interested in reading my post on "NHL Teams as Complex Systems" you can find it at:
www.integratinghockeyanalysis.com

I plan to put up one last post announcement here when my last post is done. Thanks for the feedback and support you have given me in this project.

 

[God Says No]

Cool stuff. I'll bookmark it.

 

[Vesa Awesaka]

I read some your blog and i found some of it really cool Stephan. I can see that its something you really enjoy digging into and i'm glad it makes you happy.

This may sound like an odd request so ill pose it more of an idea for all those who might want to take a could hours to go through data and make a graph...
How about a corsi/timeline line graph of karlsson's play this season.

x- variable= games

y= corsi




Add to the graph some context like most common linemate over a certain amount of games, goals/point for each game, Cameron's take over and Methot's return.

Now i'm not demanding anyone do this and i'm pretty sure most of us can figure out how the chart would look just based on watching games but it was an idea i thought of on my morning jog and i thought it would be pretty cool illatration in helping people understand the coaching impact, linemate impact and karlsson's overall play.

Note i feel that the data should be all situations not just 5 v 5 although im sure others might feel differently.

 

[YouGotAStuGoing]

I started my blog, Integrating Hockey Analysis, 11 months ago with a specific goal in mind, which is to come up with a variety of ways in which "old school" understanding of hockey and contemporary analytics can be integrated. I figured that each side had some good points, and picking one side or the other in the debate seemed limiting.

At this point I am coming to the end of the project, and I am not planning to renew the web site. I just put a post called "NHL Teams as Complex Systems," and my last post will be used to basically pull all of the strands of my argument back together into one story. Once that is done I will putter away at the Tank-Alytics page I have been working on until the blog goes dark in the last part of March (look at the top left of the web page and you'll see the tab).

Just for the record, I have tried to be respectful of the discussions here, and I have only posted links to 5 out of 30+ posts. I figured that if a post does not fit into the discussion, then why try to wedge it in here? The only exception I made to that rule was when I started a thread devoted to a post about taking my daughter to a Sens game. That seemed like the right thing to do, and the support I got here was amazing.

If you are interested in reading my post on "NHL Teams as Complex Systems" you can find it at:
www.integratinghockeyanalysis.com

I plan to put up one last post announcement here when my last post is done. Thanks for the feedback and support you have given me in this project.

It's been a pleasure to read your insights on analytics, Stef. Always an interesting read.

 

[Micklebot]

I read some your blog and i found some of it really cool Stephan. I can see that its something you really enjoy digging into and i'm glad it makes you happy.

This may sound like an odd request so ill pose it more of an idea for all those who might want to take a could hours to go through data and make a graph...
How about a corsi/timeline line graph of karlsson's play this season.

x- variable= games

y= corsi




Add to the graph some context like most common linemate over a certain amount of games, goals/point for each game, Cameron's take over and Methot's return.

Now i'm not demanding anyone do this and i'm pretty sure most of us can figure out how the chart would look just based on watching games but it was an idea i thought of on my morning jog and i thought it would be pretty cool illatration in helping people understand the coaching impact, linemate impact and karlsson's overall play.

Note i feel that the data should be all situations not just 5 v 5 although im sure others might feel differently.

Well, I think you can do most of that with waronice, but it's not loading properly on my end so I can't confirm.

Try this link, and hit the tabular by game tab.

http://war-on-ice.com/playerseason....=29&caxis=39&saxis=87&xaxisg=&yaxisg=&panel=1

As for using all the situations vice just 5vs5, I'm not sure what benefit including PP and (limited) PK time would have. It would only serve to muddy the data in my opinion, as Corsi is always much higher on the PP (and lower on the PK). Since Karlsson's ice time is skewed towards the PP (4+ mins/g PP vice 45 sec of PK) it would just artificially inflate his corsi numbers. If he happened to be getting more pp time in a segment of games, his game by game CF% would go up, but it wouldn't actually reflect his on ice play, just his usage.

 

[God Says No]

Can anyone explain how "relative goals for percentage of total" works?

The thread on Morgan Reilly has piqued my interest to see if he's actually playing better defense vs Ceci and Karlsson. Especially the last 20 games.

I got this:
http://war-on-ice.com/playertable.html?mansit=3&scoresit=8&homeawaysit=1&shotattsit=1&names=morgan,0cody0ceci,0erik0karlsson&team=&pos=1&start1=2014-12-14&xaxis=50&yaxis=80&caxis=39&saxis=87&mintoi=0&panel=2&usedaterange=1&start0=20142015&end0=20142015&end1=2015-02-04&splitseasons=0

Which I *think* shows that Reilly has been slightly better defensively than Ceci, but Karlsson blows both of them out of the water.

But I was perplexed by "relative goals for percentage of total" stat.

 

[starling]

Can anyone explain how "relative goals for percentage of total" works?

The thread on Morgan Reilly has piqued my interest to see if he's actually playing better defense vs Ceci and Karlsson. Especially the last 20 games.

I got this:
http://war-on-ice.com/playertable.html?mansit=3&scoresit=8&homeawaysit=1&shotattsit=1&names=morgan,0cody0ceci,0erik0karlsson&team=&pos=1&start1=2014-12-14&xaxis=50&yaxis=80&caxis=39&saxis=87&mintoi=0&panel=2&usedaterange=1&start0=20142015&end0=20142015&end1=2015-02-04&splitseasons=0

Which I *think* shows that Reilly has been slightly better defensively than Ceci, but Karlsson blows both of them out of the water.

But I was perplexed by "relative goals for percentage of total" stat.

"Relative" shows how the player did when compared to other players on his team. Rielly is one of the sheltered players on leafs roster (weaker opponents, more offensive zone starts), that's where his above average GF% or CF% come from.
Ceci is not sheltered in any way, hence slightly lower than average %%.

 

[God Says No]

I understand the relative part. That's not the question. I don't get what "relative goals for percentage of total" means. For example Ceci is at -12.3. What does that metric really calculate?

As for the sheltered part. I don't see that in the numbers. Care to expand?

 

[Micklebot]

Can anyone explain how "relative goals for percentage of total" works?

The thread on Morgan Reilly has piqued my interest to see if he's actually playing better defense vs Ceci and Karlsson. Especially the last 20 games.

I got this:
http://war-on-ice.com/playertable.html?mansit=3&scoresit=8&homeawaysit=1&shotattsit=1&names=morgan,0cody0ceci,0erik0karlsson&team=&pos=1&start1=2014-12-14&xaxis=50&yaxis=80&caxis=39&saxis=87&mintoi=0&panel=2&usedaterange=1&start0=20142015&end0=20142015&end1=2015-02-04&splitseasons=0

Which I *think* shows that Reilly has been slightly better defensively than Ceci, but Karlsson blows both of them out of the water.

But I was perplexed by "relative goals for percentage of total" stat.

My understanding is its: on ice GF% - off ice GF% = Rel GF %

So, if Reilly has a on ice GF% of 49% and Toronto has a GF% of 42% when Reilly is not on the ice, then Reilly's Rel GF% is +7%

Now, that's how most re stats work, but I haven't specifically looked into how Waronice does theirs. They may use total team GF% instead or Reilly's Off ice GF% but you get the general idea.

 

[God Says No]

My understanding is its: on ice GF% - off ice GF% = Rel GF %

So, if Reilly has a on ice GF% of 49% and Toronto has a GF% of 42% when Reilly is not on the ice, then Reilly's Rel GF% is +7%

Now, that's how most re stats work, but I haven't specifically looked into how Waronice does theirs. They may use total team GF% instead or Reilly's Off ice GF% but you get the general idea.

Ok. That makes sense. I looked into the appendix, but I couldn't find a definition of that metric.

So, would you say just by looking at the last 20 games, would you say my assessment that Ceci and Reilly are pretty close defensively, with Reilly having a slight edge and Karlsson blowing both of the out of the water be accurate?

Also, I don't see where Reilly is sheltered in the last 20 games. Maybe I'm looking at the wrong metrics. Thanks in advance.

 

[Micklebot]

Ok. That makes sense. I looked into the appendix, but I couldn't find a definition of that metric.

So, would you say just by looking at the last 20 games, would you say my assessment that Ceci and Reilly are pretty close defensively, with Reilly having a slight edge and Karlsson blowing both of the out of the water be accurate?

Also, I don't see where Reilly is sheltered in the last 20 games. Maybe I'm looking at the wrong metrics. Thanks in advance.

Defensive play is just about the hardest thing to measure with stats. Idk that I'd hazzard a guess as I really don't watch much Toronto these days. That said, CF% and GF% are a mix of offensive and defensive play, so I guess you could narrow it down to just the GA and CA per 60, but that comes with it's own issues, as different teams play different tempo of hockey though Goal based stats really aren't that stable after just 20 games.

With regards to being sheltered; Reilly takes more OZ starts relative to his team than Ceci by a substantial margin. That's probably where you'd see him being labelled sheltered. Now, the advantage he'd get over Ceci because of starting in the OZ more often that other Toronto players isn't much, because in raw numbers they'd be about even, but in regards of deployment, his coach is not being using him for defensive draws in the way you might argue ceci is being used.

 

[StefanW]

Cool stuff. I'll bookmark it.

I read some your blog and i found some of it really cool Stephan. I can see that its something you really enjoy digging into and i'm glad it makes you happy.

It's been a pleasure to read your insights on analytics, Stef. Always an interesting read.

Thanks guys It was fun working on it over the past 11 months.

 

[DrEasy]

If you are interested in advanced statistics there will be a great hockey analytics event at Carleton on Feb 7.
Should be pretty interesting for stats nerds. And it's free.
http://statsportsconsulting.com/ottanalytics/

It's happening right now, and I'm attending (I'm the nerdy-looking male; you can't miss me )! I took a few notes from the first few presentations (it's lunch break right now). It's rough but it's not nothing:

===

Rob Vollmer on past, present and future of hockey analytics. Need to check out his book.

RTSS: real time statistics (hits, blocked shots, etc.) were introduced in 97-98

2001-2004: focus on the WAR equivalents for hockey

Vollmer thinks Nashville's recent moves look very analytics driven

---

David Johnson (hockeyanalytics.com and puckalytics.com), creator of WOWY (With Or Without You) which looks at stats based on who you play with.

uses NHL.com stats as source (play by play etc.)

but NHL.com is manually entered data, with mistakes and omissions.

Interesting stuff about Voracek's WOWY (he didn't get better, he just got more opportunity); contrast with Jack Johnson.

Bobby Ryan's WOWY FTW!

---
Andrew Thomas, war-on-ice guy

- uses x, y coordinates from ESPN, sportsnet.ca

- wants to incorporate fan data (Passing Project)

---

Josh Weissbock, analytics in dev leagues (CHLstats.com)

PPG beats scouting reports for draft success prediction

looks like height matters too (Aragorn!)

---

panel: difficulty tracking real time data, errors, rink bias...

---

That's it for now, but if there's interest here I can try to take more thorough notes (check out the full program on the web site mentioned above). There may also be some live-tweeting going on (not me; I'm not cool enough to tweet), look for #OTTanalytics.

 

[DrEasy]

Update from Session #2 of the conference. This didn't get any love, so I kept my note-taking light.

---

Sam Ventura (CMU!), another war on ice guy

Corsi good because uses lots of events/info. Scoring chances good because focuses on most dangerous events. Idea: blend the two, use weights.

uses distance from goal, shot type, etc. obtains EGF (Expected Goals For), and EGA

correlates well to predict future EGF (from one season to the next), a little bit less for predicting future goals for GF% but does better than Corsi. True for teams and for individual players.

---

Alex Diaz, Carleton grad student

Infer shifts and penalty expirations from data.

uses it as input for market basket analysis to look for correlated events

applied to leafs pairings (Bozak effect, Phaneuf)

---

Matt Cane, puckplusplus.com

- weighting corsi components, but by predictive power (as opposed to net distance for example). shot that scores weighted more than shot that misses.

Goal against component more predictive for forwards than defensemen.

Weighted shots better predictor of FG% than Corsi.

---

Stephen Burtch

effects of coaching and changes within season

Jump in CF% and SCF% from MacLean to Cameron

under Cameron players perform better, better identification of roles for each.

Leafs: Horachek also improved CF% and SCF% from Carlyle.

---

Emmanuel Perry

Microstats, such as zone entries, which are factors in macroscopic stats such as cross.

case of misalignment between perceptions and stats: Cody Ceci. Explained by being team worst at allowing controlled entries against.

also looked at GSN line (but had to skip this for lack of time unfortch)

looked at corsi vs time spent in o-zone or d-zone. Boro and Condra get to the same result despite different corsi.

application of zone entry: auto-detect rushes, evaluate pp, descriptive/narrative potential.

---

Andrew Berkshire (Eyes on the prize)

on fan opinion and how it's influenced by tv commentary.

traditional defensive stats : blocked shots and hits but mean you don't have the puck

pb: the more you have the puck, the more you turn it over. leads to wrong perception: applies to Subban or Karlsson.

PK Subban "fixed"? turns out he's pretty much the same guy from rookie till now. And actually got worse in rel shots against. Replaced instinctive play with dumping the puck.

---

Panel: Andrew says Drew Doughty is overrated. His relative numbers (relative to his team) don't pop out. See Muzzin instead. Subban and Karlsson are better.

One more session left to go!

 

[Caeldan]

Thanks for the updates!

 

[Micklebot]

Update from Session #2 of the conference. This didn't get any love, so I kept my note-taking light.

---

Sam Ventura (CMU!), another war on ice guy

Corsi good because uses lots of events/info. Scoring chances good because focuses on most dangerous events. Idea: blend the two, use weights.

uses distance from goal, shot type, etc. obtains EGF (Expected Goals For), and EGA

correlates well to predict future EGF (from one season to the next), a little bit less for predicting future goals for GF% but does better than Corsi. True for teams and for individual players.

---

Alex Diaz, Carleton grad student

Infer shifts and penalty expirations from data.

uses it as input for market basket analysis to look for correlated events

applied to leafs pairings (Bozak effect, Phaneuf)

---

Matt Cane, puckplusplus.com

- weighting corsi components, but by predictive power (as opposed to net distance for example). shot that scores weighted more than shot that misses.

Goal against component more predictive for forwards than defensemen.

Weighted shots better predictor of FG% than Corsi.

---

Stephen Burtch

effects of coaching and changes within season

Jump in CF% and SCF% from MacLean to Cameron

under Cameron players perform better, better identification of roles for each.

Leafs: Horachek also improved CF% and SCF% from Carlyle.

---

Emmanuel Perry

Microstats, such as zone entries, which are factors in macroscopic stats such as cross.

case of misalignment between perceptions and stats: Cody Ceci. Explained by being team worst at allowing controlled entries against.

also looked at GSN line (but had to skip this for lack of time unfortch)

looked at corsi vs time spent in o-zone or d-zone. Boro and Condra get to the same result despite different corsi.

application of zone entry: auto-detect rushes, evaluate pp, descriptive/narrative potential.

---

Andrew Berkshire (Eyes on the prize)

on fan opinion and how it's influenced by tv commentary.

traditional defensive stats : blocked shots and hits but mean you don't have the puck

pb: the more you have the puck, the more you turn it over. leads to wrong perception: applies to Subban or Karlsson.

PK Subban "fixed"? turns out he's pretty much the same guy from rookie till now. And actually got worse in rel shots against. Replaced instinctive play with dumping the puck.

---

Panel: Andrew says Drew Doughty is overrated. His relative numbers (relative to his team) don't pop out. See Muzzin instead. Subban and Karlsson are better.

One more session left to go!

From all of us who didn't have a chance to go (dang responsibilities), thanks a bunch, much appreciated.

 

[dumbdick]

any talk of machine learning at this thing?

 

[DrEasy]

any talk of machine learning at this thing?

Yes, one guy from Carleton used association rule learning a.k.a market basket analysis, as mentioned above.