Understanding analytics

[dudacek]

Thought we had a thread for this but I couldn’t find it.

I remember when +/- was in vogue as a player debate tool and I remember the campaign to discredit it.

• “How can we put any stock Into that stat when they give Sam Reinhart a plus on a goal when all he did was hop over the boards”
• ”Like Lazar deserved a minus there. That stupid pass was all on Montour.”
• ”Of course Risto’s going to be a minus, he gets the hardest ice time on a bad team”
• ”Poor McCabe is a -3 because Hutton couldn’t stop a beach ball”

Basically, the argument - and it’s a sound one - is that there are way too many factors going on during a hockey game to ever use +/- as a definitive indicator of how “good” a player is.

I know advanced stats are an attempt to address that, but I get the sense sometimes that a similar argument can be applied to most advanced statistics.

I would like to discuss that here.

Edited February 28, 2021 by dudacek

[mjd1001]

there are faults in +/- for sure, but it is not totally useless in my mind. Sure, a good number of the pluses and minuses may be noise, but just as many of them (if not more) might not be.  The way I look at it...it should not be used as a 'tell all' stat, but it can show you some trends...especially if you look at it in context to other things with it.

[dudacek]

This thread was partially inspired by a snarky comment I made on the game day thread yesterday about a shot that Skinner made off a rush that was two feet wide.

I want to remove the snark and I don’t want this to be about Skinner, but I do want to use the example to start the discussion about stats based on shot attempts.

I think the situation was largely one where the defender made an excellent play to neutralize what could have been a good scoring chance, by essentially forcing Jeff to take a low percentage shot. As a coach I’d give the defender a strong pat on the back for a good play and I’d be fine with Skinner’s play selection if disappointed with his execution.

Shot attempt based analytics and high danger chance analytics give Jeff pluses on that play and the defender minuses.

When Colin Miller, faced on the point with a charging Scot Laughton, winds up and blindly blasts a puck into his shin pads he gets a plus on the shot  attempt clock.

When Rasmus Dahlin, in the same situation, spins and throws a no-look pass to Okposo on the right boards he does not.

Discuss.

Edited February 28, 2021 by dudacek

[Stoner]

I mean, in reality, it's /thread after that post.

[dudacek]

4 minutes ago, mjd1001 said:

there are faults in +/- for sure, but it is not totally useless in my mind. Sure, a good number of the pluses and minuses may be noise, but just as many of them (if not more) might not be.  The way I look at it...it should not be used as a 'tell all' stat, but it can show you some trends...especially if you look at it in context to other things with it.

Agree with this.

I think where I might be going with this thread is that analytics should be used the same way and they aren’t. 

(Ristolainen is the worst player in the NHL! Charts!)

[mjd1001]

1 minute ago, dudacek said:

Agree with this.

I think where I might be going with this thread is that analytics should be used the same way and they aren’t. 

(Ristolainen is the worst player in the NHL! Charts!)

As with many things in this world....taking one extreme point of view might be dangerous or wrong with statistics.  The charts show this so it must be true!  What I se with my own eyes shows this so it must be true!   Not so much. The real truth is likely a blend of many things.....a little of the eye test...a little of an outside point of view....a little of advanced stats...and a little of basic stats.

[Curt]

All analytics are not created equal.  +/- only tells you a fraction of the story, Shot share stats are about the same.  Scoring chance and high danger scoring chance numbers are a little better.

I think when you start looking at expected goals , it gets more useful, but it’s also tricky because there is no standard calculation for them, so different sources calculate them slightly differently.

[Thorner]

3 minutes ago, mjd1001 said:

As with many things in this world....taking one extreme point of view might be dangerous or wrong with statistics.  The charts show this so it must be true!  What I se with my own eyes shows this so it must be true!   Not so much. The real truth is likely a blend of many things.....a little of the eye test...a little of an outside point of view....a little of advanced stats...and a little of basic stats.

Yep. 

All stats only tell a portion of the story. They are not all crested equal, and that’s where a lot of the subjectivity comes in, but by and large it’s never, certainly not to me, about proving specific analytics should be the defining measure of a player, and rather that they serve as additional information carrying just as much value as the “classic” statistics. 

Just like you can’t say Risto sucks cause his advanced metrics look poor, you can’t say he’s great because he notched 40 points. 

All stats work best combined and in context as part of a larger puzzle, than they do in isolation. 

[dudacek]

13 minutes ago, Curt said:

All analytics are not created equal.  +/- only tells you a fraction of the story, Shot share stats are about the same.  Scoring chance and high danger scoring chance numbers are a little better.

I think when you start looking at expected goals , it gets more useful, but it’s also tricky because there is no standard calculation for them, so different sources calculate them slightly differently.

Expected goals is a great one for this discussion.

Great tool for looking at a player and confirming that he is “doing the right things” even if it’s not showing up on the scoresheet.

Frequently mistakenly used on the internet as a synonym for “he’s playing well.” First and foremost, playing well doesn't just mean attempting the right things, it means actually executing them.

[Thorner]

11 minutes ago, dudacek said:

Expected goals is a great one for this discussion.

Great tool for looking at a player and confirming that he is “doing the right things” even if it’s not showing up on the scoresheet.

Frequently mistakenly used on the internet as a synonym for “he’s playing well.” First and foremost, playing well doesn't just mean attempting the right things, it means actually executing them.

And the “right” things themselves are often subjective, and in the mind of the coach - often system dependant.

[GASabresIUFAN]

People here got fooled by analytics earlier this season when they said we had a good team despite the results on the ice.  I argued then that the results were a better indicator of what this team was and is.  Now the analytics have caught up to what I was seeing on the ice.  Historical stats like shooting %, O zone time, and other straight forward numbers were and are more telling.

The problem with hockey analytics is that many of the inputs are subjective like the hit counter.  The scorer in the pressbox decides if that was a hit or a high danger chance.  The Skinner play is just a symptom of that problem.  I'm not completely discounting analytics, but many of these new numbers aren't my favorite.  

Even if they were helpful, they are useless if the team won't use them.  (See Skinner benching vs putting him with better linemates).

Edited February 28, 2021 by GASabresIUFAN

[dudacek]

Do we see radical shifts in analytics when players get a new coach and system?

Ristolainen’s have improved under Ralph. Skinner’s counting stats have dropped off the map, but his analytics have not. Risto’s usage has been similar if slightly reduced, Jeff’s has changed significantly.

And not the first time I’ve asked this question, but I’ll give it another try: When a player travels, do his analytics generally travel with him?

Analytics Jimmy Vesey was a vastly different player in Buffalo than in New York.

[Weave]

1 hour ago, dudacek said:

Thought we had a thread for this but I couldn’t find it.

 

That thread became a club.  The conversation move, and likely died, there.

I could care less about stats, really.  Detracts from actually enjoying the game as a fan in my experience.  The simple enjoy of watching a rush unfold, a two on one get played by the D man, the subtle wasting of time that is running out along the boards.  Those are much more important to me than corsi, shots from danger zones, or whatever those groups of X's and letters are supposed to represent.  They certainly don't represent the flow and ebb on the ice.

That's all I got.

[SwampD]

1 hour ago, dudacek said:

Agree with this.

I think where I might be going with this thread is that analytics should be used the same way and they aren’t. 

(Ristolainen is the worst player in the NHL! Charts!)

Absolutely.

47 minutes ago, dudacek said:

Expected goals is a great one for this discussion.

Great tool for looking at a player and confirming that he is “doing the right things” even if it’s not showing up on the scoresheet.

Frequently mistakenly used on the internet as a synonym for “he’s playing well.” First and foremost, playing well doesn't just mean attempting the right things, it means actually executing them.

I love the expected goal stat, but even that has to be taken in context. It’s a league wide average-of the shots taken from here, X% go in the net. But if it’s Crosby or McDavid taking the shot from here against Hutton, it’s a much higher expected goal than if it’s Eakin shooting on Price.

4 minutes ago, Weave said:

That thread became a club.  The conversation move, and likely died, there.

I could care less about stats, really.  Detracts from actually enjoying the game as a fan in my experience.  The simple enjoy of watching a rush unfold, a two on one get played by the D man, the subtle wasting of time that is running out along the boards.  Those are much more important to me than corsi, shots from danger zones, or whatever those groups of X's and letters are supposed to represent.  They certainly don't represent the flow and ebb on the ice.

That's all I got.

Yep. They have their place, but I wouldn’t miss them if they went away.

[Curt]

33 minutes ago, GASabresIUFAN said:

People here got fooled by analytics earlier this season when they said we had a good team despite the results on the ice.  I argued then that the results were a better indicator of what this team was and is.  Now the analytics have caught up to what I was seeing on the ice.  Historical stats like shooting %, O zone time, and other straight forward numbers were and are more telling.

The problem with hockey analytics is that many of the inputs are subjective like the hit counter.  The scorer in the pressbox decides if that was a hit or a high danger chance.  The Skinner play is just a symptom of that problem.  I'm not completely discounting analytics, but many of these new numbers aren't my favorite.  

Even if they were helpful, they are useless if the team won't use them.  (See Skinner benching vs putting him with better linemates).

Are shot information based statistics subjective?

They have objective criteria for them (location, shot type, speed, where the puck came from just before the shot, etc) and the puck has the microchip in it now, so they should be accurate. 

[dudacek]

Here are two more I see frequently and don’t understand when they are juxtaposed.

”He’s still creating a ton of chances so he’s playing well”

“He’s not really playing well, he’s just got a high shooting percentage.”

The player who got 5 goals on 80 chances is not playing better offensively than the guy who got 10 goals on 40 chances over the same 20 game stretch. I get the arguments about sustainability etcetera but it is results that count.

Edited February 28, 2021 by dudacek

[Curt]

50 minutes ago, SwampD said:

I love the expected goal stat, but even that has to be taken in context. It’s a league wide average-of the shots taken from here, X% go in the net. But if it’s Crosby or McDavid taking the shot from here against Hutton, it’s a much higher expected goal than if it’s Eakin shooting on Price.

Someone can correct me if I’m wrong, but I believe that you are incorrect here.  At least some sources use a lot more than shot location to calculate expected goals.

They are now starting to use shot location, speed, type (wrist, slap, snap, on-timer, backhand), where the puck traveled from just prior to the shot, as well as now taking shooting and goalie talent into account.  It’s getting a lot more sophisticated.

[SwampD]

12 minutes ago, Curt said:

Someone can correct me if I’m wrong, but I believe that you are incorrect here.  At least some sources use a lot more than shot location to calculate expected goals.

They are now starting to use shot location, speed, type (wrist, slap, snap, on-timer, backhand), where the puck traveled from just prior to the shot, as well as now taking shooting and goalie talent into account.  It’s getting a lot more sophisticated.

Someone is going to have to prove it to me that that is what is taking place. Where is that data coming from. Is there some hockey data analysis firm with rows of people watching replays, scrutinizing every shot of every game In the NHL, as well as the myriad other data they collect? There’s no way you can get that much information from an intern counting shots real-time during the game.

Sorry if that sounds snarky. I’m legitimately asking.

[Curt]

19 minutes ago, SwampD said:

Someone is going to have to prove it to me that that is what is taking place. Where is that data coming from. Is there some hockey data analysis firm with rows of people watching replays, scrutinizing every shot of every game In the NHL, as well as the myriad other data they collect? There’s no way you can get that much information from an intern counting shots real-time during the game.

Sorry if that sounds snarky. I’m legitimately asking.

Well there is a microchip in the puck now.  That will be logging all the data, then they would use a software program to filter the data.  The shot type is probably logged manually.  The shooter/goalie talent would be based off of readily available statistics.

[SwampD]

2 minutes ago, Curt said:

Well there is a microchip in the puck now.  That will be logging all the data, then they would use a software program to filter the data.  The shot type is probably logged manually.  The shooter/goalie talent would be based off of readily available statistics.

As to this chip, where are the sensors reading it? Are there sensors in every arena? Are they in the cameras? Every camera?I’m not arguing that it is all possible, but it is way more data than I think people make it out to be.

My question is, if it’s actually being done, who is doing it and more importantly, who is paying for it?

Since some teams analytics departments consist of only two people, my guess would be EA Sports?

[Curt]

23 minutes ago, SwampD said:

As to this chip, where are the sensors reading it? Are there sensors in every arena? Are they in the cameras? Every camera?I’m not arguing that it is all possible, but it is way more data than I think people make it out to be.

My question is, if it’s actually being done, who is doing it and more importantly, who is paying for it?

Since some teams analytics departments consist of only two people, my guess would be EA Sports?

It’s something that the NHL is doing standard across all arenas.  It’s not the individual teams doing it.

https://www.nhl.com/news/nhl-plans-to-deploy-puck-and-player-tracking-technology-in-2019-2020/c-304218820

It was all supposed to be up and running this season, but I’m actually seeing they pulled the pucks early this season because they were having issues.  Probably next season it will be in full use.

So what was saying probably isn’t true.  Next season it will be true though.  I’m just ahead of the curve, lol.

Edited February 28, 2021 by Curt

[Brawndo]

1 hour ago, SwampD said:

Someone is going to have to prove it to me that that is what is taking place. Where is that data coming from. Is there some hockey data analysis firm with rows of people watching replays, scrutinizing every shot of every game In the NHL, as well as the myriad other data they collect? There’s no way you can get that much information from an intern counting shots real-time during the game.

Sorry if that sounds snarky. I’m legitimately asking.

There are hockey tracking software available that most of the public uses. There is more sophisticated stuff that the Teams Use. It watches the games in real-time and makes the calculations. One site that does this is Money Puck, the color coded circles represent each shot, the larger the circle the higher the expected goal. Here is a link to the Sabres Game Today, if you click on one of the circles, it gives you all the info that goes into calculating expected goals. 

http://moneypuck.com/g.htm?id=2020020340

 

 

[pi2000]

==============================================================
BUF even strength goal differential through 19 games = -23
BUF even strength minutes played through 19 games = 994.23
TRpm(team relative +/-) is the diff btwn actual+/- and exp+/-
exp+/- is a function of %min played at ES and team ES goal diff
%min is % of ES minutes played
==============================================================
Team  Player                   TRpm    GP    +/-  exp+/-  %min
BUF Jake McCabe                7.01    13     2   -5.01   21.8
BUF Rasmus Ristolainen         6.23    11     1   -5.23   22.7
BUF Riley Sheahan              5.32    18     1   -4.32   18.8
BUF Curtis Lazar               3.93    16     0   -3.93   17.1
BUF Jacob Bryson               3.09     3     2   -1.09    4.8
BUF Rasmus Asplund             2.88     4     2   -0.88    3.8
BUF Dylan Cozens               2.76    14    -1   -3.76   16.3
BUF Matt Irwin                 1.67     9    -2   -3.67   15.9
BUF Jeff Skinner               1.42    15    -3   -4.42   19.2
BUF William Borgen             1.41     4     0   -1.41    6.1
BUF Casey Mittelstadt          0.70     6    -1   -1.70    7.4
BUF Jack Eichel                0.63    16    -6   -6.63   28.8
BUF Tobias Rieder              0.18    16    -4   -4.18   18.2
BUF Kyle Okposo               -0.02    13    -4   -3.98   17.3
BUF Brandon Davidson          -0.09     2    -1   -0.91    3.9
BUF Henri Jokiharju           -0.45    14    -6   -5.55   24.1
BUF Brandon Montour           -0.55    16    -8   -7.45   32.4
BUF Tage Thompson             -0.55     9    -3   -2.45   10.7
BUF Cody Eakin                -0.74    18    -6   -5.26   22.9
BUF Victor Olofsson           -1.60    18    -8   -6.40   27.8
BUF Eric Staal                -2.86    18    -9   -6.14   26.7
BUF Sam Reinhart              -3.45    16   -10   -6.55   28.5
BUF Colin Miller              -4.07    18   -11   -6.93   30.1
BUF Taylor Hall               -5.00    18   -12   -7.00   30.5
BUF Rasmus Dahlin             -8.33    18   -16   -7.67   33.3
==============================================================

 

[GASabresIUFAN]

Losing back to back games 3-0 = bad team.  No analytics needed.

[PerreaultForever]

I think ALL stats have flaws when they are looked at on their own. They are no different than +/- in that regard. The flaws originally mentioned at the top are real, especially the tough minutes one, BUT if a player is a career minus player and his minus rating is below his team mates as a whole you can still draw some conclusions from it. Same goes for other stats. Zone time means absolutely nothing is all you do is cycle around the perimeter. It has to be broken down in more detail and scoring chances etc. have to be added to it. The style the team plays factors into the stats. Philly for example, aside from against us has very low shot totals but scores a lot of goals. It's how they play. On and on you go. 

Only stats that matters to me is the W's.