Understanding NHL Analytics: A Beginners Guide

[LastWordArmy]

“Advanced stats” has become a kind of buzz word around the hockey world. It’s emerging into the NHL as it did years ago with the MLB. Hockey teams have entire departments for analytics, and it has become a war (pun intended) of sorts between two sides: The Data Nerds vs. Old School Hockey. What I intend to do is break down some of the common stats used in analytics to help people get introduced into the world of NHL analytics.
Understanding NHL Analytics

When I first wanted to get into NHL analytics, the hardest part was finding a resource that helped me understand not just what the stats meant, but what a “good” version of the stat was. It’s relatively common knowledge that scoring 30 goals in a season is a good season, but what’s good in terms of Corsi or expected goals?

“Advanced stats” is often a misleading term. This is because, at their roots, a lot of the common stats being talked about are extremely simple in nature. There are some complex ways of looking at the context at times, or some large calculations involved, but the base stats aren’t that “advanced”.
One final precursor before getting into the content: the majority of these numbers are to measure 5v5 in the NHL. When talking about powerplays and penalty kills, it becomes a different beast.
Corsi

What is Corsi? This is probably the most common stat to be heard when referring to NHL advanced stats. To sum up in the most basic terms: it’s measuring the number of chances. Corsi measures shots on goal, shots wide, and blocked shots. It tends to paint a bigger picture of the entire game than just the traditional count of “shots on goal”.

There is both Corsi for (CF) and Corsi against (CA). Because of this, Corsi can be shown as a differential (C± or C+/-), or as a percentage (CF%). With that, the most common use of expressing Corsi is through CF%. It is the simplest to understand and puts it into a context that looking at raw Corsi for or against doesn’t. Individual Corsi (iCF) is also a calculatable stat. It can tell you how many shot attempts a single player has taken. Corsi in this context, however, is rarely used.

Continued in our 4 part series here....

Understanding NHL Analytics: A Beginners Guide Part 1

and

Understanding NHL Analytics: A Beginners Guide Part 2

 

[Jumptheshark]

Did not click onto all your hard work

but there are some analytics that teams have that so far more comprehensive than those who worship analytics have not been able to attain certain key stats

A buddy of mine who works for an NHL club in their analytics department(FYI not all call that department analytics, and they do not tell everyone who is in that department).

When we talk about analytics and part of the game that the so-called experts fall down on, he points out that most experts work off of the idea that all players have 20/20 vision. Most of the top goalies have better vision than others. His other qualm which he has not seen on any sites is how many hits a player receives, not tosses but is on the other end of. His other observation is the every famous Hockey IQ, people talk about it but when ask to define and explain it, that is where the fun begins. He has been involved in hockey for over 30 years and got into analytics when it was still called advance stats. He is waiting for the analytics crowns to start pushing numbers on players when they play say 3 games in 5 nights or say 20 games over a 30-day period and then see how the look at the stats when players only play 4 games in 15 days, The one thing he always points out to me is that some players perform better when they are playing more games in a shorter time and other players thrive when playing fewer games, Some players are okay in very physical games while other, when they get hit too many times in a game are useless.

One thing that has amuses him is how analytics starts talking about players equipment without asking why some players use a full shield or half shield and when it comes to d man they do not know which d man are using certain equipment while others use different types. PMD will use different pads then those who we would call stay at home d-men

 

[LeHab]

Did not click onto all your hard work

but there are some analytics that teams have that so far more comprehensive than those who worship analytics have not been able to attain certain key stats

A buddy of mine who works for an NHL club in their analytics department(FYI not all call that department analytics, and they do not tell everyone who is in that department).

When we talk about analytics and part of the game that the so-called experts fall down on, he points out that most experts work off of the idea that all players have 20/20 vision. Most of the top goalies have better vision than others. His other qualm which he has not seen on any sites is how many hits a player receives, not tosses but is on the other end of. His other observation is the every famous Hockey IQ, people talk about it but when ask to define and explain it, that is where the fun begins. He has been involved in hockey for over 30 years and got into analytics when it was still called advance stats. He is waiting for the analytics crowns to start pushing numbers on players when they play say 3 games in 5 nights or say 20 games over a 30-day period and then see how the look at the stats when players only play 4 games in 15 days, The one thing he always points out to me is that some players perform better when they are playing more games in a shorter time and other players thrive when playing fewer games, Some players are okay in very physical games while other, when they get hit too many times in a game are useless.

One thing that has amuses him is how analytics starts talking about players equipment without asking why some players use a full shield or half shield and when it comes to d man they do not know which d man are using certain equipment while others use different types. PMD will use different pads then those who we would call stay at home d-men

Visual acuity is critical for batters in baseball. Average MLBer is reported to be 20/12. Some test 20/8 which is as good as a human eye can get. Better acuity = extra fraction of a second to pick that baseball spin. Would be interesting to see more eyesight related data for nhl players. There is also more to vision than acuity which probably plays a bigger role in a fast paced game like hockey.

 

[Sinistril]

A better title would be understanding hfboards analytics. Corsi, et cetera, are not really 'advanced' statistics in any sense and have serious concerns associated with them.

 

[LastWordArmy]

Here is part 3. Which looks at goaltenders.


Goals Saved Above Average

Goals saved above average (GSAA) isn’t really an advanced stat. While it may not be advanced, it is still good to know when discussing goalies. To calculate GSAA, you compare how many goals a goaltender allowed against how many a league-average goalie would allow in the same amount of shots. Formally, it is the league’s average save percentage with the number of shots a goalie has faced. This number is how many goals a league-average goalie would allow. This number is then subtracted from the goalie that you are evaluating against. This will result in either a positive or negative number.

Continued here

Understanding NHL Analytics: A Beginners Guide Part 3

 

[LastWordArmy]

Did not click onto all your hard work

but there are some analytics that teams have that so far more comprehensive than those who worship analytics have not been able to attain certain key stats

A buddy of mine who works for an NHL club in their analytics department(FYI not all call that department analytics, and they do not tell everyone who is in that department).

When we talk about analytics and part of the game that the so-called experts fall down on, he points out that most experts work off of the idea that all players have 20/20 vision. Most of the top goalies have better vision than others. His other qualm which he has not seen on any sites is how many hits a player receives, not tosses but is on the other end of. His other observation is the every famous Hockey IQ, people talk about it but when ask to define and explain it, that is where the fun begins. He has been involved in hockey for over 30 years and got into analytics when it was still called advance stats. He is waiting for the analytics crowns to start pushing numbers on players when they play say 3 games in 5 nights or say 20 games over a 30-day period and then see how the look at the stats when players only play 4 games in 15 days, The one thing he always points out to me is that some players perform better when they are playing more games in a shorter time and other players thrive when playing fewer games, Some players are okay in very physical games while other, when they get hit too many times in a game are useless.

One thing that has amuses him is how analytics starts talking about players equipment without asking why some players use a full shield or half shield and when it comes to d man they do not know which d man are using certain equipment while others use different types. PMD will use different pads then those who we would call stay at home d-men

I think a lot of what your are talking about... vision, hockey IQ, etc... You are looking at attributes. Analytics doesn't really look at a player's individual skills, it looks at results. It doesn't care if someone has a face shield or not.

Thats going to be in the dept of visual scouting.

And of course scouting and analytics aren't two things that are competing against each other, in a good organization they are two things that work together to help management make better decisions. So many people want to make this a binary competition between the two, but that's not what it should be at all.

 

[TomasHertlsRooster]

Did not click onto all your hard work

but there are some analytics that teams have that so far more comprehensive than those who worship analytics have not been able to attain certain key stats

A buddy of mine who works for an NHL club in their analytics department(FYI not all call that department analytics, and they do not tell everyone who is in that department).

When we talk about analytics and part of the game that the so-called experts fall down on, he points out that most experts work off of the idea that all players have 20/20 vision. Most of the top goalies have better vision than others. His other qualm which he has not seen on any sites is how many hits a player receives, not tosses but is on the other end of. His other observation is the every famous Hockey IQ, people talk about it but when ask to define and explain it, that is where the fun begins. He has been involved in hockey for over 30 years and got into analytics when it was still called advance stats. He is waiting for the analytics crowns to start pushing numbers on players when they play say 3 games in 5 nights or say 20 games over a 30-day period and then see how the look at the stats when players only play 4 games in 15 days, The one thing he always points out to me is that some players perform better when they are playing more games in a shorter time and other players thrive when playing fewer games, Some players are okay in very physical games while other, when they get hit too many times in a game are useless.

One thing that has amuses him is how analytics starts talking about players equipment without asking why some players use a full shield or half shield and when it comes to d man they do not know which d man are using certain equipment while others use different types. PMD will use different pads then those who we would call stay at home d-men

Statistically speaking, teams with more rest tend to to better in terms of shot shares, goal shares, wins/losses, etc.

 

[SeaOfBlue]

I think a lot of what your are talking about... vision, hockey IQ, etc... You are looking at attributes. Analytics doesn't really look at a player's individual skills, it looks at results. It doesn't care if someone has a face shield or not.

Thats going to be in the dept of visual scouting.

And of course scouting and analytics aren't two things that are competing against each other, in a good organization they are two things that work together to help management make better decisions. So many people want to make this a binary competition between the two, but that's not what it should be at all.

This is often the impression I have been given.

I do think teams are looking to quantify attributes like Hockey IQ, but more to confirm what scouts are seeing more than anything else. What analytics are going to do when it comes to scouting is tell scouts where to look, and at which players, to make scouting more streamlined. They will still take a look at everyone possible, but having some analytics which may tell them to look at certain guys who may not have otherwise garnered much attention, or to maybe provide some further insight into higher profile prospects that may have been otherwise missed with the traditional eye test. There are hundreds of players who are available to be drafted each year from all over the world and even the largest scouting departments do not have more than 15 or so scouts. It is not going to be possible to see all of them properly.

 

[LastWordArmy]

From Part 4

What is WAR good for? A lot, actually. Today we will be breaking down wins above replacement as a means of evaluating hockey players. In any examples used in this article will come from Evolving Hockey. There are other good websites with very good WAR models such as MoneyPuck.
What is WAR?

WAR is a stat that you may be familiar with from baseball. It is a way to assess players and assign a numerical value to how much a single player contributes to their team. This value can be expressed as either a positive or a negative depending on a players impact. This stat can be used to evaluate players across positions easily by seeing how many “wins” they provide their team. This is a fantastic stat when looking at players who should win things like the Hart Trophy.

Evolving Hockey also can express this as goals above replacement (GAR). This provides similar information and just breaks it down to how many goals a player provides his team above a replacement-level player. It roughly translates from 5.6 GAR to 1 WAR based on their model.

Continued here
Understanding NHL Analytics: A Beginners Guide Part 4

 

[adsfan]

Visual acuity is critical for batters in baseball. Average MLBer is reported to be 20/12. Some test 20/8 which is as good as a human eye can get. Better acuity = extra fraction of a second to pick that baseball spin. Would be interesting to see more eyesight related data for nhl players. There is also more to vision than acuity which probably plays a bigger role in a fast paced game like hockey.

Ted Williams was supposed to be 20/15. He is still the last .400 hitter in MLB.

 

[neelynugs]

i don't follow this stuff a lot, but what site has 5 on 5 stats that go back the furthest?

 

[Filthy Dangles]

i don't follow this stuff a lot, but what site has 5 on 5 stats that go back the furthest?

natural stat trick goes back to 07-08 which is the first season the inputs for the stats were officially recorded.

 

[CutOnDime97]

A better title would be understanding hfboards analytics. Corsi, et cetera, are not really 'advanced' statistics in any sense and have serious concerns associated with them.

Any links that give an intro to more advanced ones?

 

[NeverBeNormal]

Is there any way yet to account for goaltending, besides a catch all 'average' number to compare them against which doesn't put any definition towards how difficult the save is or where the shot comes from?

 

[TomasHertlsRooster]

Is there any way yet to account for goaltending, besides a catch all 'average' number to compare them against which doesn't put any definition towards how difficult the save is or where the shot comes from?

Expected goals do account for where the shot comes from as well as a few other factors that estimate how difficult a save is.

 

[CutOnDime97]

Expected goals do account for where the shot comes from as well as a few other factors that estimate how difficult a save is.

Would you say expected goals is the best metric to evaluate a player? Good place to start for learning analytics?

 

[morehockeystats]

natural stat trick goes back to 07-08 which is the first season the inputs for the stats were officially recorded.

More precisely, 07/08 is the first season where the events were recorded with on-ice presence of all the players.
Between 1999-2007 only goals were recorded with on-ice presence.

 

[oilerbear]

Hockey is not a Money ball Sport:
Most analytics look at table scrap Affects in hockey.
Looking for Moneyball binary ( 2 outcomes) from hockey actions that have multiple 3+ success% based outcomes.

Baseball has the large affect outcomes of 100 yrs ago being played the same by all teams.
So table scrap moneyball (2 outcome analytics has value.)

Hockeynnalytics has failed to identify all large % affects thru Sequence of Events (SOE) mapping.
That allows you to identify the best and worse ways to play the gm.

Current hockey analytics people are trained academically.
Using theoretical approach to the gm.
But they are in communities that see small% affect mistakes as valueable.
Their is no real current value to their work cause of the huge% errors in their data.

High Resolution Complex outcome analysis:
Being high function Autistic I was real world trained in mapping all theoretical outcomes and identifying real world method (sports human machine play) By age 8.

My father (Offered lead in Orbit based assets for 1st yrs of European Space agency) was a lead on design of Avro Arrow, Appollo lunar lander & Service Module, Coversion of Hawk Missle to first self guided SAM, Design of first Earth Resources Technology Satelite (LandSat), Design of SeaSat basis for all orbit based Sub locating, Area 25: magnetic guided Nuclear powered Rockets. He travelled the world setting up other countries Orbit asset tracking divisions.
I was exposed to a group of friends from the projects in “late 50’s, 69’s and 70’s who designed things like Maglev.

O% mistake/ high% success approach is the only way:
I was educated that 0% mistake analysis was critical.
Mistakes have huge outcomes. Look up Aerospace occurances.
January 28, 1986
February 1, 2003

Their are fields in Academia that see 15% affects as a huge success.
A poor standard.

Multiple high resolution affects Methodology to achieve desired high% success goal:
I have 50+ theories that lead to an identification of 26 yrs of repeat final 4 roster cores based on my High danger shot density shot area observation of 50+ hrs ago and Closed shot identification of 45+ yrscago.

The champ roster theory I presented on here around April 17-20 pre VGK exp draft ( stated would be a GA cup final team) is now the only Pro sports roster proof based on high outcome affects.
VGK selected 100% of players in Expansion that fit my Roster Core Theory. They made the final which I stated.

1.00 to 2.00 GAA to win 4 gm in a Conf or Cup series:
Most Analytic people fail to understand that final 4 series winning teams only give up 4g (1.00 GAA) to 8g (2.00 GAA) in their Series 4 wins.

Get strong GA players they are cap cheap:
When you build a roster your value acquisition of Strong even & PK ga reduction players.
This yr median.
Evg/60 2.64
STG/60 7.16
7.16/2.64 = 2.71
PPG are 2.71 times easier than evg.
PK reduction is 2.71 times harder than evg reduction.
PK special teams +ve goal dif affect is 5.42 times harder than PP goal dif affect

Seek strong top 125 fwd evg and evp depth:
1.65 assist are given for each goal.
Evg are a superior direct outcome affect than EVA1
EV1 are superior to EVA2.
That is a reflection of average pass completion % compounding.
Non 1.000 values being timed by each other the failure rate potential increases for every extra pass needed.
I used this Methodology to look for strong fwd Prispects in the draft:

Homeplate theory:
Is based on shot density succes from an x,y locations.

Open / closed shots:
All non scoreable Corsi must be identified
( Blocks + misses + hit goalie = Wall) so that the open shots are identified. Pucks that hit into a Goalie have 0% chance of going in.
Pucks that deflect off a goalie Are scoreable.

hit posts with zero off data value:
Corsi that hit posts and do nit go in are not a shot.
They are a miss.
They have zero off performance value but are important in the measure of a open shot reduction analysis.
(Blocks + misses + closed shots/ CA

Their are 7 open holes:
Their are 7 holes in net elevation based Y,Z location on every plane established by a goalies set position (Angle & distance from net elevation).

7 open hole shot maps:
So their are 7 identified y,z range ( holes) maps on a single holes open shots x,y Success map. Fir a group of equal result based set positions?

A Goalie is a wall:
With arms, legs, head to Move to stop open shots.
The entire Hockey analytics world works of a belief that all shots from the same x,y location have the Same chance of going in.
If you take 2 nets and cover one completely with plywood and the other completely open.
You cannot tell a kid you have the same chance of put a shot in the net with both nets.

their are slot of standard open shot maps:
Shots from a single location can
- be open or closed
- hit 7 diffrent open shot hole locations (must be a hole not A Goalie)
- have difrent hole sizes based on set positions of goalie.

Neanderthal DNA:
Their is Bio Evolutionary targeting built in the human brain.
But instead if trying to hit the body (Mammoth/goalie) in the visualized plane. We are suppose to hit the open space which is counter intuitive targeting that must be mentally and visually practised.

Low resolution thinking leads to mistakes:
The more detailed a look at things the better we understand the best way to succeed at something.
Complex mutivariable outcome Rocket Science comes from many military non grade 12 grades who have complex Triarch design/renovate; build/ maintain; operate/run individuals.
Low resolution thought leads extensive changes. I have found theoretical design will go thru 40-45 changes before respectability in construction.

Shot mass = shot volume x expected sSH syuccess density (each set position hole size map)

Elite Fwds are identified by:
their Homeplate penetration by then are high comp% passes and open shot targeting Succes.
No fwds has the Same xSH% ( ie xG)

Elite Dmen are identified by:
-the Open xSave% ((Open shots - xG)/open shot) baseline they establish to their side.
-A dman covering a partners abandoned side does not get charged to the covering dman.
- a dman coming On Does not get charged for Coverage to a side not being established from Last Player back dman (Rover) Going off. It is still the abandoning dmans affect.

Elite Goalies are identified by:
Their +ve save% performance over the xSave% baseline established to each side.

Most of the large affects of the gm have been identified by me.
And used to get 26 yrs repeat roster and system play success.

No Real results:
Current Analytics stumbles along only using 1/2 of goal density identification for shot mass equation.
No closed shot exclusion = no real Science.

PDO has no value:
Each individual player has their own averages.
3 fwd xSH%
2 Dmen with their own xSave% to their side.
1 goalie with 2 Save% from each dman side xSave% to be measured against.
No combination of 7 Position averages are the same.

Their is no Simple War in complex sports:
In 07/08 I created a 3D graph of Desjardins Team, Comp, FO ZS
With team and comp set into
8 groups (upper/lower 1st/2nd/3rd/4th)
and 8 FO ZS groups; 8 x 8 x 8 = 512 possible player situation averages to get a season based xAvg for all 512 groups of play.

Coaches kill Corsi analytics:
Coaches choices in ZS with or without pocession leads to a 40% range in error of CF and CA data.

Eliminating Coach ZS affect on player analytics:
But quite quickly saw the need ZS to be broken up into 4 seperate zone start 3 D graphs with 4 Team, 4 Comp & 6 ZS% = 96 avg
FOZS with pocession (FO win)
FOZS without pocession ( Loss)
On the fly (bench ZS) with pocession
On the fly ( bench ZS) without pocession.
You get to identify a +ve or -ve performance for any of the 3D player situations 96 (team, comp, FOZS%) x 4 graph = 384 avg That a player is put in by the Coach.

3f - 2D -1G def sys vs. 3F - 1Rover - 1D -1G:
Off Dmen abandon 2D - 1G def of high danger shot area to chase offence. They occupy a higher% of fwd zone space. They take off zone pocession from fwds. They are a hybrid of Fwd/ Dmen based on zone space.

NZ transition Defence affects:
Corsi results are dictated By Zone entry success which has a large dependence on weather a NZ def is run or not.
- A Coach can dictate the fwds approach to NZ def which affects zone entry/ CA.
- Off dman who abandons their def responsibility affects Fwds and dpartners approach to NZ transition/ Zone entry/ CA.

Eliminate NZ def affect in performance:
4 further diferentiations must be done to get true player performance.
Fwd NZ Def run in 3F - 1R - 1D - 1G structure
fwd NZ Def run in 3F - 2D - 1G structure
Fwd NZ def Not run in 3-1-1-1 Structure
Fwd NZ def Not Run in 3 - 2 - 1 Structure

Corsi against is not a Defenceman measure:
Current Hockey analytics believes these 2 Dmen should have the same xCA
- 1 dman sees a high% of on the fly bench change without pocession while coming on after No fwd NZ def has been run in a (3 - 1 - 1 -1) attack of opp OZ and high% of def FOZS.
- 1 dman sees a lot of on the fly Bench change With pocession in a (3F - 2D - 1G) attack of opp OZ and a high% of OFF FZS.
I just laugh at how bad low resolution analytics taught in Academics is.

I literarly stated on Lowetide That Sports IQ would be a gm changer cause of the extreme complex humAn machine analytics I identified could be used.
Shook my head at low res stuff presented!

Just a short user guide of my Analytucs that nailed the first Campionship Sports proof.

PS: did this during my Rainman and Beautifulmind sessions (wife calked it) from 10 pm. To 2-4 am.
During the last 13 yrs of chasing modeling and prototype data fir engineering on the ATCO Lead on Coal Power Plant Conversion to Nat. gas.
Dealing with cancer last 2 1/2 yrs.
Though thevPlant I worked at in Alberta was a critical Node in NA VAR Network.

 

[Doctor No]

O% mistake/ high% success approach is the only way:
I was educated that 0% mistake analysis was critical.
Mistakes have huge outcomes. Look up Aerospace occurances.
January 28, 1986
February 1, 2003

Their are fields in Academia that see 15% affects as a huge success.
A poor standard.

What's your definition of "mistake" in this context?

No offense, but this isn't aerospace (and the consequences of a mistake are not even remotely similar). If you wait for a literally-perfect model to get started, you'll be waiting awhile while everyone else is benefitting.