OT: Advanced Stats Info Request

[YotesFan47]

I know many of you use advanced stats and post opinions on players based on them, but it's all gibberish to me. I've never really taken the time to learn what they are, or mean, or why certain ones are important. I guess I'm looking for some starting points because I lack so much knowledge on the subject.

I figured those of you in the know could post some of your preferred stats and what they mean to the on ice product. I'd imagine many of us could benefit from your knowledge, and it could lead to some interesting debate on why you do or don't like certain methods of measurement.

I know we get some posters from other fan bases too, you can share also you dirty lurkers!

Also, Happy New Year!

Click to read more...

 

[Freddy Sjostrom]

If you are interested in learning more, definitely following JFresh. I look at WAR (wins above replacement) you can Google a better definition, but it grades all aspects of a player. If a player is a slick offensively scoring a goal a game, but gives up 10 shots on defense, his WAR will be lower. It also has a built in component that compares players to a baseline or average.

For goalies, goals saved above expected is good. Goalies are so hard to judge, to my untrained eye anyway. You can make the argument goalie stats are team stats. Veggie may appear to be a horrible goalie to other teams, but as you see his gsax is top 5.

 

[lanky]

I really like the idea of WAR. The big shortcoming, as far as I understand it, is that any statistician can create their own WAR model. They don't have to disclose what goes into the calculation or how it works. With that much opaqueness it would be impossible to understand outliers and decide if the measure has much to offer.

 

[XX]

Remember kids - rates, never counts. 5v5 data over everything as well. Expected (aka xGF/GA/etc...) is generally more valuable than raw outcomes (simple GF/60, for example).

I really like the idea of WAR. The big shortcoming, as far as I understand it, is that any statistician can create their own WAR model. They don't have to disclose what goes into the calculation or how it works. With that much opaqueness it would be impossible to understand outliers and decide if the measure has much to offer.

This would only be a problem if you were trying to compare disparate data sets (WAR models), which you probably shouldn't ever do. It should still be internally consistent.

xGAR/WAR is fun. You can have arguments in your head about, say, the 15-16 Norris:

OEL - 27.6
Karlsson - 20.8
Doughty - 13.9 (winner)

Evolving-Hockey has free GAR/WAR data on player pages: Evolving-Hockey.com | Player Pages

 

[PainForShane]

Remember kids - rates, never counts. 5v5 data over everything as well. Expected (aka xGF/GA/etc...) is generally more valuable than raw outcomes (simple GF/60, for example).

No. Use the more appropriate comparison tool for whichever argument you're trying to evaluate. Sometimes that is rates, sometimes that is counts.

To be clear though, deciding to look at rates or counts will give you basically the same information at least 95% of the time. Probably more.

 

[YotesFan47]

If you are interested in learning more, definitely following JFresh. I look at WAR (wins above replacement) you can Google a better definition, but it grades all aspects of a player. If a player is a slick offensively scoring a goal a game, but gives up 10 shots on defense, his WAR will be lower. It also has a built in component that compares players to a baseline or average.

For goalies, goals saved above expected is good. Goalies are so hard to judge, to my untrained eye anyway. You can make the argument goalie stats are team stats. Veggie may appear to be a horrible goalie to other teams, but as you see his gsax is top 5.

Without looking it up, I'm going to assume the above expected is some combination of team and player metrics assembled in a formula to make a prediction?

 

[YotesFan47]

Remember kids - rates, never counts. 5v5 data over everything as well. Expected (aka xGF/GA/etc...) is generally more valuable than raw outcomes (simple GF/60, for example).



This would only be a problem if you were trying to compare disparate data sets (WAR models), which you probably shouldn't ever do. It should still be internally consistent.

xGAR/WAR is fun. You can have arguments in your head about, say, the 15-16 Norris:

OEL - 27.6
Karlsson - 20.8
Doughty - 13.9 (winner)

Evolving-Hockey has free GAR/WAR data on player pages: Evolving-Hockey.com | Player Pages

Why is expected more valuable? I've also heard WAR can be as misleading as +\-. Now maybe the person that said it is a numb-skull, but it was said to me at one point.

Do people pay sites to get those fancy, colorful graphs?

 

[PainForShane]

I know many of you use advanced stats and post opinions on players based on them, but it's all gibberish to me. I've never really taken the time to learn what they are, or mean, or why certain ones are important. I guess I'm looking for some starting points because I lack so much knowledge on the subject.

I figured those of you in the know could post some of your preferred stats and what they mean to the on ice product. I'd imagine many of us could benefit from your knowledge, and it could lead to some interesting debate on why you do or don't like certain methods of measurement.

I know we get some posters from other fan bases too, you can share also you dirty lurkers!

Also, Happy New Year!

This is cool, thanks for asking this Q. Hopefully this thread helps a bunch of ppl become better fans of this amazing game. And hopefully this gets the lurkers out of hiding! If I'd seen this way back when I might have come out of hiding a little earlier.

I think the most important thing is to actually understand what the particular stats measure and then use the proper advanced stat / comparison depending on whatever you're trying to figure out. And whatever the advanced stats say, use the eye test to verify whenever possible (especially before you bring up a potentially controversial opinion).

That's not really an actionable point, but you definitely want to keep the big picture in mind at all times.

***

In terms of the actual stats, there are things that try to measure everything. As of now it's pretty accepted that we do a better job of measuring ability to score compared to ability to prevent goals (esp by non-goalies), but there are decent metrics around shots / shot attempts allowed (and from where) so that as a team you can get a top level sense of how things look at either end of the ice. There are also stats that try to measure luck (PDO, arguably on-ice shooting percentage), but the point is there's really no one stat that encapsulates everything, though some stats like GAR/WAR/SPAR and xGAR try.

And yes, just saw your post and really any stat (WAR, +/- etc) can be misleading if you don't know what you're doing / use it inappropriately. It doesn't mean the stats are inherently bad, it just means you have to know when it's appropriate to use them and for what purpose.

***

I think what I would do if I were just starting out is pick one stat at a time and just understand what that one stat is. Like, really understand it -- try to find some short blog post or youtube video that explains why it's good and what it's limitations are. And then go to a site like Natural Stat Trick and try to play around with players / teams you think would be good with respect to that one stat, so you get a sense of the variance of it from game to game / season to season as appropriate. And after that one stat, just move onto the next one. In total I think there are probably, I don't know maybe 20 or so relevant stats you want to get a good sense of? You don't even need to get that far, if you understand even one advanced stat pretty well, you'll be a much more aware fan than otherwise.

Personally I think I would start with Corsi and Fenwick, not because they're better than other stats but because a lot of other stats are essentially built on top of them (this will make sense once you start getting deeper into this stuff). This is obviously my opinion, you can honestly start wherever.

To repeat though -- the key is to understand the individual stat before you try to use it. Think of each stat like a tool -- if you don't understand what they're for, you might unknowingly pull out a chainsaw when all you need is a lockpick.

 

[YotesFan47]

This is cool, thanks for asking this Q. Hopefully this thread helps a bunch of ppl become better fans of this amazing game. And hopefully this gets the lurkers out of hiding! If I'd seen this way back when I might have come out of hiding a little earlier.

I think the most important thing is to actually understand what the particular stats measure and then use the proper advanced stat / comparison depending on whatever you're trying to figure out. And whatever the advanced stats say, use the eye test to verify whenever possible (especially before you bring up a potentially controversial opinion).

That's not really an actionable point, but you definitely want to keep the big picture in mind at all times.

***

In terms of the actual stats, there are things that try to measure everything. As of now it's pretty accepted that we do a better job of measuring ability to score compared to ability to prevent goals (esp by non-goalies), but there are decent metrics around shots / shot attempts allowed (and from where) so that as a team you can get a top level sense of how things look at either end of the ice. There are also stats that try to measure luck (PDO, arguably on-ice shooting percentage), but the point is there's really no one stat that encapsulates everything, though some stats like GAR/WAR/SPAR and xGAR try.

And yes, just saw your post and really any stat (WAR, +/- etc) can be misleading if you don't know what you're doing / use it inappropriately. It doesn't mean the stats are inherently bad, it just means you have to know when it's appropriate to use them and for what purpose.

***

I think what I would do if I were just starting out is pick one stat at a time and just understand what that one stat is. Like, really understand it -- try to find some short blog post or youtube video that explains why it's good and what it's limitations are. And then go to a site like Natural Stat Trick and try to play around with players / teams you think would be good with respect to that one stat, so you get a sense of the variance of it from game to game / season to season as appropriate. And after that one stat, just move onto the next one. In total I think there are probably, I don't know maybe 20 or so relevant stats you want to get a good sense of? You don't even need to get that far, if you understand even one advanced stat pretty well, you'll be a much more aware fan than otherwise.

Personally I think I would start with Corsi and Fenwick, not because they're better than other stats but because a lot of other stats are essentially built on top of them (this will make sense once you start getting deeper into this stuff). This is obviously my opinion, you can honestly start wherever.

To repeat though -- the key is to understand the individual stat before you try to use it. Think of each stat like a tool -- if you don't understand what they're for, you might unknowingly pull out a chainsaw when all you need is a lockpick.

You did a better job articulating what I'm trying to learn haha. As a long time fan of the game, and someone who enjoys statistics, I'd love to develop a deeper understanding of available metrics. I remember wanting to start this process a few years ago but found myself doing what I do, delving too deep into too many different metrics which caused confusion, and ultimately led to me not investigating further. I want to correct my process this time, plus I find I learn best in group settings where figures can be associated to relevant information.

A major spark currently is Chychrun. You see posters who have never/rarely watched the guy play take dumps on him, but from my eye tests and basic metrics, find him to be a very valuable player. I want to have a deeper knowledge on the advanced metrics available to provide deeper insights to folks who may not only lack viewings, but understanding of the advanced metrics necessary to help make an astute judgement.

So based on this picture, Chychrun is our best player, that actually plays on our team, for shot attempts when on the ice. If I'm understanding this correctly, his FF% means he would also be one of our worst for unblocked shot attempts when on the ice, meaning he lets a lot of shots through? Or do I have that backwards?

 

[XX]

Why is expected more valuable? I've also heard WAR can be as misleading as +\-. Now maybe the person that said it is a numb-skull, but it was said to me at one point.

Do people pay sites to get those fancy, colorful graphs?

Expected gives you a larger, more representative sample because it measures all attempts and not just outcomes. A player that has a high raw xGF/60 is generating some kind of offense. They might be Tage Thompson blowing it past goalies or it might be Christian Fischer style hit the goalie square in the chest offense, but it's still there to a degree and you can check it against their actual scoring.

Easy example: A goalie is getting rave reviews and Vezina mentions on twitter, but their team is a gong show and their stats are pretty pedestrian. If we're trying to objectively measure this goalie's performance somehow, what would it look like? What are we trying to find? We're roughly trying to find how much a given goalie is exceeding expectations (is this goalie playing out of his damn mind?) if their raw normal stats like SV% and GAA are not telling us much.

If a goalie lets in 2 shots on 20 total, that's generally bad. If they only let in two on 60 shots against, that's a heroic performance. Same thing goes for saving point blank quality chances vs flip ins from the blue line. It's possible for the amount of shots a goalie on a bad team sees to completely dwarf that of a goalie on a defensively strong team. We want to measure, if possible, how much better a given goalie is doing than a league average then, using the quality of the shots themselves as the measure rather than the raw totals (total shots) and outcomes (goals against).

Enter Goals Saved Above Expected. Because we have improved tracking data statisticians can now measure roughly the league average chance of a shot from a certain place on the ice going in, and a goalie routinely beating this average by making a save is exceeding expectations.

Of goalies who have played 20 games, the 5v5 above expected per 60 leaders are:
Hellebuyck (Vezina)
Saros
Ullmark
Hart
Sorokin
Vasilevskiy (Vezina)
Vejmelka

You use 5v5 data when comparing players because shorthanded data is not created equally or evenly distributed, making it mostly meaningless for cross comparisons. It introduces too much noise.

On Moneypuck we can see Vejmelka has the highest raw xGAA in the league in all situations, so he is seeing a lot of quality chances against every game. So while his 3.16 GAA and .907 SV% are pedestrian at best, we can easily deduce he is actually one of the better starters in the league and has been playing that way with this advanced metric.

 

[SniperHF]

JFresh

HFBoards poster, he's still semi active around here.

 

[PainForShane]

You did a better job articulating what I'm trying to learn haha. As a long time fan of the game, and someone who enjoys statistics, I'd love to develop a deeper understanding of available metrics. I remember wanting to start this process a few years ago but found myself doing what I do, delving too deep into too many different metrics which caused confusion, and ultimately led to me not investigating further. I want to correct my process this time, plus I find I learn best in group settings where figures can be associated to relevant information.

A major spark currently is Chychrun. You see posters who have never/rarely watched the guy play take dumps on him, but from my eye tests and basic metrics, find him to be a very valuable player. I want to have a deeper knowledge on the advanced metrics available to provide deeper insights to folks who may not only lack viewings, but understanding of the advanced metrics necessary to help make an astute judgement.

View attachment 629809

So based on this picture, Chychrun is our best player, that actually plays on our team, for shot attempts when on the ice. If I'm understanding this correctly, his FF% means he would also be one of our worst for unblocked shot attempts when on the ice, meaning he lets a lot of shots through? Or do I have that backwards?

This is awesome. Cool discussion to have, popped open a beer as well. This is a long post (tried to cut the length but wasn't really able to), but lots of content aimed at the advanced stats noob, apologies in advance but I think it will be helpful (especially to the interested lurker, we know you are out there!!!

Also, I selectively bolded some lines so easier to skip around in case you already know some of the info. Maybe not enough but at least it's something.

Anyway a couple of things (before we get to the question):

1) First of all these are really, REALLY bad team Corsi / Fenwick numbers. An absolutely average team will be 50% on both CF and FF percentage, if we look at our team it looks like our CF% is somewhere between 40 and 45% as a team? To @XX this is one of those times it helps to see counts because otherwise you might disproportionally weight the three games that Timmins' and Jenik had when you're doing an eyeball estimate.

Or I guess you can look up what the team Corsi percentage is directly (turns out it is 42.8% at even strength which is... really bad). To put that in perspective, Bergeron will be a first ballot HOFer largely because of his 2-way play, Corsi / Fenwick are decent approximations of that. Bergeron's even strength CF% is 58.4... which is pretty close to the opposite of our team Corsi (100 - 42.8% = 57.2%). So according to this widely-used and unbiased top level statistical approximation, our team is so bad at ES that the average of all our opponents' possession play this year is essentially the level of Patrice Bergeron.

I'm laughing as I type this because it aligns with the eye test -- and at least now we can quantify how bad our team is. This is one reason why playing with advanced stats can be fun, if used right you can discover some things.

***

2a) Regarding the definition of Corsi and Fenwick as statistics, Corsi measures all shot attempts when a player is on the ice - which includes shots on goal, missed shots, blocked shots - these all count as Corsi events. Computing the Corsi For % is what you'd expect, namely the number of events your team has divided by total events (when said player is on the ice). So if Chychrun is on the ice for 15 shot attempts taken and 10 shot attempts against, his Corsi For % for that game would be (15 / 25) 60%. Fenwick measures pretty similar things as Corsi, the only difference is that blocked shots do not count as Fenwick events.

Re: Fenwick, if Chychrun is on the ice for 15 shot attempts taken (but 5 shots are blocked) and 10 shots attempts allowed (and all 10 of those shots get through), his Fenwick For % for that game would be ((15 - 5 ) / (20)) or 50%. His CF% percentage would still be 60% because as stated above blocked shots count as Corsi events but not as Fenwick events. Aside from that one little difference, Corsi and Fenwick are designed to measure the same thing.

Higher numbers are better for both Corsi and Fenwick for percentages (CF%, FF% respectively), and the percentages vary quite a bit from game to game.

2b) Some ppl argue about which stat (Corsi or Fenwick) is a "better" indicator of what is happening in a game. To me, it doesn't really matter because the VAST majority of the time we're just looking for a high level summation anyway, and also in the vast majority of cases (easily 95%+), Fenwick and Corsi for percentages will be basically equivalent, esp over a whole season.

***

But, both Corsi or Fenwick have problems, esp if we have tunnel vision and don't consider overall context.

1) First, they're really dependent on quality of linemate and quality of competition. If I'm some bum and I play with GREAT linemates and defenders at ES, I'd expect to have relatively high Corsi / Fenwick numbers. Why? Because my linemates would be good at getting shots off (and through in the case of Corsi), and if my D don't allow any shot attempts against, I'll barely have any Corsi / Fenwick events meaning my CF% and FF% will be high. On the other hand, if my linemates are crap and I'm playing in the D-zone all game, of course I'm more likely to be in the ice for shot attempts so my Corsi / Fenwick would be lower. Same idea wrt whether I'm mostly playing against other teams' top six or bottom six.

2) Second, Corsi / Fenwick don't account for shot quality or shot location. As an example if I gain the blue line and immediately shoot (missing the net), and the puck is picked up by the opposition and they miss (or score on) a 2-1... well, according to Fenwick and Corsi that's one Corsi event for and one against, so CF% and FF% are both 50% which implies an even game. But that is obviously not what's happening, Corsi / Fenwick just measure the number of Corsi or Fenwick events, nothing to do with the quality of the events themselves. There are other stats which do try to control for this sort of thing but Corsi and Fenwick do not.

3) Lastly there are other problems with Corsi / Fenwick that aren't immediately as obvious immediately but will skew the numbers. For instance if I'm a John Tortorella team and I'm 100% ok with giving up those point shots... well, when other teams take those shots, everyone's Corsi / Fenwick numbers will be worse because every shot counts as an event against. But if I have a coach that's always yelling, "SHOOT THE PUCK" then I'll probably have better Corsi / Fenwick numbers because more events for.

Specifically to this convo, what happens if I'm Chychrun this year and I'm deployed more often in the O-zone than the other Arizona D? Well, then I'd expect my Corsi / Fenwick numbers to be slightly higher than my teammates' because the coach is giving me more offensive opportunities. But what if Chych is spending lots of time against other players' top competition (which he is)...? Well, then I'd expect his Corsi / Fenwick to be slightly lower. The question is whether all these effects balance each other out... and honestly I have no idea. I'd want to play around with the numbers more before definitively saying one thing or the other.

Anyway there are quite a few factors that can influence the #s either way, that's why it's important to keep the big picture and use common sense when you look at stats.

***

So now your question (finally!). How to analyze the table?

The first thing I noticed is how truly horrible our possession numbers are (see above), the second thing is for my own sake try to see whether anyone's Corsi / Fenwick are materially different. Like I expect, Fenwick and Corsi numbers are basically the same (assuming a decent sample size), no more than a percentage point or two difference for anybody (again @XX this is another example where I'm glad I can see counts because it's obvious that Timmins, Jenik and Mayo didn't play much so I'm not even going to look into the difference between their Corsi / Fenwick, it's due to only a few events so almost certainly random chance).

And now the Q -- is Chych our best player because he has one of the best CF% on our team. I would say... maaaybe. It's not clear because of those edge effects (#3 above). Chychrun is starting more in the O-zone than other Arizona D-men (that is what oZS% measures, we'll get to that later), but also he's averaging >3 shots a game which will juice his Corsi / Fenwick numbers a bit. That's not a bad thing Chychrun should be shooting (his shot is amazing!) but it means we should expect his Corsi / Fenwick numbers to be slightly higher. But, it looks like he plays most of his minutes against top competition (I looked this up to shut up some main board clowns), so that would lower his percentages. And he's putting up points which is good, also +11... but his Corsi / Fenwick percentages are below 50 so it's not like we're even close to dominating when he's on this ice.

And yes, a big part of that is probably our lack of offensive depth (after Keller / Schmaltz / arguably Crouse / Macelli there is a big dropoff) but it doesn't make it less true. But also his PDO of 105 is unsustainably high but not that surprising given a small sample size (PDO is a stat that is supposed to measure luck, it's not 100% accurate but 105 on our team is extreme). We haven't discussed PDO yet but maybe we'll get there.

So for the reasons above, if I used advanced stats alone, I'm not sure Chychrun's our best player. He might be, but I don't think the answer is clear just based on the stats. But, combine above average possession numbers (wrt other Yotes players) with the eye test as he continues to launch bombs and still amazing in transition... yes, Chychrun is our best player. That's why you need to use both the eye test and advanced stats.

***

My other takeaway with the stats shown above -- I'm actually surprised at how good Valimaki looks based only on the numbers. Much worse deployment than Chych in terms of zone starts, but pretty much the same Corsi / Fenwick (ie possession) numbers. That's surprising to me (in a good way). Personally I'd want to look a little more into who Valimaki matches up against and also what he looks like in games before deciding anything, but based on the table you showed, I find Valimaki's numbers to be surprisingly good, so I'll pay more attention to him in the coming games and see what my eyes can tell me and if they back up these stats.

And THAT is another reason to get familiar with these types of stats. To change our expectations and maybe direct our focus in places it wasn't before.

Either way definitely worth the time needed to get smart about this stuff. Thanks again for starting the thread!

 

[PainForShane]

You did a better job articulating what I'm trying to learn haha. As a long time fan of the game, and someone who enjoys statistics, I'd love to develop a deeper understanding of available metrics. I remember wanting to start this process a few years ago but found myself doing what I do, delving too deep into too many different metrics which caused confusion, and ultimately led to me not investigating further. I want to correct my process this time, plus I find I learn best in group settings where figures can be associated to relevant information.

A major spark currently is Chychrun. You see posters who have never/rarely watched the guy play take dumps on him, but from my eye tests and basic metrics, find him to be a very valuable player. I want to have a deeper knowledge on the advanced metrics available to provide deeper insights to folks who may not only lack viewings, but understanding of the advanced metrics necessary to help make an astute judgement.

View attachment 629809

So based on this picture, Chychrun is our best player, that actually plays on our team, for shot attempts when on the ice. If I'm understanding this correctly, his FF% means he would also be one of our worst for unblocked shot attempts when on the ice, meaning he lets a lot of shots through? Or do I have that backwards?

That was a long post. You (and everyone else) probably need a TLDR. So here is a shorter summary:

Your question was, "based on the picture above is Chychrun our best player" my answer is something like, "based on the picture he's one of our better players but I'm not sure if he's the best" because of a few reasons:

1) Chych's Corsi / Fenwick numbers are higher than our other players, but Valimaki's are basically just as good
2) We expect Chychrun's Corsi / Fenwick numbers to be higher than most, because he shoots a lot and has significantly more offensive zone starts (oZS%) compared to other Yotes defenders besides Ghost Bear (his D partner)
3) However, we also expect Chychrun's Corsi / Fenwick numbers to be lower because he plays a lot of even strength minutes against opponents' top lines

The effects of 2 and 3 will partially offset each other. have no idea which factor would have a higher impact.

4) Also, there is different advanced stat (PDO) that supposedly measures luck in scoring. This stat in my opinion doesn't really do a great job of this. Nonetheless, Chychrun's PDO is very high which means his actual scoring numbers (Goals / Assists / Points) are probably slightly higher than they should be due to goalie performance while he's on the ice. This PDO stat should not affect Corsi or Fenwick numbers

5) Valimaki's Corsi / Fenwick numbers are basically the same as Chychrun's, and he has much lower offensive deployment and not an extreme PDO. But, I don't know anything about who Valimaki usually matches up against at even strength

***

So, based solely on the advanced stats I think we can say Chychrun is better than most of our roster. But, solely based on the stats above I don't know for sure if we can say he is our best player, because Valimaki or maybe because one of our guys with lower possession numbers is a stud in some other valuable way (ie if Moser is Chris Pronger-like in the D-zone, we'd need more info to prove or disprove that).

Combined with the eye test, yes I would say Chych is our best player. But I'd need that eye test also.

Additional context / justification incl in that very long post above. Happy new year everyone!

 

[YotesFan47]

This is awesome. Cool discussion to have, popped open a beer as well. This is a long post (tried to cut the length but wasn't really able to), but lots of content aimed at the advanced stats noob, apologies in advance but I think it will be helpful (especially to the interested lurker, we know you are out there!!!

Also, I selectively bolded some lines so easier to skip around in case you already know some of the info. Maybe not enough but at least it's something.

Anyway a couple of things (before we get to the question):

1) First of all these are really, REALLY bad team Corsi / Fenwick numbers. An absolutely average team will be 50% on both CF and FF percentage, if we look at our team it looks like our CF% is somewhere between 40 and 45% as a team? To @XX this is one of those times it helps to see counts because otherwise you might disproportionally weight the three games that Timmins' and Jenik had when you're doing an eyeball estimate.

Or I guess you can look up what the team Corsi percentage is directly (turns out it is 42.8% at even strength which is... really bad). To put that in perspective, Bergeron will be a first ballot HOFer largely because of his 2-way play, Corsi / Fenwick are decent approximations of that. Bergeron's even strength CF% is 58.4... which is pretty close to the opposite of our team Corsi (100 - 42.8% = 57.2%). So according to this widely-used and unbiased top level statistical approximation, our team is so bad at ES that the average of all our opponents' possession play this year is essentially the level of Patrice Bergeron.

I'm laughing as I type this because it aligns with the eye test -- and at least now we can quantify how bad our team is. This is one reason why playing with advanced stats can be fun, if used right you can discover some things.

***

2a) Regarding the definition of Corsi and Fenwick as statistics, Corsi measures all shot attempts when a player is on the ice - which includes shots on goal, missed shots, blocked shots - these all count as Corsi events. Computing the Corsi For % is what you'd expect, namely the number of events your team has divided by total events (when said player is on the ice). So if Chychrun is on the ice for 15 shot attempts taken and 10 shot attempts against, his Corsi For % for that game would be (15 / 25) 60%. Fenwick measures pretty similar things as Corsi, the only difference is that blocked shots do not count as Fenwick events.

Re: Fenwick, if Chychrun is on the ice for 15 shot attempts taken (but 5 shots are blocked) and 10 shots attempts allowed (and all 10 of those shots get through), his Fenwick For % for that game would be ((15 - 5 ) / (20)) or 50%. His CF% percentage would still be 60% because as stated above blocked shots count as Corsi events but not as Fenwick events. Aside from that one little difference, Corsi and Fenwick are designed to measure the same thing.

Higher numbers are better for both Corsi and Fenwick for percentages (CF%, FF% respectively), and the percentages vary quite a bit from game to game.

2b) Some ppl argue about which stat (Corsi or Fenwick) is a "better" indicator of what is happening in a game. To me, it doesn't really matter because the VAST majority of the time we're just looking for a high level summation anyway, and also in the vast majority of cases (easily 95%+), Fenwick and Corsi for percentages will be basically equivalent, esp over a whole season.

***

But, both Corsi or Fenwick have problems, esp if we have tunnel vision and don't consider overall context.

1) First, they're really dependent on quality of linemate and quality of competition. If I'm some bum and I play with GREAT linemates and defenders at ES, I'd expect to have relatively high Corsi / Fenwick numbers. Why? Because my linemates would be good at getting shots off (and through in the case of Corsi), and if my D don't allow any shot attempts against, I'll barely have any Corsi / Fenwick events meaning my CF% and FF% will be high. On the other hand, if my linemates are crap and I'm playing in the D-zone all game, of course I'm more likely to be in the ice for shot attempts so my Corsi / Fenwick would be lower. Same idea wrt whether I'm mostly playing against other teams' top six or bottom six.

2) Second, Corsi / Fenwick don't account for shot quality or shot location. As an example if I gain the blue line and immediately shoot (missing the net), and the puck is picked up by the opposition and they miss (or score on) a 2-1... well, according to Fenwick and Corsi that's one Corsi event for and one against, so CF% and FF% are both 50% which implies an even game. But that is obviously not what's happening, Corsi / Fenwick just measure the number of Corsi or Fenwick events, nothing to do with the quality of the events themselves. There are other stats which do try to control for this sort of thing but Corsi and Fenwick do not.

3) Lastly there are other problems with Corsi / Fenwick that aren't immediately as obvious immediately but will skew the numbers. For instance if I'm a John Tortorella team and I'm 100% ok with giving up those point shots... well, when other teams take those shots, everyone's Corsi / Fenwick numbers will be worse because every shot counts as an event against. But if I have a coach that's always yelling, "SHOOT THE PUCK" then I'll probably have better Corsi / Fenwick numbers because more events for.

Specifically to this convo, what happens if I'm Chychrun this year and I'm deployed more often in the O-zone than the other Arizona D? Well, then I'd expect my Corsi / Fenwick numbers to be slightly higher than my teammates' because the coach is giving me more offensive opportunities. But what if Chych is spending lots of time against other players' top competition (which he is)...? Well, then I'd expect his Corsi / Fenwick to be slightly lower. The question is whether all these effects balance each other out... and honestly I have no idea. I'd want to play around with the numbers more before definitively saying one thing or the other.

Anyway there are quite a few factors that can influence the #s either way, that's why it's important to keep the big picture and use common sense when you look at stats.

***

So now your question (finally!). How to analyze the table?

The first thing I noticed is how truly horrible our possession numbers are (see above), the second thing is for my own sake try to see whether anyone's Corsi / Fenwick are materially different. Like I expect, Fenwick and Corsi numbers are basically the same (assuming a decent sample size), no more than a percentage point or two difference for anybody (again @XX this is another example where I'm glad I can see counts because it's obvious that Timmins, Jenik and Mayo didn't play much so I'm not even going to look into the difference between their Corsi / Fenwick, it's due to only a few events so almost certainly random chance).

And now the Q -- is Chych our best player because he has one of the best CF% on our team. I would say... maaaybe. It's not clear because of those edge effects (#3 above). Chychrun is starting more in the O-zone than other Arizona D-men (that is what oZS% measures, we'll get to that later), but also he's averaging >3 shots a game which will juice his Corsi / Fenwick numbers a bit. That's not a bad thing Chychrun should be shooting (his shot is amazing!) but it means we should expect his Corsi / Fenwick numbers to be slightly higher. But, it looks like he plays most of his minutes against top competition (I looked this up to shut up some main board clowns), so that would lower his percentages. And he's putting up points which is good, also +11... but his Corsi / Fenwick percentages are below 50 so it's not like we're even close to dominating when he's on this ice.

And yes, a big part of that is probably our lack of offensive depth (after Keller / Schmaltz / arguably Crouse / Macelli there is a big dropoff) but it doesn't make it less true. But also his PDO of 105 is unsustainably high but not that surprising given a small sample size (PDO is a stat that is supposed to measure luck, it's not 100% accurate but 105 on our team is extreme). We haven't discussed PDO yet but maybe we'll get there.

So for the reasons above, if I used advanced stats alone, I'm not sure Chychrun's our best player. He might be, but I don't think the answer is clear just based on the stats. But, combine above average possession numbers (wrt other Yotes players) with the eye test as he continues to launch bombs and still amazing in transition... yes, Chychrun is our best player. That's why you need to use both the eye test and advanced stats.

***

My other takeaway with the stats shown above -- I'm actually surprised at how good Valimaki looks based only on the numbers. Much worse deployment than Chych in terms of zone starts, but pretty much the same Corsi / Fenwick (ie possession) numbers. That's surprising to me (in a good way). Personally I'd want to look a little more into who Valimaki matches up against and also what he looks like in games before deciding anything, but based on the table you showed, I find Valimaki's numbers to be surprisingly good, so I'll pay more attention to him in the coming games and see what my eyes can tell me and if they back up these stats.

And THAT is another reason to get familiar with these types of stats. To change our expectations and maybe direct our focus in places it wasn't before.

Either way definitely worth the time needed to get smart about this stuff. Thanks again for starting the thread!

This is where I snipped the table btw: 2022-23 Arizona Coyotes Roster, Stats, Injuries, Scores, Results, Shootouts | Hockey-Reference.com

Not a problem at all my friend! I love learning more and more as I age, its just too bad my brain isn't the same level of sponge it was 20-30 years ago.

I apricate you including so much information. Maybe it didn't read as, but my initial understanding of Corsi and Fenwick was completely false. After reading your post and looking at the chart again, I see where I had made a mistake and now totally get how they operate and why they have value. This definitely gives me some things I want to look for from certain players in our coming games.

It's funny you found the Valimaki figures surprising, I did as well. I know he's made some slick plays, but he looks sloppy on the ice to me. It's like he runs half a second behind everyone else (mentally). I'll need to pay closer attention to him.

Yea, PDO is interesting. You said it's a luck measurement? I'm not sure what the save percentage part of the equation is for a skater, Ill need to look up what that entails.

 

[PainForShane]

This is where I snipped the table btw: 2022-23 Arizona Coyotes Roster, Stats, Injuries, Scores, Results, Shootouts | Hockey-Reference.com

Not a problem at all my friend! I love learning more and more as I age, its just too bad my brain isn't the same level of sponge it was 20-30 years ago.

I apricate you including so much information. Maybe it didn't read as, but my initial understanding of Corsi and Fenwick was completely false. After reading your post and looking at the chart again, I see where I had made a mistake and now totally get how they operate and why they have value. This definitely gives me some things I want to look for from certain players in our coming games.

It's funny you found the Valimaki figures surprising, I did as well. I know he's made some slick plays, but he looks sloppy on the ice to me. It's like he runs half a second behind everyone else (mentally). I'll need to pay closer attention to him.

Yea, PDO is interesting. You said it's a luck measurement? I'm not sure what the save percentage part of the equation is for a skater, Ill need to look up what that entails.

Ha, all good! Yeah hockey-reference is a good place to start imo. I think I might have started there too (well, also with google). Re: hockey-reference, I find the interface very clean and not overwhelming, if you mouse over the categories it'll also pop up how the stats are calculated.

Re: PDO, yeah, it's supposed to measure luck but there are enough things wrong with its calculation that I personally don't like it, different discussion though. At a top level it's measured by my team's on ice shooting percentage + my team's on ice save percentage (when I'm on the ice)... which supposedly is not correlated to the player so a high PDO apparently means the player is getting lucky wrt to goal scoring / goals allowed. But like... I would expect someone like Scott Stevens (and his D partner) to have a higher on ice save percentage (and therefore a higher PDO) because I would think they'd be better at making the average shot less dangerous, you know? Which is the opposite of luck. Anyway that's a different discussion let's tackle it later, you can probably tell I'm not a big fan of the stat but no time to write another mammoth post.

Anyway, Chychrun's PDO is extremely high so probably have to listen to it a bit even though I don't want to. This is usually how these things work, there's always so much fuzziness around the edges. Another reason to really understand each stat before blindly plugging them in

 

[YotesFan47]

Ha, all good! Yeah hockey-reference is a good place to start imo. I think I might have started there too (well, also with google). Re: hockey-reference, I find the interface very clean and not overwhelming, if you mouse over the categories it'll also pop up how the stats are calculated.

Re: PDO, yeah, it's supposed to measure luck but there are enough things wrong with its calculation that I personally don't like it, different discussion though. At a top level it's measured by my team's on ice shooting percentage + my team's on ice save percentage (when I'm on the ice)... which supposedly is not correlated to the player so a high PDO apparently means the player is getting lucky wrt to goal scoring / goals allowed. But like... I would expect someone like Scott Stevens (and his D partner) to have a higher on ice save percentage (and therefore a higher PDO) because I would think they'd be better at making the average shot less dangerous, you know? Which is the opposite of luck. Anyway that's a different discussion let's tackle it later, you can probably tell I'm not a big fan of the stat but no time to write another mammoth post.

Anyway, Chychrun's PDO is extremely high so probably have to listen to it a bit even though I don't want to. This is usually how these things work, there's always so much fuzziness around the edges. Another reason to really understand each stat before blindly plugging them in

Seems to me it would be relevant as an add on stat. For example, Chychrun has the high Corsi for our team, showing he's better at helping his team keep possession, and PDO elaborates that he's also keeping SA to lower danger areas and his team's SF are more effective when he's on the ice. Would that be a correct evaluation? Basically, he plays more of a bend, don't break defense while also being a constant scoring threat (either actively or passively) when on the ice.