The Advanced Stats Thread

[van22]

Here's a zone start question....

They only count starts off of face offs correct?

Is there a place we can find the actual raw number of zone starts... and not just a percentage?

Feel with about 60 total zone starts to be handed out per game for all ES / PP / SH situations between 3 different pairings, that zone start % is one of the most abused or misinterrupted stats around relative to it's actual significance.. ( also neutral zone starts I believe is also excluded, so there go 20-33% of the 'handed out' zone starts)

i.e. I have a feeling for some players that one single zone start could effect their percentage by 5-10%...

Part B .. is there a place to see zone start % as a percentage to the total team' s starts?

This site has number of absolute zone starts - it's under the report "on ice corsi stats" - you can sort by players/situations/teams/etc:
http://stats.hockeyanalysis.com/rat...aters&minutes=50&disp=1&sort=PCT&sortdir=DESC

War-on-ice also has them by selecting "faceoffs" under columns to display.

 

[Alan Jackson]

I think analytics can be useful, but I think where some run into problems is using them as a predictive measure for individual players, and ignoring the effect that randomness has on individual possession numbers.

I mean, if you're the Toronto Maple Leafs, it should be easy to improve: just sign and trade for players with good possession numbers, right?

Except it isn't that easy, because a player with good possession numbers playing on the Rangers 3rd line might not be able to duplicate those numbers playing on the Toronto Maple Leafs 2nd line.

I think there are players who drive possession, but for a lot of (most) players, your possession numbers are almost entirely derived from the situation you're in rather than a reflection of your play.

 

[racerjoe]

Yay! Thanks for this excited to continue my learning.

 

[DL44]

This site has number of absolute zone starts - it's under the report "on ice corsi stats" - you can sort by players/situations/teams/etc:
http://stats.hockeyanalysis.com/rat...aters&minutes=50&disp=1&sort=PCT&sortdir=DESC

War-on-ice also has them by selecting "faceoffs" under columns to display.

Thanks!

So 5-on-5... on average... we are looking at dmen getting between 4-6 o-zone starts per game.

This is the reason this stat always bugged me when people referenced it strongly in arguments one way or the other... but its neat to see what samples we're talking about laid out like that..


One thing i learned tho...

I didn't realize that neutral zone faceoffs were the most common zone start 5-on-5.
The offside effect..
Initial logic would indicate that most plays die with a save or a puck over the end boards... didn't think of the offsides... which would be (are) pretty big factors.

 

[Verviticus]

man its kind of depressing to see that the team is starting in the dzone more than the ozone now

 

[Drop the Sopel]

I think there are players who drive possession, but for a lot of (most) players, you're possession numbers are almost entirely derived from the situation you're in rather than a reflection of your play.

Yup. Each skater represents around 10% of what happens on the ice, in terms of possession results. The majority of a players results aren't derived from the player themselves, and have more to do with quality of linemates/teammates and the level of opposition.

The stats are useful to some extent, you just have to take them with a grain of salt, based on how difficult it is to separate individuals from the team.

 

[Verviticus]

well that was the thinking behind the use of WOWY (with or without you) numbers. to isolate a player's effect

for instance: joe thornton obviously makes everyone with him better. derek dorsett drags them into the deep to drown, like an anchor

 

[DL44]

man its kind of depressing to see that the team is starting in the dzone more than the ozone now

Some PP effect...

395 mins on the PP..
450 mins on the PK..
doesn't fully explain the discrepancy tho..


5-on-5 its essentially even.. (difference of 20 over 4000 starts)

 

[Drop the Sopel]

well that was the thinking behind the use of WOWY (with or without you) numbers. to isolate a player's effect

for instance: joe thornton obviously makes everyone with him better. derek dorsett drags them into the deep to drown, like an anchor

So a superstar is a better possession player than an enforcer?

Yeah, I could see that.

 

[Verviticus]

i think its pretty clear that was illustrative

 

[Mr. Canucklehead]

Good resources for advanced stats
http://stats.hockeyanalysis.com/
- for general stats

http://www.behindthenet.ca/nhl_stat...+3+5+4+6+7+8+13+14+29+30+32+33+34+45+46+63+67
- for QoC stuff

http://chlstats.pythonanywhere.com/
-CHL advanced stats

http://hockeystats.ca/
-game by game stats


What is Corsi/Fenwick TL;DR

(CF)Corsi for: Shot attempts for (Shots missed + Shots Blocked + Shots on net)
(CA)Corsi against: Shot attempts against (Shots missed + Shots blocked + Shots on net)
(CF%)Corsi for percentage: CF/(CF+CA)*100%
(FF)Fenwick for: Shots on net + Shots missed for
(FA)Fewnwick agaisnt: Shots on net + shots missed against
(FF%)Fenwick for percentage: FF/(FF+FA)*100%

So someone feel free to troubleshoot my analysis here. I'm trying to get a handle on advanced stats vs. my own eye test. For the sake of this, I'm going to pick on Luca Sbisa; he looks categorically awful to my eye, but I want to see how he stacks up "advanced stats" wise vs. his teammates.

For starts; using the HockeyAnalysis site and using 5 on 5 play as an example, for Fenwick he ranks 5th in terms of Corsi (ahead of only Chris Tanev when it comes to the regular top-six defender rotation; he's also ahead of the smattering of other defenders we used, them being Biega, Clendening, Stanton and Corrado).

Now, when I factor in the Behind the Net Statistics and look at his Corsi relative to his Quality of Competition, he is again 5th out of the 6 Canucks regular defenders (ahead of Yannick Weber this time, while Tanev leaps up to 2nd on the team, and Corrado slots ahead of him in this chart as well, technically ranking him 6th overall).

So; am I correct in assuming that the higher your QoC, the higher caliber of players you are typically on the ice against? Meaning that Tanev, for example, being placed higher on the chart relative to his QoC means he faces the tougher assignments, and still comes out ahead in terms of advanced stats? (not surprised by that, just looking to make sure I understand that)

Would I also further infer that Sbisa plays against tougher opponents than Weber, and has better advanced stats to show for it? Or am I completely off base?

 

[Alan Jackson]

well that was the thinking behind the use of WOWY (with or without you) numbers. to isolate a player's effect

for instance: joe thornton obviously makes everyone with him better. derek dorsett drags them into the deep to drown, like an anchor

I agree, but if you put Dorset with the Sedins, and put Thornton out with 4th liners, and I'll bet the WOWY numbers would change drastically.

The numbers are helpful, but they don't tell the whole story, and they can't always anticipate the random nature of the sport.

 

[Tiranis]

I agree, but if you put Dorset with the Sedins, and put Thornton out with 4th liners, and I'll bet the WOWY numbers would change drastically.

I don't understand what you're saying... this happens all the time and we can still isolate the good vs the bad players.

Literally every single one of your complaints from the original post has already been addressed by the advanced stats community.

 

[Verviticus]

So; am I correct in assuming that the higher your QoC, the higher caliber of players you are typically on the ice against? Meaning that Tanev, for example, being placed higher on the chart relative to his QoC means he faces the tougher assignments, and still comes out ahead in terms of advanced stats? (not surprised by that, just looking to make sure I understand that)

Would I also further infer that Sbisa plays against tougher opponents than Weber, and has better advanced stats to show for it? Or am I completely off base?

generally yeah. if the numbers are pretty close it doesn't matter too much. i tend to look at things like QoC and QoT in tiers - players tend to be grouped in clumps and you can see "ok this group plays the hardest competition, this group the second hardest"

edit: he definitely doesn't have better stats to show for it - sbisa is the worst regular canucks player by far (excluding stanton)

though

for Fenwick he ranks 5th in terms of Corsi

doesn't make much sense

so here's two things:

corsi for canucks D
fenwick for canucks d

if you want to compare players, its best to pick one of three things

A: fenwick/corsi for (a measure of offence)
B: f/c against (a measure of defence)
C: f/c% (a measure overall)

I prefer C and I basically always use corsi unless there's something i specifically want from fenwick. i like larger samples in general (fenwick excludes blocked shots), and honestly, its pretty rare that a player can consistently post large discrepancies between the stats to the point where its significant

 

[Alan Jackson]

I don't understand what you're saying... this happens all the time and we can still isolate the good vs the bad players.

Literally every single one of your complaints from the original post has already been addressed by the advanced stats community.

My apologies if I'm not expressing myself clearly. And they're not necessarily "complaints", per se. I think the advanced stats community is doing good work, and I agree that the data can be very useful. But they can't tell us everything about a player. They can't.

In an alternate universe, where Derek Dorset plays 82 games with the Sedins, Derek Dorset will have positive possession numbers. Those numbers would not be replicated if he's in his current role. Players tend to play the majority of their shifts with set line-mates, and a shift here or there with different personnel may not provide enough of a sample to generate accurate WOWY data.

I think the possession numbers are better at judging how a player is playing currently, in his current role, on his current team. I don't think they're necessarily a good predictor of how that player will play on a new team, in a new role, with different players.

And to be frank, if possession is the end-all and be-all, we should just measure actual possession (time how long each player possesses the puck, with a stopwatch). That, compared with personal shot attempts, might be a better gauge than what we have now.

 

[Verviticus]

sure, but they can't always be together. dorsett will occasionally be with someone else and be an anchor and the sedins will occasionally be with someone else and fly away. thats the point of WOWY, to isolate those specific things and make them easier to see

go back to my links: look at the CF%s under "with this player" and "without this player". the links weren't just measuring how good a player is, but how good a player is with every other player.

actual possession time would own and every single stats geek wants it. its very hard to track. ****, they can't even track shots properly

and a shift here or there with different personnel may not provide enough of a sample to generate accurate WOWY data.

this happens very rarely and there are other ways to isolate that information. you can do a "with player x" and the sum of all of the "without player x" if you need a larger sample. its a valid criticism but it basically never comes up, so its not worth worrying about too much

 

[Drop the Sopel]

edit: he definitely doesn't have better stats to show for it - sbisa is the worst regular canucks player by far (excluding stanton)

Looks like Sbisa's numbers are better than Stanton and Corrado's, and not much worse than Bieksa's. Though penalties drawn/taken is an important stat that often gets glossed over - Sbisa is much better than Bieksa/Stanton in this regard and it affects goal differential.

Stanton is a guy they tried to shelter with few defensive zone starts and easy competition and yet he was torn to shreds it appears, while taking a ton of minor penalties. Stanton was a major disappointment this season - an advanced stats nightmare.

 

[carolinacanuck]

serious question..

has anyone advanced statted (is that a word) gretzky? is nhl.com gonna go back and analyze the data from the past?

 

[Verviticus]

serious question..

has anyone advanced statted (is that a word) gretzky? is nhl.com gonna go back and analyze the data from the past?

they would love to but they only started tracking actual shot attempts in.. 2006? people have tried for sure but information from that era is very spotty

also, at some level of talent the system is going to breakdown a bit. just like how in jr and the AHL high end shooting talent seems more repeatable, there's a good chance that the oilers teams from the 80s were just so much better that what we have wouldn't be useful

Looks like Sbisa's numbers are better than Stanton and Corrado's, and not much worse than Bieksa's. Though penalties drawn/taken is an important stat that often gets glossed over - Sbisa is much better than Bieksa/Stanton in this regard and it affects goal differential.

Stanton is a guy they tried to shelter with few defensive zone starts and easy competition and yet he was torn to shreds it appears, while taking a ton of minor penalties. Stanton was a major disappointment this season - an advanced stats nightmare.

okay? bieksa sucks and stanton had a bad year. we know this

 

[Tiranis]

In an alternate universe, where Derek Dorset plays 82 games with the Sedins, Derek Dorset will have positive possession numbers. Those numbers would not be replicated if he's in his current role. Players tend to play the majority of their shifts with set line-mates, and a shift here or there with different personnel may not provide enough of a sample to generate accurate WOWY data.

Yeah but this literally never happens. Even Daniel Sedin spent 77 ES minutes (6 games worth!) this season without his brother and that's by far the most inseparable duo in the league.

I don't think possession stats are perfect or even great but the criticisms levelled against them are usually incredibly weak.

 

[Drop the Sopel]

okay? bieksa sucks and stanton had a bad year. we know this

You said Sbisa was worse than Bieksa by far. When I look at the advanced stats, I don't really see that. Are you looking at some numbers that say Bieksa was much better than Sbisa? Want to make sure I'm not missing anything in there...

Did Stanton have a bad year, or was this season a good reflection of what to expect going forward? I think that's the question.

 

[Alan Jackson]

Yeah but this literally never happens. Even Daniel Sedin spent 77 ES minutes (6 games worth!) this season without his brother and that's by far the most inseparable duo in the league.

I get your point, but for the sake of argument, 77 minutes over 82 games works out to maybe a shift or two per game. I don't think 1 or 2 shifts with random linemates in the middle of a line change is enough data to support anything.

 

[Verviticus]

You said Sbisa was worse than Bieksa by far. When I look at the advanced stats, I don't really see that. Are you looking at some numbers that say Bieksa was much better than Sbisa? Want to make sure I'm not missing anything in there...

Did Stanton have a bad year, or was this season a good reflection of what to expect going forward? I think that's the question.

i... ok, if you dont see that, that's alright

i hope stanton had a bad year. he had a good year last year.

 

[DL44]

Yeah but this literally never happens. Even Daniel Sedin spent 77 ES minutes (6 games worth!) this season without his brother and that's by far the most inseparable duo in the league.

And that's magical spin on numbers that people become weary of when people start in with new stats....

77 mins without his brother...

6 games worth of mins without his brother! (seems significant)

or is actually...

56 secs/game... (much less significant)

Or is that 56 sec broken up into 9 or 10 smaller segments due to lines changes.... (pretty much insignificant)

Same thing. but it isn't.

Edit.. I see Jackson beat me to it...

I get your point, but for the sake of argument, 77 minutes over 82 games works out to maybe a shift or two per game. I don't think 1 or 2 shifts with random linemates in the middle of a line change is enough data to support anything.

 

[Verviticus]

...they're a very specific and extremely rare case, and the information of "which brother is better" is completely irrelevant. why do we care? we're not going to make personnel decisions with it.