I'm not sure how to share the image to here, but I think this link should work:
Eichel compared to Pastrnak
All-Purpose Advanced Stats, Player Usage, etc. Thread
- Thread starter Zip15
- Start date
I'm not sure how to share the image to here, but I think this link should work:
Eichel compared to Pastrnak
Aladyyn
they praying for the death of a rockstar
new toy based on Corey Sznajder's data: Tableau Public
small sample size from this season, but hey, it's fun
small sample size from this season, but hey, it's fun
Aladyyn
they praying for the death of a rockstar
Hey guys, can you help a brother out? I'm having a debate about plus-minus versus pretty much any other stat with respect to predictability. I KNOW I've read studies comparing it to like, corsi, for example.
Do you guys have the links? I'm swamped with work tonight and can't spend all night searching and figured maybe someone has something bookmarked.
ie, show me the work that shows why the analytics community has distaste for that stat. I understand it has crap predictability, but I want to convince someone else of that.
Do you guys have the links? I'm swamped with work tonight and can't spend all night searching and figured maybe someone has something bookmarked.
ie, show me the work that shows why the analytics community has distaste for that stat. I understand it has crap predictability, but I want to convince someone else of that.
Aladyyn
they praying for the death of a rockstar
struckbyaparkedcar
Guilty of Being Right
Late to the party on this but...
For starters, +/- sucks at being a descriptive goal differential stat. It doesn't normalize for ice time, which generally leads to punishing high-minute players on bad teams. For example, Jack Eichel and Tomas Hertl have the same 45% goal differential 5v5. Eichel is -25, Hertl -1.
Then, it also randomly includes non-ES events, for the benefit of PK specialists and detriment of PP guys, so it's even noisier and less descriptive than raw goal differential.
Finally, GF% as a whole isn't as predictive as shot-based metrics because goaltending factors into things too much. By removing save percentage from your equation, you get a clearer picture of what the 18 skaters are doing.
PS: Here's a good article testing the predictive properties of the different stats. TL;DR - SCF rate is the best goals-for predictor, CA/60 is better at predicting goals-against.
Are Scoring Chances Better Than Corsi?
struckbyaparkedcar
Guilty of Being Right
The newer version of this lets you sort by year.new toy based on Corey Sznajder's data: Tableau Public
small sample size from this season, but hey, it's fun
Tableau Public
Most of this is probably from Oct-Nov, but still. Thanks, Phil.
Husko
Registered User
Shocking
joshjull
Registered User
You actually think this?
joshjull
Registered User
This could be a really great tool if it had season long numbers. Instead its comparing 22 random games for ROR last year to 12 random games this year. In Jack's case its 15 random games to 12.
You say it could be mostly from Oct/Nov. The 12 random games could be entirely from that time or not at all from that time.
struckbyaparkedcar
Guilty of Being Right
Oh totally. These are definitely on the interesting data point side of things.
I saw these in January, so I think the data is very skewed to early in the year.
Appreciated.Late to the party on this but...
For starters, +/- sucks at being a descriptive goal differential stat. It doesn't normalize for ice time, which generally leads to punishing high-minute players on bad teams. For example, Jack Eichel and Tomas Hertl have the same 45% goal differential 5v5. Eichel is -25, Hertl -1.
Then, it also randomly includes non-ES events, for the benefit of PK specialists and detriment of PP guys, so it's even noisier and less descriptive than raw goal differential.
Finally, GF% as a whole isn't as predictive as shot-based metrics because goaltending factors into things too much. By removing save percentage from your equation, you get a clearer picture of what the 18 skaters are doing.
PS: Here's a good article testing the predictive properties of the different stats. TL;DR - SCF rate is the best goals-for predictor, CA/60 is better at predicting goals-against.
Are Scoring Chances Better Than Corsi?
Aladyyn
they praying for the death of a rockstar
Yes, I think he's one of the best players in the league.
Do I come off as that big of an idiot?
@struckbyaparkedcar , do you have results like those analyses for plus minus versus corsi/shot metrics? I know they get mentioned as being less predictive early on in that article, but they don't link to that analysis.
joshjull
Registered User
I give him full marks for even trying to create these things with the enormous amount of time he has to commit to breakdown games.
Last edited:
joshjull
Registered User
The article, and general consensus, is that shot based metrics (Corsi, we'll say) are better predictors for future goals for/against/success in general than plus minus.
I really want to see the studies, presented in the exact same or similar format to the ones comparing corsi to scoring chances in that article, that back that conclusion up. I know they have happened and I have seen them before years ago, but I can't find them.
Let's just say I'm conversing with people who won't believe me when I say that it is true, they need to see it themselves.
joshjull
Registered User
I've never heard of studies about plus/minus being a predicator of anything. I've only known it to be used as a bad measure of how good or bad a player was defensively.
There are studies out there, otherwise the stats community would not state with confidence that it's a worse predictor stat than shot based metrics, because they would have no proof for such a claim then.
Also, I have very little understanding of statistical analysis at this point, so this is a question from ignorance - the r^2 values for literally any metric appears to be quite low, so why should I be convinced that any of them are good predictors, with those low numbers? What is the proof that there's meaningful correlation to reality with any of them?
joshjull
Registered User
I'll take your word there are studies but I'm at a loss on how you could predict anything based on a +/- model.
I don't put much stock in predictive modeling for hockey. Its pretty much educated guesswork. Too many variables that influence things that cannot be quantified properly or at all.
I guess I'm not the best guy to answer your questions
jc17
Registered User
- Jun 14, 2013
- 11,035
- 7,765
Bumping this so I dont pollute the other threads with off-topic stuff.
Admittedly in some desperation to have hope for the season I'm trying to convince myself the roster might be better than expected. I've already mentioned that there's chance Risto could be effective as seen by his stats when you separate him from the other Sabres "anchors", in a way you can't do with basic WOWY.
Looking at Vesey, who's acquisition has left a lot of us unimpressed especially with a 46 CF%, 47 xGF%, and negative WAR, maybe there's a chance his surface level numbers dont tell the whole story.
NYR was really bad last year, especially at even strength. Here's what I got separating Vesey from some of NY's anchors.
Similar to Risto, Vesey's stats are respectable when away from the lousy players on his team, but the lousy players stay lousy away from him. Now, the argument can be made that its predictable that players will look good away from bad players. I think its important to notice that Vesey without the anchors, out-performs the rest of the team in every category.
Not saying this is proof he will be good, I haven't been super impressed in the pre-season, but I'm also not ready to write him off.
Admittedly in some desperation to have hope for the season I'm trying to convince myself the roster might be better than expected. I've already mentioned that there's chance Risto could be effective as seen by his stats when you separate him from the other Sabres "anchors", in a way you can't do with basic WOWY.
Looking at Vesey, who's acquisition has left a lot of us unimpressed especially with a 46 CF%, 47 xGF%, and negative WAR, maybe there's a chance his surface level numbers dont tell the whole story.
NYR was really bad last year, especially at even strength. Here's what I got separating Vesey from some of NY's anchors.
Similar to Risto, Vesey's stats are respectable when away from the lousy players on his team, but the lousy players stay lousy away from him. Now, the argument can be made that its predictable that players will look good away from bad players. I think its important to notice that Vesey without the anchors, out-performs the rest of the team in every category.
Not saying this is proof he will be good, I haven't been super impressed in the pre-season, but I'm also not ready to write him off.
itwasaforwardpass
I'll be the hyena
- Mar 4, 2017
- 5,330
- 5,142
Fair point. Howden was Vesey's most common line mate and he was a drag. It was mentioned before somewhere but thank you for laying out the numbers. Similarly, Tage being tied to Sobotka is one thing that gives me some hope about Tage.
