Comparing Jets first 18 games this and last year

[Derfel*]

That first chart... it just... it just drops off the bottom there... like way down... right out... the bottom.

I'm afraid it's going to show up in my reply!

 

[mcpw]

That first chart... it just... it just drops off the bottom there... like way down... right out... the bottom.

I'm afraid it's going to show up in my reply!

Your post started a new page so it didn't.

 

[garret9]

I'm not as concerned. HSC only looks at particular shots, reducing sample and making it not very predictive.
It's also not overly descriptive either.
43% of the Jets goals last year were not from HSC.

I'd prefer this:

 

[mcpw]

It's actually very descriptive if used in the right context (not: "Jets have low HSCF% so Jets=bad team which will regress to 46 GF% over large sample").

SACF% 51.9 (8th)
xGF% 49.5 (16th)
SAHSCF% 46.0 (24th)

Conclusion: Jets have such a big shot quality problem that their good Corsi becomes almost irrelevant as their good rating drops as soon as you take quality into account. Therefore, what you write, "So, the good news is the Jets were a worse even strength team last year at the 18 game point" only has very limited validity, as it only seems to be based on Corsi.

Edit: wait, is xG all situations?

 

[garret9]

% of SA that are HSC differential (HSCF/CF - HSCA/CA) explains about 3% of the variation in the NHL you have between difference in Corsi and goal% (goal% - Corsi%)

 

[garret9]

 

[mcpw]

% of SA that are HSC differential (HSCF/CF - HSCA/CA) explains about 3% of the variation in the NHL you have between difference in Corsi and goal% (goal% - Corsi%)

That sounds like HSC is almost useless.

What's the average sv% on HSC events vs non-HSC events?

Edit: not sure why you posted that graph. I never claimed a strong correlation between the two.

 

[garret9]

That sounds like HSC is almost useless.

What's the average sv% on HSC events vs non-HSC events?

Edit: not sure why you posted that graph. I never claimed a strong correlation between the two.

HSC is score close. It's useful in telling you what happened in a particular subset of conditions but misses a large chunk of the picture. Corsi still better than Corsi Close, and SA Corsi is better than both.

The graph is because, since SA Corsi and/or xG are better predictive, the use is more descriptive, but the descriptiveness is still very limited.

 

[mcpw]

HSC is score close. It's useful in telling you what happened in a particular subset of conditions but misses a large chunk of the picture. Corsi still better than Corsi Close, and SA Corsi is better than both.

?

I don't think war-on-ice HSC is score close. Can you give a source on that?

The graph is because, since SA Corsi and/or xG are better predictive, the use is more descriptive, but the descriptiveness is still very limited.

Once again, I never said there was a strong correlation so I don't know what point you're trying to make.

 

[garret9]

?

I don't think war-on-ice HSC is score close. Can you give a source on that?



Once again, I never said there was a strong correlation so I don't know what point you're trying to make.

The HSC and Score Close was an analogy.
Score Close adds some description due to score effects, but loses more due to sample. HSC adds some description due to shot quality, but loses more due to sample.


I'm saying if there is not strong correlation on a non-predictive metric, there is not that much concern on it.

 

[mcpw]

The HSC and Score Close was an analogy.

Honestly, that's an awful analogy. One is distinguishing an average shot while tied from an average shot while up three. The other is distinguishing an average shot from the crease from an average shot from the red line. Huge difference.

I'm saying if there is not strong correlation on a non-predictive metric, there is not that much concern on it.

That's not how it works.
CF% and GF% have a strong correlation, but it's not an ideal descriptive/predictive metric because of hidden variables.
One of these hidden variables is shot quality, we know that. That's the main reason why xG looks to be better than Corsi.
HSCF% should be one of these hidden variables.

I'll ask you again: what's the average sv% for HSC and nonHSC events?
Expected GA from Corsi description:
<GA> = CA*[1-E(CSv%)]
Expected GA from HSC-nonHSC description:
<GA> = HSCA*[1-E(HSCSv%)] + (CA-HSCA)*[1-E(nHSCSv%)]
with E(.) league average.

Compare these two for
a) a team with HSCA/CA = 20%
b) a team with HSCA/CA = 25%
How much of a difference over a full season for typical values of full-season CA? That should be the indicator of whether HSC is relevant or not.

Edit: I think you're assuming that I believe HSCF% > CF% in some way, descriptive or predictive. That's not the case. I just believe that HSCF% is able to provide some context. High HSCF% is good, low HSCF% is bad, it's a relevant information which is not captured by the Corsi description, but plays a role in the xG description.

 

[garret9]

Honestly, that's an awful analogy. One is distinguishing an average shot while tied from an average shot while up three. The other is distinguishing an average shot from the crease from an average shot from the red line. Huge difference.



That's not how it works.
CF% and GF% have a strong correlation, but it's not an ideal descriptive/predictive metric because of hidden variables.
One of these hidden variables is shot quality, we know that. That's the main reason why xG looks to be better than Corsi.
HSCF% should be one of these hidden variables.

I'll ask you again: what's the average sv% for HSC and nonHSC events?
Expected GA from Corsi description:
<GA> = CA*[1-E(CSv%)]
Expected GA from HSC-nonHSC description:
<GA> = HSCA*[1-E(HSCSv%)] + (CA-HSCA)*[1-E(nHSCSv%)]
with E(.) league average.

Compare these two for
a) a team with HSCA/CA = 20%
b) a team with HSCA/CA = 25%
How much of a difference over a full season for typical values of full-season CA? That should be the indicator with whether HSC is relevant or not.

Edit: I think you're assuming that I believe HSCF% > CF% in some way, descriptive or predictive. That's not the case. I just believe that HSCF% is able to provide some context. High HSCF% is good, low HSCF% is bad, it's a relevant information which is not captured by the Corsi description, but plays a role in the xG description.

Sigh. You don't get what I'm saying at all. I'm probably not communicating well.

It's a fine analogy IMO in discussing sample. Maybe I'll try a different example and it'll clear it up...

A single goal is far more predictive how a team or player will perform in the future than a single shot on goal. About 5 times stronger in fact.
Also, a team that does better in GA/SA = 20% will do comparably different than a team with GA/SA = 25%.

... if you see what direction this is going towards, it's PDO.

I don't believe you think HSC>Corsi, at all.

I'm just saying HSC provides some of the context of what happened, which is why I use it in conjunction with non-HSC... much like how I'll use shooting percentages to discuss context on performance of similar shot metric teams.
However, I don't really care too much about it. It's not very predictive, so I don't expect sustainability in too much. While adjusting for quality (like xG) you do better, looking at only the high quality ends up with too much small sample and bias as it ignores a lot of what is still a huge part of the game. Basically, quite often in hockey restricted statistics << all sample < adjusted.
And in the non-predictive usage, as I noted earlier, 60% of goals come from non-HSC. Wins still has a lot to do with beating the opposition in non-HSC shots. So a team doing well in HSC does not necessarily do well in the game, although they do better than had they not.

If I was a coach, I worry a bit more about it. Not because it tells me how a team will do, but shows some areas I can look at tape to help instruct/teach my team.

 

[mcpw]

Sigh. You don't get what I'm saying at all. I'm probably not communicating well.

Could be. I'm coming from the statistics side without having done a lot of actual hockey calculations, so bear with me

It's a fine analogy IMO in discussing sample. Maybe I'll try a different example and it'll clear it up...

Still no, imo. A close shot attempt should have the same success rate as a non-close shot attempt (am I right about that? It should be very close at least). A HSC shot attempt has a different success rate from a non-HSC shot attempt. It is therefore a biased subsample.

It's like defining success as rolling 9 or more with two dice, and then looking at the subsample of rolling a '5' with the first one, as opposed to rolling dice on a Monday.

A single goal is far more predictive how a team or player will perform in the future than a single shot on goal. About 5 times stronger in fact.
Also, a team that does better in GA/SA = 20% will do comparably different than a team with GA/SA = 25%.

... if you see what direction this is going towards, it's PDO.

I know that several people call PDO "luck", when that's just wrong. Tanguay and Lundqvist have high PDO because of who they are. Pavelec has a low PDO because of who he is. Of course, PDO includes what we assume to be statistical fluctuations (which could also be hidden variables we haven't discovered yet) - i.e. luck.

I'm just saying HSC provides some of the context of what happened, which is why I use it in conjunction of non-HSC... much like how I'll use shooting percentages to discuss context on performance of similar shot metric teams.
However, I don't really care too much about it. It's not very predictive, so I don't expect sustainability in too much.

You keep using the term "not predictive" for HSC. What do you actually mean by it?
a) predictive for success - low correlation of past HSCF% with future GF%. That's what I assumed what you meant, and that's why I repeated that I don't think HSCF% > CF% for judging teams.
b) now that I think about it, by "not predictive" you could mean low autocorrelation, i.e. past HSCF% and future HSCF% (or, even better, past and future HSCF/CF and HSCA/CA) are (almost) uncorrelated, i.e. any value of HSCF% is "unsustainable" (EDIT: to be clearer, it could also mean that HSCF/CF and HSCA/CA are not autocorrelated) which is very different from CF%. Is that what you mean? In that case, I too would be less concerned about it.
But then again, you talk about shooting percentages - and they should show autocorrelation (see Alex Tanguay, he has a long-range autocorrelation), right? Obviously the fluctuations in sh% are huge compared to the offset from league average.

While adjusting for quality (like xG) you do better, looking at only the high quality ends up with too much small sample and bias as it ignores a lot of what is still a huge part of the game.

I don't dispute that. Unfortunately, xG is still a black box for me (I don't think there's public access to a database with all the necessary info you need to calculate xG), so I only have access to what DTMAH posts on Twitter. So as an approximation I'm going by what I think is the most relevant ingredient (correct me if I'm wrong here): Going by WOI stats, we have
Sv%L 0.974
Sv%M 0.925
Sv%H 0.835
(That's on shots, not shot attempts, obviously ) So, highest priority should be to not allow high-danger shots - confirmed by eyetest, keeping shots to the outside is a good thing. Looking at the shot distribution for several goalies, I find that, for a league-average goalie, most goals against come from high-danger shots. Therefore, I believe that the distinction between HSC and non-HSC is the most important ingredient in the xG formula (I know that it still distinguishes between several classes of HSC). In conclusion, looking at HSCF% can tell you whether a team performs better or worse than its CF% indicates - that should be the first step from CF% to xG%, to quantify the effect of HSC. Super-extreme example: If a team has 60CF% and 90% HSCA/CA over a season, you would probably expect that it failed to makebthe playoffs. Predictive value isbhidden in the HSCA autocorrelation which I haven't calculated yet.

And in the non-predictive usage, as I noted earlier, 60% of goals come from non-HSC. Wins still has a lot to do with beating the opposition in non-HSC shots. So a team doing well in HSC does not necessarily do well in the game, although they do better than had they not.

Again, non-predictive future GF% or autocorrelation? Obviously, HSC is only one part of the game. Another super-extreme example: a team with 65% 5v5 CF%, 0.850 goaltending, 3% PP, 60% PK, and 0% OT wins probably missed the playoffs. Still, we use 5v5 metrocs as prediction and description for success, as 5v5 is the biggest part of the game.

2014-15 WPG
5v5 goals 143-123 (266 goals)
AllSit goals 223-204 (427 goals)

266/427 = 62.3%
We're using metrics describing less than 2/3 of goals.
So 40% of goals come from 5v5 HSC, you said.
That's still ~25% of all goals. That's very significant.
Obviously, if the HSC part is great and the rest is awful, the team is probably bad. Same vice versa.

If I was a coach, I worry a bit more about it. Not because it tells me how a team will do, but shows some areas I can look at tape to help instruct/teach my team.

I hope PauMau does

 

[garret9]

I think we agree having high HSCA and low HSCF is bad.

We disagree on the extent, how useful it is in analysis, and how telling it is of the future (in HSC or in Goals).

 

[mcpw]

I think we agree having high HSCA and low HSCF is bad.

We disagree on the extent, how useful it is in analysis, and how telling it is of the future (in HSC or in Goals).

That's why I asked you about the HSCsv% in the first place. That can settle it.

 

[garret9]

That's why I asked you about the HSCsv% in the first place. That can settle it.

Not necessarily. How sustainable 18-19 game sample is would matter to me as well.

 

[mcpw]

Not necessarily. How sustainable 18-19 game sample is would matter to me as well.

sv% and autocorrelation it is, then.

 

[garret9]

Won't take long to get sv%. Just have to not be busy at work.

Autocorrelation will take a bit longer.

 

[MrBoJangelz71]

Curious if there is any information or calculating of our strength of schedule played thus far, compared to our strength of schedule last season up to this point.

12 of 19 on the road, 4 back to backs already, all on the road, along with playing top teams in the league.

This was the hardest stretch of games we will play this season, and as long as we stay around .500, better days are ahead.

Also, hopefully some can remember attacking this team hard early on last season, only to eat crow later on. Hopefully those same people will be a little more refrained with their criticism of our play this season

 

[Gm0ney]

Curious if there is any information or calculating of our strength of schedule played thus far, compared to our strength of schedule last season up to this point.

12 of 19 on the road, 4 back to backs already, all on the road, along with playing top teams in the league.

This was the hardest stretch of games we will play this season, and as long as we stay around .500, better days are ahead.

Also, hopefully some can remember attacking this team hard early on last season, only to eat crow later on. Hopefully those same people will be a little more refrained with their criticism of our play this season

I thought I saw "venue adjusted" fancy stats somewhere. Some work has been done to look at home ice advantage. It exists (most teams get more points at home vs. the road, take less penalties at home, win more faceoffs at home)...not sure how strong a factor it is. More important, I think, would be the strength of your opponents - and we've been playing some pretty tough competition lately. Although we lost to the Flyers, too...

Why are you so concerned about people eating crow and being less critical? We're all on the Jets bandwagon for life, but that doesn't mean we can't ***** about the ride...

 

[mcpw]

Won't take long to get sv%. Just have to not be busy at work.

Autocorrelation will take a bit longer.

Just did a quick and dirty calculation. Looks like 18-game HSCA/CA can't predict future values well. So it looks like it's just a "here's what has gone wrong" metric.

Now get that unsustainbly high value down, PauMau

 

[mcpw]

here are the pictures: first 18 games vs rest of season

CA

HSCA/CA

 

[MrBoJangelz71]

I thought I saw "venue adjusted" fancy stats somewhere. Some work has been done to look at home ice advantage. It exists (most teams get more points at home vs. the road, take less penalties at home, win more faceoffs at home)...not sure how strong a factor it is. More important, I think, would be the strength of your opponents - and we've been playing some pretty tough competition lately. Although we lost to the Flyers, too...

Why are you so concerned about people eating crow and being less critical? We're all on the Jets bandwagon for life, but that doesn't mean we can't ***** about the ride...

I am old school. You support your team, moreso when things are going south.

At work, I cannot tolerate blamers, finger pointers, bus thrower underers. You accomplish absolutely nothing with people of this mindset.

Sure, they challenge your views, but not in a constructive manner. Their mandate is to be critical of everything that failed. They wait for the failure, then stand up on their milk crate, and point out how they saw it from a mile away. Then, when things improve, they sit back and wait for the next bump.

And I get it, this is a forum for discussing our team, the good and the bad. There are plenty of posters on here that apply constructive criticism in a reasonable manner. By no means do I believe this team is above criticism.

Its those that obsessively wait for us to experience some hard times so they can make silly statements as though they are absolute proclamations based off of a slew of games that didn't go our way.

But hey, if 7 games is all you need to determine that everything this organization has done up to this point is a massive failure, enjoy.

 

[Jetfaninflorida]

I am old school. You support your team, moreso when things are going south.

At work, I cannot tolerate blamers, finger pointers, bus thrower underers. You accomplish absolutely nothing with people of this mindset.

Sure, they challenge your views, but not in a constructive manner. Their mandate is to be critical of everything that failed. They wait for the failure, then stand up on their milk crate, and point out how they saw it from a mile away. Then, when things improve, they sit back and wait for the next bump.

And I get it, this is a forum for discussing our team, the good and the bad. There are plenty of posters on here that apply constructive criticism in a reasonable manner. By no means do I believe this team is above criticism.

Its those that obsessively wait for us to experience some hard times so they can make silly statements as though they are absolute proclamations based off of a slew of games that didn't go our way.

But hey, if 7 games is all you need to determine that everything this organization has done up to this point is a massive failure, enjoy.

Great speech, boss. Need help getting down from your milk crate?

Please don't go to hard on me. It's in jest.

 

[MrBoJangelz71]

Great speech, boss. Need help getting down from your milk crate?

Please don't go to hard on me. It's in jest.

You are far from the worst, your not even remotely in that range.