2018-19 stats and underlying metrics thread

[YWGinYYZ]

It’s almost like Corsi is a limited stat that like everything else is extremely flawed and shouldn’t be blindly followed.

I mean just look at the GF% clearly shows the skill difference.

Of course it's a limited stat - even people active in the hockey stat community state that this is so, and why there's ongoing improvements to the models being made. Corsi simply measures shot attempts taken - no more, no less. Everything else (xG, etc) is a refinement on top of it.

It's still in its infancy, and it's a much more complex game than baseball, where it's easier to isolate player vs. player, etc.

 

[WPGChief]

It’s almost like Corsi is a limited stat that like everything else is extremely flawed and shouldn’t be blindly followed.

I mean just look at the GF% clearly shows the skill difference.

GF% only really shows the skill difference in net in this specific example...

 

[Tom ServoMST3K]

Wouldn't a team's performance above/below their xGF show "skill" level?

Like some players are just better shooters.

Or is it just chalked up to variance?

 

[Whileee]

Wouldn't a team's performance above/below their xGF show "skill" level?

Like some players are just better shooters.

Or is it just chalked up to variance?

Statistical (random) error.

Measurement error (systematic or non-systematic).

Talent.

 

[Mortimer Snerd]

I think there are a few drivers:

1) Players (Blake) have admitted to conserving themselves to make it through the grind through to June. (the concern is they can't rediscover the 60-minute form of last year's home stretch)

2) Injuries to Buff, Ehlers and Morrissey have hampered their play over the past two months.

3) Missing the Toby-Buff pairing has really hampered our d-zone play from the second pairing perspective.

Which team hasn't had injuries?

When the injury load is sharply worse than normal you might be able to use that excuse. I don't think this is that year.

 

[Maukkis]

It’s almost like Corsi is a limited stat that like everything else is extremely flawed and shouldn’t be blindly followed.

I mean just look at the GF% clearly shows the skill difference.

The skill of having better goaltending than you could reasonably expect, while your defense is garbage?

 

[Tom ServoMST3K]

Statistical (random) error.

Measurement error (systematic or non-systematic).

Talent.

I assume these are in order of effect?

 

[raideralex99]

Of course it's a limited stat - even people active in the hockey stat community state that this is so, and why there's ongoing improvements to the models being made. Corsi simply measures shot attempts taken - no more, no less. Everything else (xG, etc) is a refinement on top of it.

It's still in its infancy, and it's a much more complex game than baseball, where it's easier to isolate player vs. player, etc.

Exactly MLB is one on one.
Corsi is 6 vs 6 ... way to many variables ... plus throw in the B2B or 3 games in 4 nights or a week layoff schedule and your stats are contaminated.
A bigger stat which should be talked about and I have been saying it for years and they finally did improve it ... is the scheduling.
Certain team benefiting from the schedule ... team A gets to play 13 games where a team played the night before or their 3rd game in 4 nights while team B only plays 5 games with an advantage.
I first noticed this with the Jets 1.0. They use to go into Alberta play a great game against Gretzky's Oilers only to lose in the 3rd period and than play the Flames the next night and get blown out but when the playoffs came the Flames had no advantage and the Jets would beat them in the series.

 

[AKAChip]

Exactly MLB is one on one.
Corsi is 6 vs 6 ... way to many variables ... plus throw in the B2B or 3 games in 4 nights or a week layoff schedule and your stats are contaminated.
A bigger stat which should be talked about and I have been saying it for years and they finally did improve it ... is the scheduling.
Certain team benefiting from the schedule ... team A gets to play 13 games where a team played the night before or their 3rd game in 4 nights while team B only plays 5 games with an advantage.
I first noticed this with the Jets 1.0. They use to go into Alberta play a great game against Gretzky's Oilers only to lose in the 3rd period and than play the Flames the next night and get blown out but when the playoffs came the Flames had no advantage and the Jets would beat them in the series.

Okay... but these things balance out over a long sample size. This argument is extremely flawed.

 

[YWGinYYZ]

There's always variation in games - what you're describing is a single event that has a 55-45 chance of happening one way or the other on any given night.

Corsi is 6 vs 6 ... way to many variables ... plus throw in the B2B or 3 games in 4 nights or a week layoff schedule and your stats are contaminated.

Not really - that's not how statistical analysis works.

 

[Whileee]

I assume these are in order of effect?

Don't know, really. There usually isn't enough information in the public domain to assess.

 

[garret9]

There's always variation in games - what you're describing is a single event that has a 55-45 chance of happening one way or the other on any given night.



Not really - that's not how statistical analysis works.

The idea that computers can’t suss out things when hockey is 6v6 is absolutely and certainly ridiculous.

Either one of two things have to be true:

1) it’s just multivariable/linear algebra and we can solve who tilts the ice

2) it’s completely random and hockey cannot be solved... therefor there is no player better than the other and you might as well also fire all your scouts and pick players randomly and pay league minimum

There is no other alternatives

 

[garret9]

Exactly MLB is one on one.
Corsi is 6 vs 6 ... way to many variables ... plus throw in the B2B or 3 games in 4 nights or a week layoff schedule and your stats are contaminated.
A bigger stat which should be talked about and I have been saying it for years and they finally did improve it ... is the scheduling.
Certain team benefiting from the schedule ... team A gets to play 13 games where a team played the night before or their 3rd game in 4 nights while team B only plays 5 games with an advantage.
I first noticed this with the Jets 1.0. They use to go into Alberta play a great game against Gretzky's Oilers only to lose in the 3rd period and than play the Flames the next night and get blown out but when the playoffs came the Flames had no advantage and the Jets would beat them in the series.

The fact that hockey is 6v6 doesn’t make statistics impossible. Just makes it more involved and even more important.

If science can’t figure out how certain players are better than others at a silly game of putting a rubber thing in a net because there are 12 variables than we as a civilization wouldn’t have gotten even this far.

It’s simple math:
Deep Dive on Regularized Adjusted Plus-Minus I: Introductory Example

 

[garret9]

It’s almost like Corsi is a limited stat that like everything else is extremely flawed and shouldn’t be blindly followed.

I mean just look at the GF% clearly shows the skill difference.

The issue is GF% represents a smaller portion of the picture than Corsi does.

Or... if you blindly followed one, you’d at least be wrong less often using Corsi.

On ice sv% at the player level acts as stable as a RNG. GF actually has useful information but GA is fairly poor performing commodity thanks to the inconsistency that is goaltending.

Corsi isn’t flawed. It actually perfectly measures what it does: shot volume.

People’s analysis using Corsi can be flawed because:
1) people are flawed
2) Corsi isn’t everything but it’s part of that everything

I've even shown before how much of everything Corsi is:
A closer look at Corsi, how much it matters, and what it...

Micah McCurdy also has shown how Corsi is the driving factor of showing who wins playoffs:
HV Article

Here we see that good regular season goaltending is crucial---that is, teams with sub-par regular season goaltending have virtually never gone on long playoff campaigns. Shooting talent, on the other hand, appears to mean very little, as the blue patch spreads horizontally through a huge swathe of five standard deviations or so. The blander colours are also a symptom of this spread; we are more sure that a shot-production juggernaut will have a deep playoff run than we are of a team with strong goaltending doing the same.

 

[winnipegger]

1) it’s just multivariable/linear algebra and we can solve who tilts the ice

2) it’s completely random and hockey cannot be solved... therefor there is no player better than the other and you might as well also fire all your scouts and pick players randomly and pay league minimum

There is no other alternatives

I get what you're saying, but can't it be a combination of those statements.

Take for example the game of chess....it is not solvable (kind of like hockey imo). HOWEVER. You can objectively say which moves are better than other moves, and you can prove it. You can't say "on move #1 you do this and you will win the game". You CAN say "on move #1 you do this and you increase your likelihood of winning the game". I think this is the crux with all of this scientifically based hockey analysis. You can't exactly deny the descriptive power of it (at least if you want to stay on valid conversational grounds). But the prescriptive power of it is limited...since you can't separate and isolate each variable and test them independently. Hope that makes sense I have been hitting the bottle tonight.

 

[LaScheifers]

I get what you're saying, but can't it be a combination of those statements.

Take for example the game of chess....it is not solvable (kind of like hockey imo). HOWEVER. You can objectively say which moves are better than other moves, and you can prove it. You can't say "on move #1 you do this and you will win the game". You CAN say "on move #1 you do this and you increase your likelihood of winning the game". I think this is the crux with all of this scientifically based hockey analysis. You can't exactly deny the descriptive power of it (at least if you want to stay on valid conversational grounds). But the prescriptive power of it is limited...since you can't separate and isolate each variable and test them independently. Hope that makes sense I have been hitting the bottle tonight.

you would be correct in your assessment as it's not so absolute as one or the other. Its multivariable and random.

 

[Neuf]

Jets have played fewest 3 point games this season at 10. they have collected 58.4% of available points, good for 5th in the league.

Florida is at the top of 3 point games at 19. Last year they had the fewest.

Nothing meaningful here, just a stat I've been watching. Some teams and divisions have a lot more points up for grabs than others.

Central 1,111
Pacific 1,272
Metropolitan 1,271
Atlantic 1,277

 

[DashingDane]

I just listened to a great podcast episode where professional poker player Annie Duke spoke about decision making strategy. She talks a lot about utilizing data to inform those decisions. Her main example is Pete Carroll superbowl decision that most people consider a mistake. I highly recommend listing as I think it's relevant to hockey. Invest Like the Best by Patrick O'Shaughnessy on Apple Podcasts.

It made me wonder if there are any ways to measure decision making in hockey and thought it would be interesting to discuss in general.

Mods: feel free to delete if this is the wrong thread but the podcast made me think of underlying metrics and how we use them.

 

[garret9]

Jets have played fewest 3 point games this season at 10. they have collected 58.4% of available points, good for 5th in the league.

Florida is at the top of 3 point games at 19. Last year they had the fewest.

Nothing meaningful here, just a stat I've been watching. Some teams and divisions have a lot more points up for grabs than others.

Central 1,111
Pacific 1,272
Metropolitan 1,271
Atlantic 1,277

Generally speaking, win% in 1 goal games or OT/SO games are other ways to look at over/under performance that is likely to regress aside from underlying things like xGoals and Corsi.

 

[Maukkis]

- The unnamed small logo above Laine is Hayes.
- While Copp has retained his great level of play from last year, the same just cannot be said about Tanev and Lowry (Tanev is above Lowry between bad and overused).
- The top guns are all in the good quadrant (minus Laine), so that is good. Their scoring rates at 5v5 are from those of best in the business, but at least they remain above average.
- Roslovic is the only one in bad, and the sample size is quite significant too. Steps forward were expected of him, and hands down - he has not delivered. Hopefully we'll see more next year.

- Full credit to Chiarot for having put together a decent season. Thinking the scoring rate is a bit of an outlier, though, but convincing overall.
- We. Need. To. Dump. Kulikov.
- Morrissey and Trouba have switched scoring rates, if you will. Trouba has been way over 1 point / 60 at 5v5 for the past few years, but the 5v5 scoring woes and an increasing amount of tough assignments have probably taxed his rate (despite the raw point total being at an all-time high).

 

[AKAChip]

So the eye test matches the stats with Lowry. He's a bum. Tanev has got to be one of the most lowkey overrated Jets in history.

 

[Whileee]

So the eye test matches the stats with Lowry. He's a bum. Tanev has got to be one of the most lowkey overrated Jets in history.

They're good defensively, not good offensively. Nothing new, really.

 

[Duke749]

They're good defensively, not good offensively. Nothing new, really.

I’m not sure what some are expecting out of Lowry honestly. Come playoffs I think that line does exactly what we want to and need to do.

 

[AKAChip]

They're good defensively, not good offensively. Nothing new, really.

Lowry's defensive stats this year have been beyond abysmal as well. He's at a -7.2 CF% and a -7.5 FF% after being well in the positives last season. Granted a lot of that has to do with being very bad on offense but you don't have numbers that terrible as a premiere defensive player. Tanev is having the "best" season of his career by those metrics at -8.2 and -8.1 respectively.

 

[DRW204]

I’m not sure what some are expecting out of Lowry honestly. Come playoffs I think that line does exactly what we want to and need to do.

Expecting better than 13 pts (at 5v5), 0.9 pts/60 (down from 2.1 last year) and a positive at 5v5 +/-. Defensively he's still superb but not as great as last year. Offensively it's a night and day difference