Toggle sidebar

Speculation: Caps General Discussion (Coaching/FAs/Cap/Lines etc) - 2021 "Season" Pt. 2

Status
Not open for further replies.
SherVaughn30

SherVaughn30

Registered User
Jan 12, 2010
4,845
3,002
Los Angeles
filinski77 said:
Anyone else feel like Ovi is being shadowed on the power play a lot more this season than in the past?

So far it seems almost impossible to get a good shot opportunity over to him.
Click to expand...
It's the Caps own fault because they don't execute enough player movement in the o-zone once they enter the zone. PK's just blanket OV's side of the ice. They just sit on him, knowing he will get that one-timer pass from Carlson. It's too easy and predictable to stop now.
 
g00n

g00n

Retired Global Mod
Nov 22, 2007
30,567
14,616
twabby said:
Yes, the sample is lopsided. As is like every 11 game sample in every other season. Could be a long west-coast road trip, could be a long homestand, could be a bunch of crappy or difficult teams in a row, could be that the coach is trying new lines, could be a bunch of injuries, could be a flu bug going around, could have a pregnant wife at home, etc. You can make up a million reasons why you don't like the sample, or that it shouldn't count. Rarely are you going to have an 11 game stretch and say "hey that looks random enough!"

So yes, in my opinion the onus is on you to prove your claim that his xGF numbers so far don't have any predictive value. And similarly, you would need to also prove that his point totals so far are indicative of his future totals in this sample, and that also your eye-test during this sample is predictive as well in order to evaluate them by these metrics. Because hey, the sample's really lopsided. How do we know that his counting stats aren't artificially suppressed by playing these two teams? How do we know that your eye-test isn't horribly skewed by him mainly playing against the Penguins and Sabres?

It seems like you aren't applying the same level of scrutiny to your methods of evaluation of the player so far.
Click to expand...

This is just nonsense. You're the one who asserted xGF was such a great predictor even based on 11 games, as supported by your link that I already addressed. Was that data based on playing the same team 6 times? Or the same 2 teams for ~80% of the games?

I don't need to provide counter-statistics to tell you it's crap. So here:

It's crap.

Also feel free to point to my "eye test" assertions of unprovable fact. Because there aren't any.
 
  • Like
Reactions: CapitalsCupReality, Calicaps and zappa4ever
twabby

twabby

Registered User
Mar 9, 2010
13,710
14,623
g00n said:
This is just nonsense. You're the one who asserted xGF was such a great predictor even based on 11 games, as supported by your link that I already addressed. Was that data based on playing the same team 6 times? Or the same 2 teams for ~80% of the games?

I don't need to provide counter-statistics to tell you it's crap. So here:

It's crap.

Also feel free to point to my "eye test" assertions of unprovable fact. Because there aren't any.
Click to expand...

I asserted that it has a correlation of about r^2 = 0.23 to future GF, nothing more. A noticeable, but not incredibly strong predictor. It's possible he'll regress because it has happened for plenty of players. But I'm not betting on it. It's based on thousands of different 10 game samples (different than the 11 he has played, but close enough IMO to use the 0.23 as the r^2), then averaged. This includes samples of playing the same team 4-5 times, or only a couple of teams instead of 11 different teams. Again, if you want to say that the current sample of games he has played, a sample in which he has significantly outperformed his teammates, is so unique that we can disregard the predictive value that has been shown, then fine! But it doesn't pass any level of scrutiny.
 
  • Like
Reactions: ALLCAPSALLTHETIME
HTFN

HTFN

Registered User
Feb 8, 2009
12,265
10,898
twabby said:
Yes, the sample is lopsided. As is like every 11 game sample in every other season. Could be a long west-coast road trip, could be a long homestand, could be a bunch of crappy or difficult teams in a row, could be that the coach is trying new lines, could be a bunch of injuries, could be a flu bug going around, could have a pregnant wife at home, etc. You can make up a million reasons why you don't like the sample, or that it shouldn't count. Rarely are you going to have an 11 game stretch and say "hey that looks random enough!"

So yes, in my opinion the onus is on you to prove your claim that his xGF numbers so far don't have any predictive value. And similarly, you would need to also prove that his point totals so far are indicative of his future totals in this sample, and that also your eye-test during this sample is predictive as well in order to evaluate them by these metrics. Because hey, the sample's really lopsided. How do we know that his counting stats aren't artificially suppressed by playing these two teams? How do we know that your eye-test isn't horribly skewed by him mainly playing against the Penguins and Sabres?

It seems like you aren't applying the same level of scrutiny to your methods of evaluation of the player so far.
Click to expand...
I think you're being willfully ignorant here, and the premise you're bucking against is fairly basic. This isn't any other season, where the opposition is highly randomized outside of a few home and home series. The way that this season is being put together is unique enough that we won't really be able to say a lot about the statistics until after it's concluded. It's ripe for autopsy, but not something you can honestly take this "business as usual" statistical approach to and be as certain as you are.
 
  • Like
Reactions: CapitalsCupReality and g00n
g00n

g00n

Retired Global Mod
Nov 22, 2007
30,567
14,616
twabby said:
I asserted that it has a correlation of about r^2 = 0.23 to future GF, nothing more. A noticeable, but not incredibly strong predictor. It's possible he'll regress because it has happened for plenty of players. But I'm not betting on it. It's based on thousands of different 10 game samples (different than the 11 he has played, but close enough IMO to use the 0.23 as the r^2), then averaged. This includes samples of playing the same team 4-5 times, or only a couple of teams instead of 11 different teams. Again, if you want to say that the current sample of games he has played, a sample in which he has significantly outperformed his teammates, is so unique that we can disregard the predictive value that has been shown, then fine! But it doesn't pass any level of scrutiny.
Click to expand...

Sure it does. It belies common sense to believe that playing the same team for more than half your games is going to predict how anyone will play against all the other teams, all other things being close enough to equal.

You also said way more than "it has a correlation of about r^2 = 0.23 to future GF". To paraphrase, you said 11 games was more than adequate to judge a player's expected future performance based on xGF, which you supported with an article that merely showed xGF was slightly less shitty than other clearly shitty fancy stats.

What's more that analysis was based on 10 game sets from a handful of seasons, repeated 1,000 times. Do you know how many different combinations of 10 games you can get from an 82 game season? It's a lot more than 1000. It's more like 2 trillion.

That's a sample size so small it makes 11 games look enormous.

So what are the chances you got a good representation of data where the players were facing the same team 6 times out of 10 in that set, during seasons where they faced the same team 4-5 times at most? Based on 1,000 trials of 2 trillion possible outcomes?

Nearly zero.

So I say again, the stat is low-confidence. aka crap.
 
Last edited:
  • Like
Reactions: CapitalsCupReality
txpd

txpd

Registered User
Jan 25, 2003
69,649
14,131
New Bern, NC
Anyone remember Martin Fehervary? A bunch of us had him as high as the 2nd pair this season. He seems so forgotten
 
Raikkonen

Raikkonen

Dumb guy
Aug 19, 2009
10,720
3,171
Russia
GMBM got many quality defenders lately. We cant even lose much in ED now. Still see a need to trade Siege and/or Kempny and/or Jensen.

Interesting, TVR sits instead of Jensen while many thought latter is the worse player.
 
twabby

twabby

Registered User
Mar 9, 2010
13,710
14,623
g00n said:
Sure it does. It belies common sense to believe that playing the same team for more than half your games is going to predict how anyone will play against all the other teams, all other things being close enough to equal.

You also said way more than "it has a correlation of about r^2 = 0.23 to future GF". To paraphrase, you said 11 games was more than adequate to judge a player's expected future performance based on xGF, which you supported with an article that merely showed xGF was slightly less shitty than other clearly shitty fancy stats.

What's more that analysis was based on 10 game sets from a handful of seasons, repeated 1,000 times. Do you know how many different combinations of 10 games you can get from an 82 game season? It's a lot more than 1000. It's more like 2 trillion.

That's a sample size so small it makes 11 games look enormous.

So what are the chances you got a good representation of data where the players were facing the same team 6 times out of 10 in that set, during seasons where they faced the same team 4-5 times at most? Based on 1,000 trials of 2 trillion possible outcomes?

Nearly zero.

So I say again, the stat is low-confidence. aka crap.
Click to expand...

The problem with assuming the schedule makes all of the observations so far meaningless is that it ignores all of the non-schedule related factors that could explain a player's performance between one season and the next: offseason workout, coaching, maturity level changing, etc. all are pretty much independent of the following season's schedule. Are you taking any of these into account when concluding that his current results aren't useful in predicting the rest of his season?

For example, Kuznetsov was a much better player under Barry Trotz than he was under Todd Rierden. Every year under Barry Trotz was better than both years he had under Todd Rierden in terms of shot differential metrics, and especially on the defensive side of things. Now Todd Rierden is gone and another experienced coach with a track record of success has been brought in. Is it really that unreasonable to assume that Peter Laviolette has led to a quick turnaround in Kuznetsov's game, especially when most posters agree that Todd Rierden was not a very good coach? People are high on Laviolette and Kuznetsov is one of the most talented players on the team and in the NHL, according to many posters here. Isn't that a reasonable possibility to explain his excellent shot metrics so far? Why are we to conclude that it's the schedule, and therefore his numbers so far are meaningless? I don't think Peter Laviolette is going anywhere any time soon, and I don't think Todd Rierden is going to make his triumphant return to Washington.

Also, I believe you still have yet to address the possibility of random chance being a factor in the game of hockey. If you do accept random chance being a factor in hockey, then you necessarily will conclude that there is no perfect model that will predict future outcomes with 100% certainty. Perhaps the theoretical maximum predictability after 10 games is closer to r^2 = 0.23 than you expect? How do you know one can do much better than this? Have you tested it?

Regarding the bolded above: averages converge very quickly when performing samples. It is rarely if ever needed to exceed a thousand or so as a sample size. If you follow political polling you will see most sample sizes are usually only a few hundred to maybe a thousand or so participants even though there are 239 million eligible voters in the United States, for example. Any sample greater than a thousand or so is unnecessary because there is no point in drilling down to hundredths or thousandths of point in accuracy. Is there really a big difference between r^2 being 0.23 under the method used in the article posted previous, or do you really need it to be shown that it's actually 0.226863428? It'd be a waste of time. Especially since, unlike political polling, it is possible to get truly unbiased random samples of data under the method the author used, so the sample size would need to be even smaller to get a good confidence interval.
 
  • Like
Reactions: traparatus, Ridley Simon and ALLCAPSALLTHETIME
g00n

g00n

Retired Global Mod
Nov 22, 2007
30,567
14,616
twabby said:
The problem with assuming the schedule makes all of the observations so far meaningless is that it ignores all of the non-schedule related factors that could explain a player's performance between one season and the next: offseason workout, coaching, maturity level changing, etc. all are pretty much independent of the following season's schedule. Are you taking any of these into account when concluding that his current results aren't useful in predicting the rest of his season?

For example, Kuznetsov was a much better player under Barry Trotz than he was under Todd Rierden. Every year under Barry Trotz was better than both years he had under Todd Rierden in terms of shot differential metrics, and especially on the defensive side of things. Now Todd Rierden is gone and another experienced coach with a track record of success has been brought in. Is it really that unreasonable to assume that Peter Laviolette has led to a quick turnaround in Kuznetsov's game, especially when most posters agree that Todd Rierden was not a very good coach? People are high on Laviolette and Kuznetsov is one of the most talented players on the team and in the NHL, according to many posters here. Isn't that a reasonable possibility to explain his excellent shot metrics so far? Why are we to conclude that it's the schedule, and therefore his numbers so far are meaningless? I don't think Peter Laviolette is going anywhere any time soon, and I don't think Todd Rierden is going to make his triumphant return to Washington.

Also, I believe you still have yet to address the possibility of random chance being a factor in the game of hockey. If you do accept random chance being a factor in hockey, then you necessarily will conclude that there is no perfect model that will predict future outcomes with 100% certainty. Perhaps the theoretical maximum predictability after 10 games is closer to r^2 = 0.23 than you expect? How do you know one can do much better than this? Have you tested it?

Regarding the bolded above: averages converge very quickly when performing samples. It is rarely if ever needed to exceed a thousand or so as a sample size. If you follow political polling you will see most sample sizes are usually only a few hundred to maybe a thousand or so participants even though there are 239 million eligible voters in the United States, for example. Any sample greater than a thousand or so is unnecessary because there is no point in drilling down to hundredths or thousandths of point in accuracy. Is there really a big difference between r^2 being 0.23 under the method used in the article posted previous, or do you really need it to be shown that it's actually 0.226863428? It'd be a waste of time. Especially since, unlike political polling, it is possible to get truly unbiased random samples of data under the method the author used, so the sample size would need to be even smaller to get a good confidence interval.
Click to expand...

Randomness? Coaching? I'm pretty sure in searching for excuses for flaws in your statistical models you argued yourself in a circle, and whether you realize it or not came to the conclusion that there are enough unquantifiable, mitigating factors in predicting outcomes that raw statistical representation isn't enough to tell the story.

You can parse out 22 or 23% to a million decimal places and it's still only halfway to coin-flip. Rounding is not sample size.

Regarding sample size, there are formulas and calculators for that. You can look them up, and test to see if 1,000 vs 239 million is the same as 1,000 vs 2.1 Trillion. (Hint: the sample size needed is a hell of a lot more than 1,000).

There's also the fact that polling questions have set outcomes among populations that are generally known to behave in certain patterns, so even the mathematically relevant sample size to produce the desired confidence level might shrink. That's not the same as digging for performance metrics represented by highly variable numbers as your data rather than answers to closed, multiple-choice questions.

But again, do the math and see what you find.
 
  • Like
Reactions: HTFN
twabby

twabby

Registered User
Mar 9, 2010
13,710
14,623
g00n said:
Randomness? Coaching? I'm pretty sure in searching for excuses for flaws in your statistical models you argued yourself in a circle, and whether you realize it or not came to the conclusion that there are enough unquantifiable, mitigating factors in predicting outcomes that raw statistical representation isn't enough to tell the story.

You can parse out 22 or 23% to a million decimal places and it's still only halfway to coin-flip. Rounding is not sample size.

Regarding sample size, there are formulas and calculators for that. You can look them up, and test to see if 1,000 vs 239 million is the same as 1,000 vs 2.1 Trillion. (Hint: the sample size needed is a hell of a lot more than 1,000).

There's also the fact that polling questions have set outcomes among populations that are generally known to behave in certain patterns, so even the mathematically relevant sample size to produce the desired confidence level might shrink. That's not the same as digging for performance metrics represented by highly variable numbers as your data rather than answers to closed, multiple-choice questions.

But again, do the math and see what you find.
Click to expand...

Again, you appear to be conflating the "how" and the "what". I don't know how Kuznetsov has turned around his numbers. Maybe Laviolette has helped him regain his form. Maybe Kuznetsov himself did some soul-searching this offseason and decided he needed to train better and dedicate himself to the game moreso than the past. Perhaps a wizard came and cast a spell on him making him a dominant player in terms of xGF%. It'd be an odd spell, but I suppose it's possible. Maybe it's all of the above.

Unfortunately we are not privy to the "how" (at least I am not), but we do know that his on-ice performance metrics are significantly different, and better, than they were for the prior two seasons. I'm not trying to claim how he did it, I'm just saying it's more likely than not that his level of play is going to continue based on past data. I don't care if it's mostly attributable to Peter Laviolette, or Kuzy's wife, or his kids, or a wizard. I'm just saying he is likely to continue to have a positive net effect on the team's success as measured by on-ice goal differential. As someone who wants the Capitals to keep winning, that's all that I can ask for.

Regarding sample size:

Sample Size Calculator - Confidence Level, Confidence Interval, Sample Size, Population Size, Relevant Population - Creative Research Systems

upload_2021-3-2_22-49-1.png

upload_2021-3-2_22-50-45.png


Sample Size Calculator

upload_2021-3-2_23-2-8.png


Indeed at some point once the population size becomes large enough the sample size needed doesn't depend on the population size at all. It does depend on the margin of error you are looking for: I think 3% is a reasonable margin of error, so maybe r^2 is as low as 0.20 or as high as 0.26, but for some reason I don't think this would alter your argument. I consulted the formulas and calculators as you asked, and I came to the same conclusion I did beforehand: 1000 is plenty big as a sample size, even if there were an infinite population size. Perhaps you have a better calculator, and perhaps these sites are both using the incorrect formula? I'd love to learn something new!

And while the answers to polling are typically multiple choice, and the data presented in our discussion are "highly variable numbers," fortunately we can still build confidence intervals around "highly variable numbers." That's the whole point of building confidence intervals. Confidence interval
 

Attachments

  • upload_2021-3-2_22-48-4.png
    upload_2021-3-2_22-48-4.png
    9.6 KB · Views: 1
Last edited:
  • Like
Reactions: ALLCAPSALLTHETIME
Status
Not open for further replies.

Ad

Latest posts

Upcoming events

Ad

Ad

Top Bottom