Greg Beyer

421 posts

Greg Beyer

@GregBeijer

Katılım Mart 2013

69 Takip Edilen28 Takipçiler

Greg Beyer@GregBeijer·4h

@JFreshHockey @HilsNoglander Found it, I believe. This doesn’t prove there’s no correlation, because you didn’t isolate players traded to better teams. It’s testing for a league-wide linear relationship across all team switches, so pooling positive and negative team changes together can wash out correlation

English

Greg Beyer@GregBeijer·1d

@JFreshHockey @HilsNoglander Can I see your test? I mean CBJ relative to the Canadiens team he left is basically a juggernaut.

English

369

JFresh@JFreshHockey·1d

great point, for example you can see here how Kyrou's numbers are bad because he plays for the Blues

X - WE TALK HOCKEY@wetalkhockeyy

I don’t fuck with analytics because of shit like this. Like if you follow the Habs, this is nothing crazy. You put a talented player on a shitty team, obviously his numbers are gonna look trash. Kyrou has 43 points this year. He’d be a PPG in Pittsburgh easily

English

1.1K

189.7K

Greg Beyer@GregBeijer·8h

@JeffVeillette @5andagame_ I mean as it should, these aren’t really comparable. One is a confident declaration by a major media outlet, the other is some minor pandering. Criticism should be warranted and the Habs are on the up, I don’t understand why you would expect there to be much right now.

English

214

Jeff Veillette@JeffVeillette·8h

@GregBeijer @5andagame_ People were mocking it the second it came out, too. This isn't a graphic that got rediscovered last week

English

777

Jeff Veillette@JeffVeillette·9h

The actual Montreal roster is generally pretty easy to like, but posts like this are why I’ve said over the years that Habs discourse is basically what people accuse Leafs discourse of being with a fraction of the criticism

PuckEmpire@puckempire

The dynasty of the early 2020's vs. the dynasty of the late 2020's...? TAMPA VS MONTREAL IN ROUND ONE. WHO YOU GOT?

English

242

42.5K

Greg Beyer@GregBeijer·8h

@JeffVeillette @5andagame_ Because they’ve failed to live up to these expectations for a decade. Same will happen to the Habs if things don’t pan out.

English

763

Jeff Veillette@JeffVeillette·8h

@5andagame_ And that cover has been ripped on and mocked for a decade straight now, I don't think that conflicts with the point here

English

1.7K

Greg Beyer@GregBeijer·1d

@JFreshHockey @HilsNoglander Would like to see the methodology of your test. Cant seem to find it, I vaguely remember it examining player underlying metrics before and after being traded. Don’t remember if you isolated players specifically traded to better teams though.

English

JFresh@JFreshHockey·1d

@GregBeijer @HilsNoglander there isn't one, we tested it two summers ago. there are examples of players who did better analytically on better teams, there are examples of players who did worse analytically on better teams. Monahan didn't exactly join a juggernaut in CBJ

English

1.7K

Greg Beyer@GregBeijer·1d

@JFreshHockey @HilsNoglander Wouldn’t call it a myth, there are plenty of supporting examples. Drouin resurgence with Colorado, Monahan with Columbus. Could be fit or team structure, but I do imagine there is a fair correlation between players playing better on better teams overall.

English

1.9K

JFresh@JFreshHockey·1d

@HilsNoglander nothing wrong with not liking analytics. but the "players on bad teams have bad stats" myth is a pet peeve

English

139

5.3K

Greg Beyer@GregBeijer·3d

@JFreshHockey Honestly I just don’t trust the Sportlogiq model. How has it compared to everyone else’s model in terms of point projections to actuals in the past 3-4 years?

English

123

JFresh@JFreshHockey·3d

The Jack Adams usually goes to a coach whose team most overperforms preseason expectations, but that means that goaltending or hot shooting plays a big part. Here's one (flawed) solution - comparing the preseason expectation to the team's performance filtering those two out.

English

291

116.1K

Greg Beyer@GregBeijer·4d

@JFreshHockey @SeanCFoley12 What have the correlation coefficients been over the years?

English

JFresh@JFreshHockey·4d

@SeanCFoley12 this has been an outlier year. every season it correlates very closely with goals, except this one. that probably also has something to do with why the standings have been so insanely unpredictable

English

539

JFresh@JFreshHockey·4d

Final Goal vs. xGoal differential numbers from SportLogiq this season, shared via TSN

English

169

74.2K

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato Highly doubt Makar’s production was simply redistributed. Even in that explanation, the split is still being shaped by deployment/context effects rather than cleanly identifying player value.

English

yolo_pinyato_is_dead@yolo_pinyato·24 Mar

@GregBeijer lol not even close since that would imply that the same phenomenon happened over thousands of minutes and only exclusively to Cale makar and absolutely nobody else for the entirety of his career. The points were simply redistributed throughout the lineup

English

yolo_pinyato_is_dead@yolo_pinyato·23 Mar

Slafkovsky is on pace to break 70 points which is a big jump in production for him, but the average forward would be on pace for 80 pts w/ his deployment (and the avg 21 year old would be on pace for 72). MTL is stacked offensively, but Slaf is more passenger than play driver.

yolo_pinyato_is_dead@yolo_pinyato

every "slafkovsky is good now" post is just a screen shot of his points because people don't know how to calibrate someone being the 6th best offensive player on a top 5 offense in the league lol

English

88.6K

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato Fairly stable isn’t exogenous. Lineups are still assigned by talent and role, so QoT remains endogenous; saying it gets more random lower in the lineup doesn’t solve that. And nothing I’ve said has been anecdotal; I’ve tried to keep the critique strictly logical.

English

yolo_pinyato_is_dead@yolo_pinyato·24 Mar

@GregBeijer top of lineups it is far more random further down the roster. So while this repartee has been fun and informative I don’t think you’ve couched your claims with much more than anecdotes and conjectures and so in lieu of you doing additional work and research on your own time

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato @IneffectiveMath QoT is also endogenous. And it being 5x–10x larger than QoC and close to isolated skill, doesn’t rescue the model it just means context is doing even more of the explanatory work.

English

yolo_pinyato_is_dead@yolo_pinyato·24 Mar

@GregBeijer Many of those same posts and studies also indicate the far more significant impact of QoT relative to QoC (about 5x-10x) and in the last screenshot from @IneffectiveMath a nearly similar impact to individual skill level and while teammate distribution is fairly stable at the

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato Ah yes, Devon Toews with a career high of 57 points is “replacing” Makar. I believe that whole w/o MacKinnon was debunked and attributed to partial changes off the rush, joint offensive zone starts etc etc. W/O metrics need to be contextualized.

English

yolo_pinyato_is_dead@yolo_pinyato·24 Mar

@GregBeijer makar is actually probably the worst example because there's a career long body of evidence that Colorado across an extended sample had similar or better results with toews replacing an injured makar in '21 and '23 and makar's scoring and impacts falling off a cliff w/o mackinnon

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato All other underlying inputs (teammates, competition, coaching, and deployment are still not random and jointly determined, making it unreliable. At best, this just shows public QoC is a poor metric; making your model as a whole weaker.

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato Your screenshots just argue that public QoC metrics unreliable, overly compressed, and weaker than commonly assumed. Could be true, but that is a measurement error argument, not a solution to endogeneity.

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato This really boils down to a simple question: does Makar get that usage because he’s Makar, or does that usage create Makar? You can’t assume the latter, build a baseline off it, and then project it onto average populations.

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato What studies show QoC is randomly assigned, this is completely illogical, Ovi has randomly assigned QoC? I’m not an expert, and I’m not claiming to be. I’m just critiquing your model, and pointing to why the conclusions you’re drawing are circular because of endogenous variables.

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato @AndrewBerkshire @kay_imw I never claimed I can identify the exact amount of the endogeneity from your chart alone. I’m saying the structure of the setup makes it unavoidable in principle as we discussed QoC/QoT are not randomly assigned. It’s a polluted dataset and thus unfit for evaluative claims

English

yolo_pinyato_is_dead@yolo_pinyato·24 Mar

@GregBeijer @AndrewBerkshire @kay_imw I have gone out of my way to point out that model in particular was a simplified bridge between RAPM and traditional stats yet you claim the level of endogeneity with certainty but have yet to provide any proof of it or approaches that you yourself would take to get around it

English

Greg Beyer@GregBeijer·24 Mar

@yolo_pinyato @AndrewBerkshire @kay_imw Whole narrative is built on endogenous context, so the conclusion ends up being circular. Your original graph has the exact same issues.

English

yolo_pinyato_is_dead@yolo_pinyato·23 Mar

@AndrewBerkshire @kay_imw bc he had 47 in 60 during the last full run of the data as I acknowledged in next post and yes, deployment can have that much of an impact on point performance especially given how much slaf's most common teammates have leveled up this year x.com/yolo_pinyato/s…

yolo_pinyato_is_dead@yolo_pinyato

About 70-75% of a players points can be explained by their teammates/competition/ice time, so a player with Makar's deployment would be expected to have ~45 pts this year (2nd highest behind Bouchard.) He's outscoring that by +10 pts. Werenski/Hutson are +25/+30. Seider is +5.

English

122

Greg Beyer@GregBeijer·23 Mar

@yolo_pinyato Similar to WAR doesn’t really mean much. Argument wasn’t about if context adjusted can be useful in principle, obviously it can. My objection was with your specific setup

English

yolo_pinyato_is_dead@yolo_pinyato·23 Mar

@GregBeijer You can derive enough additional useful information beyond raw points in a similar fashion to creating a wins above replacement metric that follow a similar approach

English

Greg Beyer@GregBeijer·23 Mar

@yolo_pinyato That actually reinforces my point. If his deployment is beyond even an average 1st liner’s, then ‘average forward’ is an even less realistic counterfactual. You’re comparing him to a player who basically doesn’t exist, then adding opponent adjustments on an even noisier sample.

English

yolo_pinyato_is_dead@yolo_pinyato·23 Mar

@GregBeijer And I agree it is rare for the average player to get Slafs deployment but his deployment is beyond that of an even an avg 1st liner and given the amount of times lines are switched up (and because the metric is also adjusting for opponent which is an even smaller set of samples)

English

Keşfet

@JFreshHockey @HilsNoglander @JeffVeillette @5andagame_ @SeanCFoley12 @yolo_pinyato @elonmusk @BarackObama