Speculation: Top 5 leading scorers for 2018-2019 - #3

[Nithoniniel]

and you are welcomed to use a variable weighted to icetime with an elite center. When I see REITM actually drop for Marner this year which is consistent with your description of what happens, I will also tout the virtues of this.

Seriously, go back and read up on what RelTM is because it's clear that you don't understand the stat. If you did, you wouldn't be trying to bring up the linemates argument.

It's quite simple. RelTM measures how much you help your linemates. How much you improve their numbers. Do you think that Matthews needs a lot of help? Because that's what your argument says, that Nylander plays with an elite center a lot, and it is much easier to improve his numbers than if you played with a more limited linemate who need you more.

 

[Notsince67]

Seriously, go back and read up on what RelTM is because it's clear that you don't understand the stat. If you did, you wouldn't be trying to bring up the linemates argument.

It's quite simple. RelTM measures how much you help your linemates. How much you improve their numbers. Do you think that Matthews needs a lot of help? Because that's what your argument says, that Nylander plays with an elite center a lot, and it is much easier to improve his numbers than if you played with a more limited linemate who need you more.

You assume I haven't read about it. I have. They each have their contextual place in a conversation but the criticism of RELTM has been the observation that players who are together a lot have almost identical REL TM numbers. Nowhere have I read that WOWY is thrown out because of REL TM numbers.
This is consistent with my view that weighing heavily against a lot of icetime with an elite centerman might help quantitatively to predict point contributions, but it fails to illustrate qualitative effects on individual team-mate play.
I am confident enough to predict that the WOWY differences next year between the 2 players will remain, but the REL TM numbers of Marner will rise.

 

[Nithoniniel]

You assume I haven't read about it. I have. They each have their contextual place in a conversation but the criticism of RELTM has been the observation that players who are together a lot have almost identical REL TM numbers. Nowhere have I read that WOWY is thrown out because of REL TM numbers.
This is consistent with my view that weighing heavily against a lot of icetime with an elite centerman might help quantitatively to predict point contributions, but it fails to illustrate qualitative effects on individual team-mate play.
I am confident enough to predict that the WOWY differences next year between the 2 players will remain, but the REL TM numbers of Marner will rise.

Players that spend almost all the time together makes it hard to assign credit for any stat. Including WOWY.

Both WOWY and RelTM provides quantitative information, neither is qualitative in nature. I have no idea where you are going with that.

Care to elaborate on how example their WOWY differences will remain the same but the RelTM for Marner will go up? Because I don't see how that would happen. You can see WOWY look the same while RelTM goes up if a player played a lot with guys he was not successful with one year, and the next year play with guys he is successful with. That's not applicable here.

 

[Notsince67]

Players that spend almost all the time together makes it hard to assign credit for any stat. Including WOWY.

Both WOWY and RelTM provides quantitative information, neither is qualitative in nature. I have no idea where you are going with that.

Care to elaborate on how example their WOWY differences will remain the same but the RelTM for Marner will go up? Because I don't see how that would happen. You can see WOWY look the same while RelTM goes up if a player played a lot with guys he was not successful with one year, and the next year play with guys he is successful with. That's not applicable here.

I already did. When looking at Wowy as a stat by itself, the impact of your play varies player by player because you can look quite good with a crap player. The overall net team impact does differ though. That is ok because I am not looking at overall net team contribution...just individual qualitative contribution. I can compare apples and oranges players because my metric is not how much better but how many players are better. Your metric aims to look at overall team impact of an individual on a weighted individual based on icetime...which is ok but kind of masks the situation if you play with 1 person 85% of the time really well and 25% with others really poorly.

 

[Nithoniniel]

I already did. When looking at Wowy as a stat by itself, the impact of your play varies player by player because you can look quite good with a crap player. The overall net team impact does differ though. That is ok because I am not looking at overall net team contribution...just individual qualitative contribution. I can compare apples and oranges players because my metric is not how much better but how many players are better. Your metric aims to look at overall team impact of an individual on a weighted individual based on icetime...which is ok but kind of masks the situation if you play with 1 person 85% of the time really well and 25% with others really poorly.

Couple of comments:
1) This does not address your claim that his RelTM would go up but WOWY remain unchanged.
2) What you are talking about is not qualitative data. The amount of players improved is quantitative data. Their experience of playing with Marner would be qualitative data.
3) By only looking at the number of players improved, you draw conclusions based on what happens in sample sizes so small that they are pretty much all noise. It's absolute folly from a statistical analysis standpoint.

As for point #1, I'm going to guess that your reasoning is that his WOWY will remain the same in terms of the amount of players he improved. That would be ignorant to how extremely unreliable such measurements in small sample sizes are. It's not something one should expect. All those tiny samples might actually look bad next season, and it would say nothing about Marner's impact. It would just mean that for example a Johnsson overlapped his shift when they were getting pinned in those few minutes, instead of in the offensive zone.

 

[Notsince67]

Couple of comments:
1) This does not address your claim that his RelTM would go up but WOWY remain unchanged.
2) What you are talking about is not qualitative data. The amount of players improved is quantitative data. Their experience of playing with Marner would be qualitative data.
3) By only looking at the number of players improved, you draw conclusions based on what happens in sample sizes so small that they are pretty much all noise. It's absolute folly from a statistical analysis standpoint.

As for point #1, I'm going to guess that your reasoning is that his WOWY will remain the same in terms of the amount of players he improved. That would be ignorant to how extremely unreliable such measurements in small sample sizes are. It's not something one should expect. All those tiny samples might actually look bad next season, and it would say nothing about Marner's impact. It would just mean that for example a Johnsson overlapped his shift when they were getting pinned in those few minutes, instead of in the offensive zone.

It would prove your point for sure. I'm not so sure it happens. It might seem like a bad sample size and I would question the results if they looked random. In Marner's case the far majority of the team shows positive results. Beyond that, his Rel TM will begin to converge with the a high end player like Tavares. We can revisit when sufficient data becomes available.
I used to validate credit scoring models and while they do degrade over changes in the population behavior, they did manage to differentiate risk from a good bad perspective for a long time. How they translated to overall losses is what shifted. Weighing factors needed to be recalibrated. I suspect the same thing here. REL TM will be the metric that shifts but the underlying behavior behavior will not...just the relevant importance of each variable.
I would be willing to admit I am wrong...are you?

 

[member 300185]

1-Marner
2-Tavares
3-Matthews
4-Nylander
5-Kadri
6-Rielly
7-Gardiner

 

[Nithoniniel]

It would prove your point for sure.

No it would not. It wouldn't prove anything. The sample sizes are too small.

It might seem like a bad sample size and I would question the results if they looked random. In Marner's case the far majority of the team shows positive results.

It doesn't seem like a bad sample size. It is a bad sample size. That's simple fact. We're dealing with sample sizes that are far, far off the minimum that anyone in the field would suggest. What you basically are saying here is that numbers subject to incredible variation happen to say what you want, so you lend them credence.

I would be willing to admit I am wrong...are you?

Of course. But I see no reason to, given that everything from goal metrics, shot metrics of all types, and measured individual actions support what I'm saying. I'm not going to toss all that aside because Marner did better when on the ice with guys he barely played a regular shift with than Nylander did.

You're tossing aside overwhelming statistical evidence contrary to your position in favor of appeals to authority and a statistical analysis on such shaky grounds that nobody who works in the field would give it a second thought. So forgive me if I am a bit skeptical to your claim here.

 

[Notsince67]

No it would not. It wouldn't prove anything. The sample sizes are too small.


It doesn't seem like a bad sample size. It is a bad sample size. That's simple fact. We're dealing with sample sizes that are far, far off the minimum that anyone in the field would suggest. What you basically are saying here is that numbers subject to incredible variation happen to say what you want, so you lend them credence.


Of course. But I see no reason to, given that everything from goal metrics, shot metrics of all types, and measured individual actions support what I'm saying. I'm not going to toss all that aside because Marner did better when on the ice with guys he barely played a regular shift with than Nylander did.

You're tossing aside overwhelming statistical evidence contrary to your position in favor of appeals to authority and a statistical analysis on such shaky grounds that nobody who works in the field would give it a second thought. So forgive me if I am a bit skeptical to your claim here.

That is completely your opinion. Please show me evidence that WOWY has been completely discredited and isn't worthy of any analytical assessment.
Does Sean Tierney populate his chart for the sake of having nothing better to do?
Can you describe how many minutes are considered statistically significant?
Which player data should be excluded?
I'm afraid the appeals to authority seem to be your trademark on this topic. You dismiss the data you don't like with no real explanation other than the data is too small. Is it all the data or some of the data? Can you give some guidance of reference? I have laid out my explanation why I think Rel TM can be biased around a conversation about who drives a line yet on my assessment of published data (wowy), you insist the data is crappy because you know a guy (or know a guy who knows a guy)and he says so.

Sigh...do you even dispute the fact that Marner is likely to have a better Rel TM playing with an elite center?

 

[Nithoniniel]

Sigh...do you even dispute the fact that Marner is likely to have a better Rel TM playing with an elite center?

Well yes, because I understand how RelTM works. Jesus. For his RelTM to improve playing with Tavares, he needs to help Tavares more than he did Bozak and Kadri. Tavares does not need help, Bozak did. Best way to get a high RelTM is to play a lot with someone who fall apart without you. You think that'll happen for Tavares without Marner? Learn what the damn stats do.

you insist the data is crappy because you know a guy (or know a guy who knows a guy)and he says so.

When people start making up things to discredit those they argue with, it's quite obvious they have lost the plot.

I'm done with you now.

 

[Notsince67]

Well yes, because I understand how RelTM works. Jesus. For his RelTM to improve playing with Tavares, he needs to help Tavares more than he did Bozak and Kadri. Tavares does not need help, Bozak did. Learn what the damn stats do.


When people start making up things to discredit those they argue with, it's quite obvious they have lost the plot.

I'm done with you now.

The first real criticism of RELTM I saw was when Weber played with Josi. They had nearly identical numbers. Were they identical players? Of course not but their relative contribution was measured pretty closely based on their collective success. It is similar to Rielly and Hainsey. Without Haisey, Rielly CF% soars closer to 60% that 50%. Can you not see that playing with a better player elevates your CF% and that weighting to minutes will completely wash out any situational analysis.
The point of WOWY is completely about line construction analysis. It is absolutely important for situational decisioning.

As for your comment of : "Learn what the damn stats do."
From: Revisiting Relative Shot Metrics – Part 1

Problems
In an abstract sense, all relative to teammate methods inevitably suffer from “multicollinearity” when players spend a large amount of time together. This has been noted in various ways by Manny here, Brian MacDonald here, Shane Jensen here and here, A.C. Thomas et al. here, and Dawson Sprigings here among others.
Manny quite succinctly dubbed this the “Sedin Paradox”. With a large enough sample, the multicollinearity issue isn’t as problematic; however, in small samples and with player-pairs like the Sedins, this can be a huge problem. Even during a full season, pairs of players often play so much time together (90%+) that their relative numbers will be heavily influenced by what one teammate did in a relatively small number of minutes away from that player. This problem becomes apparent with the way the weighted average is calculated (a player’s teammates who played the most with that player are given the most weight in the calculation).


This is what I have been saying. Doctor...heal thyself.

Edit: From the same document....
WOWYs are rather difficult to use outside of situational analysis. They can be great for looking at how lines are constructed, or how a given player impacted his teammates (who’s a drag, etc.), but the output is rather cumbersome – WOWYs are not meant to be “summed”.

Seems like I know what the stats do quite well

 

[lukeleim]

Mitchell Marner 82GP, 28G, 56A, 84PTS
Auston Matthews 79GP, 44G, 39A, 83PTS
John Tavares 82GP, 39G, 41A, 80PTS
William Nylander 82GP, 32G, 40A, 72PTS
Morgan Rielly 80GP, 11G, 52A, 63PTS
Nazem Kadri 82GP, 30G, 25A, 55PTS
Jake Gardiner 82GP, 7G, 42A, 49PTS
Patrick Marleau 82GP, 28G, 21A, 49PTS
Andreas Johnson 78GP, 27G, 20A, 47PTS
Zach Hyman 82GP, 17G, 23A, 40PTS

 

[Nithoniniel]

Even during a full season, pairs of players often play so much time together (90%+) that their relative numbers will be heavily influenced by what one teammate did in a relatively small number of minutes away from that player

And this is not the case with Nylander and Matthews. They have a large enough sample size away from each other that multicollinearity is not an issue. Swing and a miss.

Seems like I know what the stats do quite well

You are of course referring to the articles that state that RelTM is the superior stat? Which has been my point since this discussion started?

Without Haisey, Rielly CF% soars closer to 60% that 50%. Can you not see that playing with a better player elevates your CF% and that weighting to minutes will completely wash out any situational analysis

Again, this shows you don't understand RelTM. It is not relative CF%. They are not the same thing. Now your argument is that Nylander, like Hainsey in your example, benefits from playing with a superior player. Look at Hainsey. His CF% in these numbers would rise, for sure. You know what happens with his RelTM? It would notice that Rielly without Hainsey is performing much better, and would use that relative to give Hainsey a much lower RelTM.

Being carried by Rielly gives Hainsey a worse RelTM. That's why his RelTM was one of the worst on the team. Because the stat correctly noticed that Hainsey was carried by a greater player, who did much better without him. The opposite is true for Nylander.

You claim that Marner will get a better RelTM when he upgrades from Bozak to Tavares. Like I described above, this would only happen if the difference between Bozak with Marner and Bozak without Marner was smaller than it is for Tavares, ie that Tavares is more dependent on Marner. And that's not very likely, is it?

You talk again about weighing. The reason the stat is weighed is to properly calculate a particular player's impact on the aggregated numbers of his linemate. Another thing you seem to misunderstand.

And this is why I tell you to learn about the stats. This is now the fourth time I've described how the stat work. You claim to have read how the stat work. I know you are a clever guy, so I can only deduce that you are being purposefully obtuse at this point. And I cannot fathom why, since the only thing this means is that a player on your team is better than you perhaps thought.