• Moderators, please send me a PM if you are unable to access mod permissions. Thanks, Habsy.

Leafs sign John Tavares to 7 year/ $77m deal

leafman101 said:
Well it is when people rely solely on p/60 to judge a player. Its still just a counting stat, not a predictive one. Its flawed to rely on it solely to judge offensive talent. Thats the point. Probability of future performance is important, and in the interest of accuracy, should not be completely ignored.
Click to expand...

Except that's not what the piece was saying. They were saying that when there is more data in your sample, the better confidence you can have in it that past performance being repeatable in the future. Their entire premise (as illustrated in the first paragraph) is that a relatively small sample of P/60 (in this case, just one season of Hischier) doesn't produce high confidence in future likelihood of repeating that production (keeping in mind of course that the small sample also lacks the confidence to predict increases in production, which are less likely than regressing to production that is closer to average, but still possible). On the flip side of that, they argue that Kucherov's 3 yrs worth of data generate a result that is far more predictive. They're right.

Another thing that absolutely doesn't support your assertion here, is that this analysis actually uses P/60 as it's only measure of offensive performance. The authors here are taking for granted the opposite thing you're claiming the piece suggests. They're taking for granted that P/60 is an excellent measure of offensive talent, and it's their assertion that "true talent" can be better determined by utilizing a bigger sample size of that one metric.

If anything, that increases the importance of P/60 in evaluation. But sample size is ****ing important.

With all of that said, the JT argument has never rested solely on P/60. P1/60, CF%, XGF, etc have all been part of the broader discussion.
 
Yeah man, this piece is saying the exact opposite of what you think it is:

Every time a skater manages to gain a primary point, the time it took for him to do so is noted. The excess time left between the skater’s most recent primary point and the moment he last got off the ice is simply added to the time it took to score the skater’s first primary point (this is perfectly acceptable thanks to the memoryless property of the exponential distribution). Each one of those lengths of time is a piece of information that says something about the skater’s scoring talent. A player’s collection of primary points and the time required for their production are treated as independent, identically distributed exponential random variables.
Click to expand...

That is a far better, more eloquent defence for the importance of P1/60 than I can personally put forward.
 
plenty is missing - JVR had better p/60 numbers than Tavares the past couple years.

but we know why that's not a relevant comparison.

but I try my best to make comparisons that are rigorous and relevant.
 
JVR also had better p1/60 numbers than when he played more minutes.

Panarin had better p1/60 numbers than he did playing with Kane last year, over similar minutes, but against lesser competition.

Kessel basically put up the same P1/60 (aside from his shitty last year here) over different, teams, teammates, completion and ice time.

There are a lot of differing results. Information is missing.
 
zeke said:
Basically I think there's probably an argument to be made that substantially easier icetime and better linemates would take care of the fewer minutes, but I was hoping somebody had done some legwork on this.
Click to expand...

I'm not a mathemagician, but here's my take on it.

I don't think Marner's QoT will change much at all if the lines are what they're rumoured to be. Marleau-Kadri in 5 on 5 ice is pretty similar in offensive quality to Hyman-Tavares. So any improvements in Marner's scoring rates will be driven by individual improvement or random fluctuation moreso than QoT.

If we look at JT, he's likely to play with the best offensive linemate he's ever had (Marner), but also one of the worst he's ever had (Hyman). Bailey appeared to be a 1.7ish P/60 guy in absensce of Tavares, and Lee is harder to pin down but he appears to be at least a 1.7 guy in absence of JT as well. Marner appears to be at least a 1.9+ guy and probably a 2.0+ (potentially more if the 2nd half Marner is the real Marner going forward). High talent players do tend to develop weird synergies that can drive scoring rates higher than they would normally be, so I would be surprised if both JT and Marner pushed into 2.1+ territory, especially with slightly lesser QoC (I don't expect JT's usage to be sheltered at all, I just expect Matthews and Kadri's to be less severe and the load more evenly shared...Kadri probably won't be taking 60% of his faceoffs in the defensive zone this year, there's just no Bozak around to have to hide anymore).

Where I think JT's QoT improves significantly is on the PP. If there's going to be a significant increase in his scoring rates, it's going to be playing with Mitch and Kadri on that 1A unit, replacing JVR in that low slot position for tips and such.
 
Brayden Schenn this year had a significantly higher P1/60 playing more minutes, against tougher competition, with better teammates.
 
Tavares over the last 3 years had 2.08, 1.28 and 1.61 P1/60, despite almost identical minutes, teammates and competition.
 
leafman101 said:
It is saying that half the equation is missing when simply relying on P/60.
Click to expand...

Read it again, or at the minimum quote me the section you think is supporting this.

It uses primary points as literally it's only means of measuring offensive talent, and defends it's usage of P1/60 in the passage I quoted. They are literally treating primary points and the time intervals between them as the holy grail in measuring true talent. It's the underpinning feature of every statistical distribution they generation to try to provide that confidence score at the end.
 
leafman101 said:
Brayden Schenn this year had a significantly higher P1/60 playing more minutes, against tougher competition, with better teammates.
Click to expand...

and that's their argument in the piece. That small samples can't generate strong confidence in future production. It doesn't generate enough confidence to consider that a true measure of skill.

A larger sample can though.
 
leafman101 said:
Tavares over the last 3 years had 2.08, 1.28 and 1.61 P1/60, despite almost identical minutes, teammates and competition.
Click to expand...

Yep, which sums to 1.69 over the 3 yr total. Which is probably a pretty good measure of his true talent.

Brayden Schenn over the same time frame produces a 1.24.
 
MindzEye said:
Read it again, or at the minimum quote me the section you think is supporting this.

It uses primary points as literally it's only means of measuring offensive talent, and defends it's usage of P1/60 in the passage I quoted. They are literally treating primary points and the time intervals between them as the holy grail in measuring true talent. It's the underpinning feature of every statistical distribution they generation to try to provide that confidence score at the end.
Click to expand...

Well the post the equation.
posterior.png


It is insanely more complicated that just factoring in sample size.

It turns out that Kucherov has a sample that is about three times larger (3359 minutes vs. 1077 minutes) and so it is reasonable to expect that his observed scoring rate is likely more indicative of his true scoring talent. However, the degree to which we should feel more comfortable with the data being in Nikita’s favor and how that factors into our comparison of the two players is unclear.
Click to expand...

That is what they are trying to determine. What is the probability of repetition. And the conclusion clearly states that it is more than simply scoring rates + sample size:

Knowing the probability that one skater is the more efficient even strength scorer is significantly more enlightening than simply eye-balling their raw scoring rates and respective sample sizes. We can generate the same plot for any two skaters, as long as they play the same position. Here is Travis Dermott vs. Jake Gardiner…
Click to expand...

...which is exactly what we have been doing here. Shouldn't be any shocker that our looking at the stats on Corsica doesn't offer the same level of insight as the insane depth these guys go into, and is flawed.
 
Last edited:
zeke said:
how about this - put out some names that you think saw a signficiant increase or decrease in production due to playing on a more or less stacked team, and I'll do some legwork on it.
Click to expand...

For in a season ...Miller/McD in Tampa
 
Yeah, I'm not claiming to be able to do what they're doing by hitting 3 buttons on corsica bro.

But they're utilizing P1/60 and sample size in minutes to generate a probability of whichever player producing a higher P1/60 than another player in the future.

The entire exercise depends on primary points/60 being an accurate representation of past production. The entire piece presumes that P1/60 is the most meaningful measure of past offensive production.
 
MindzEye said:
Yep, which sums to 1.69 over the 3 yr total. Which is probably a pretty good measure of his true talent.

Brayden Schenn over the same time frame produces a 1.24.
Click to expand...

Ok but what is the predictive value of that? Particularly in the face of changing ice time, competition, and teammates.

Does that mean Tavares will score 1.6 P1/60 next year? Is it possible for him to score 2.08 P1/60 again next year? If so does that mean he became better player? Does that mean he lived up to the hope as zeke would suggest?
 
MindzEye said:
Yeah, I'm not claiming to be able to do what they're doing by hitting 3 buttons on corsica bro.

But they're utilizing P1/60 and sample size in minutes to generate a probability of whichever player producing a higher P1/60 than another player in the future.

The entire exercise depends on primary points/60 being an accurate representation of past production. The entire piece presumes that P1/60 is the most meaningful measure of past offensive production.
Click to expand...

Well I am not sure thats trues, as xGF has produced the best predictive value of anything currently publicly available. But that used to be Corsi. And used to be ppg before that.

Point being, just because it is the "most meaningful measure of past offensive production" does not mean it has the most predictive value for future performance. What that paper does is try to provide predictive value by calculating the probability of a player scoring all possible P/60 based on past production and league wide distributions. They are going to great lengths to add value, not display the value of P/60 itself.

Which actually answers zeke's original question, can Tavares improve his P/60 next year and actually give you the liklihood of that. Although I have no idea what to do with their code to figure that out.
 
Last edited:
leafman101 said:
JVR also had better p1/60 numbers than when he played more minutes.

Panarin had better p1/60 numbers than he did playing with Kane last year, over similar minutes, but against lesser competition.

Kessel basically put up the same P1/60 (aside from his shitty last year here) over different, teams, teammates, completion and ice time.

There are a lot of differing results. Information is missing.
Click to expand...


I think we can do better than throw our hands in the air and say "who knows"?

All three of those examples seem to show that lesser competition was the biggest factor, to me.
 
leafman101 said:
Brayden Schenn this year had a significantly higher P1/60 playing more minutes, against tougher competition, with better teammates.
Click to expand...

1yr sample isn't very useful - and Schenn is a great example. He had an insane first 20-30gms like he's never had before, and then settled down to his usual pace the rest of the way.
 
zeke said:
I think we can do better than throw our hands in the air and say "who knows"?

All three of those examples seem to show that lesser competition was the biggest factor, to me.
Click to expand...

I agree. I tend to think competition is always one of the bigger, if not the biggest, factors in these things where generally people never seem to talk about competition.
 
zeke said:
1yr sample isn't very useful - and Schenn is a great example. He had an insane first 20-30gms like he's never had before, and then settled down to his usual pace the rest of the way.
Click to expand...

Agreed, I'm just picking players that changed teams, roles etc. to see what happens.
 
You must log in or register to reply here.
Back
Top