2018-19 stats and underlying metrics thread
- Thread starter garret9
- Start date
But hasn't Tanev been improving? I'm skeptical of summary analyses that combine all games and contexts early in a career, unless there is no trajectory. If there is upward or downward trajectory, it should be reflected. It seems that Dano's trajectory has been downward since his rookie season, whereas Tanev has been improving. Ignoring that seems to miss an important part of the story.It's no secret that Tanev kills offesne... Simply look at shot location:
Trends? Seems to me that Dano has been trending down and Tanev trending up. Very early to get a good read on Lemieux, though he seems to lack some key attributes necessary to be a really effective player.Stats update:
Analysis
- Perreault is good at everything and the best of the bunch, as expected. He tilts the ice more so in shot quantity than he does in shot quality, but he is the top-six talent in the bottom-six for the Jets.
- Lowry is all defense when it comes to shot volume, and including shot quality makes him look even better. That said, he does well at making the Jets few shots they make come from prime scoring areas.
- Copp is the Jets best pure defensive player. His defensive impact exceeds even Lowry. While he's not void of any offense, he is not exceptionally good at it either.
- Dano is very similar to Copp, although most of his defensive value comes from shot quality. Dano may be a better point producer than Copp, but they are quite similar in producing chances for the team.
- Petan's results interest me. It does remind one that offensive/defensive skill is not the same as impact, and individual performance is not the same as team performance. Part of it is likely Petan been trying to play a certain way to appease Maurice et al., and part of it is likely just to do with RAPM accounting for quality of linemates but not chemistry with linemates.
- Roslovic hasn't done well in improving his linemates offense, but it should be noted that in terms of individual scoring performance he has been the best of the entire bunch listed here.
- Appleton has done well treading above water, but it is a small sample so his confidence intervals are the largest.
- Tanev is extremely adept at defense, and extremely inept at offense. Again, this is a reminder that individual offense is different than team offense with the individual on the ice. Tanev works well with Lowry and Copp as they become specialized, but outside of that usage he becomes more harm than helpful.
- It is early, and things can change, but the early returns of Lemieux have not been as positive as Petan, Roslovic, or Appleton. Unlike Tanev, he doesn't excel in one particular area so he cannot be used as a specialist.
garret9
AKA#VitoCorrelationi
Difficult to read into trends when sample size is 2-4: random variation can look like trends.
Ex:
Dano looks like he’s trending down if you look just at this and last season... but he’s trending up if you group by 2 seasons at a time as last season was his best.
Ex:
Dano looks like he’s trending down if you look just at this and last season... but he’s trending up if you group by 2 seasons at a time as last season was his best.
garret9
AKA#VitoCorrelationi
Dano’s best season was last year...
Issue with trajectory:
1) people sometimes (often) get fooled by fluctuations
2) people have a bad guess at how trajectories are moving
I don’t mind using Marcels or something similar IF it improves projection... but it actually doesn’t always.
Rounding out this point, current performance and trajectory probably needs a closer look for players at the beginning of their careers. Should be relatively straightforward to look at adjustments / reliability around early career performance.
i think this is a huge part of it that gets overlooked. petan is definitely doing more than he needs to and straying further from his true game just so that he can stay in the lineup. the perks of having a coach with terrible roster management and talent evaluation abilities.
i think a little bit of the same mentality has been slowly creeping into laine's game as well lately. not good
garret9
AKA#VitoCorrelationi
Trajectory sure, but don't be fooled by randomness. In fact, I'd say ignoring trajectory is better than trying to include it if you are not letting a computer model it because we are really bad at thinking randomness is a pattern...
left is not random
right is random
left is not random
right is random
I understand randomness well.Trajectory sure, but don't be fooled by randomness. In fact, I'd say ignoring trajectory is better than trying to include it if you are not letting a computer model it because we are really bad at thinking randomness is a pattern...
left is not random
right is random
But I'd like to know if performance and trajectories at the beginning of NHL careers can be considered part of a randomly distributed set of parameters when combined over the first 2-4 seasons. If not, then comparing players by collapsing a few years together and ignoring trends / trajectories is overly simplistic. The graph below strongly suggests that performance at the beginning of forwards' careers are not randomly distributed over time.
garret9
AKA#VitoCorrelationi
It depends if you are asking how good a player did and how bad they will do.
Regardless, one issue with trajectory models vs something more like a Marcel or model using priors is the more information you use, the smaller your cohort pool becomes, the more likely you are ending up with a spurious correlation.
At this time, with what we have, I find it much better to just relatively compare people in similar situations... Like for example Lemieux has been still bad whether as a NHL player or compared to early use players like Petan and Appleton.
Regardless, one issue with trajectory models vs something more like a Marcel or model using priors is the more information you use, the smaller your cohort pool becomes, the more likely you are ending up with a spurious correlation.
At this time, with what we have, I find it much better to just relatively compare people in similar situations... Like for example Lemieux has been still bad whether as a NHL player or compared to early use players like Petan and Appleton.
Right, but how confident are we that comparisons at that stage with those sample sizes are reliable?
garret9
AKA#VitoCorrelationi
All confidence is relative.
I will say that I'd be more confident in small samples from those measures than any other comparison method.
garret9
AKA#VitoCorrelationi
Yes.
Just like strength of relation is quantifiable, but is also relative.
Take for example complex in world correlation vs in a controlled lab.
Point being all small sample evaluation would be low.
Eye test, professional or otherwise, model, etc. but some are lower than others.
You unfortunately can't play 300 NHL games to see which prospects will play best in the NHL as a test.
Just like strength of relation is quantifiable, but is also relative.
Take for example complex in world correlation vs in a controlled lab.
Point being all small sample evaluation would be low.
Eye test, professional or otherwise, model, etc. but some are lower than others.
You unfortunately can't play 300 NHL games to see which prospects will play best in the NHL as a test.
Psych0dad
Registered User
Agreed and appreciated!
I'm sure there are simulations available, but naturally they rely completely on rather inadequate data input and can only predict to some level of accuracy.
- Mar 10, 2010
- 34,904
- 31,383
Maukkis
EZ4ENCE
- Mar 16, 2016
- 10,596
- 7,344
Different models value Hellebuyck quite differently. I think Corsica has him as one of the worst starters still (although slowly improving), whereas EW and Moneypuck hold him in a higher regard. As with most things, the truth is probably somewhere in the middle.
AlphaLackey
Registered User
Here's a question: someone brought up the issue of Hellebuyck's SV% versus his salary, and it reminded me that for whatever reason, "how good *should* Hellebuyck be doing?" has proven a tricky one to answer. Does anyone have any historical data on SV% expectations relative to salary?
Doing a quick "by hand" calculations of Cap Friendly's top 7 matches to Hellebuyck for comparable contracts, and taking the line of best fit, suggests a .915 is "right" for a 6.1M, but this is clearly an insufficient sample size hampered by the fact that the two biggest salaries (Bob and Rask) are near the bottom of the pool. Check out that R^2 :/
So yeah, if there's been any significant work done on this, I'm all ears!
Doing a quick "by hand" calculations of Cap Friendly's top 7 matches to Hellebuyck for comparable contracts, and taking the line of best fit, suggests a .915 is "right" for a 6.1M, but this is clearly an insufficient sample size hampered by the fact that the two biggest salaries (Bob and Rask) are near the bottom of the pool. Check out that R^2 :/
So yeah, if there's been any significant work done on this, I'm all ears!
garret9
AKA#VitoCorrelationi
AlphaLackey
Registered User
I'm really hoping that we can change that sooner than later
I guess there's really nothing better than a linear fit of save percentage versus salary, weighted by raw # of shots against?
WPGChief
Registered User
I think the closest you could get to your answer is through Matt Cane - he does contract predictions for RFAs and UFAs yearly and has been pretty bang-on on most, with the exception of goaltenders up to this past year. However, I can't find a specific thread on Twitter or blog post that references exactly what he used for this year in how he improved those goalie predictions, unfortunately. Here are some posts though:
The Time Value of Money and Player Valuation
Quick Post: Do Past Sv% Variables Predict Future Sv% Variables?
Predicting Free Agent Salaries
The Time Value of Money and Player Valuation
Quick Post: Do Past Sv% Variables Predict Future Sv% Variables?
Predicting Free Agent Salaries
Maukkis
EZ4ENCE
- Mar 16, 2016
- 10,596
- 7,344
Results shown above are affected by SH% and time on ice. This information provided by garret.
- Our net negative players include three of the six defensemen you are going to see a lot with Byfuglien out of the lineup. Fun?
- An assortment of fourth liners near the break-even point fits in well with the notion of our fourth line being underutilised. Hard to have an impact without ice time. Appleton has been impressive, though.
- The only impact Laine provides is on the PP, where he is very good. Other than that, a no show. f*** paying him anything close to 10.
- I am genuinely surprised to see Ehlers' EV impact being that low.
- Little and Lowry? WTF?!
- Trouba and Morrissey practically neck and neck, with most of the difference coming from Morrissey's PP minutes. You could have predicted this kind of impact from our top pairing, and you also could have reacted to that by locking them up in the summer.
Joe Morrow > Mathieu Perreault.
Results shown above are affected by SH% and time on ice. This information provided by garret.
- Our net negative players include three of the six defensemen you are going to see a lot with Byfuglien out of the lineup. Fun?
- An assortment of fourth liners near the break-even point fits in well with the notion of our fourth line being underutilised. Hard to have an impact without ice time. Appleton has been impressive, though.
- The only impact Laine provides is on the PP, where he is very good. Other than that, a no show. **** paying him anything close to 10.
- I am genuinely surprised to see Ehlers' EV impact being that low.
- Little and Lowry? WTF?!
- Trouba and Morrissey practically neck and neck, with most of the difference coming from Morrissey's PP minutes. You could have predicted this kind of impact from our top pairing, and you also could have reacted to that by locking them up in the summer.
Good to know...
Results shown above are affected by SH% and time on ice. This information provided by garret.
- Our net negative players include three of the six defensemen you are going to see a lot with Byfuglien out of the lineup. Fun?
- An assortment of fourth liners near the break-even point fits in well with the notion of our fourth line being underutilised. Hard to have an impact without ice time. Appleton has been impressive, though.
- The only impact Laine provides is on the PP, where he is very good. Other than that, a no show. **** paying him anything close to 10.
- I am genuinely surprised to see Ehlers' EV impact being that low.
- Little and Lowry? WTF?!
- Trouba and Morrissey practically neck and neck, with most of the difference coming from Morrissey's PP minutes. You could have predicted this kind of impact from our top pairing, and you also could have reacted to that by locking them up in the summer.
Looks like more evidence that we should be playing the Petan, Roslovic and Appelton line and giving them some decent minutes. Ours third line has been sub par this year and Lowry hasn't been as good as he was the last two years. Maurice needs to recognize this and adapt his usage accordingly based on this metric our fourth line has been more impactful and should be played more.
Moreover it looks like Lemieux is going back in tomorrow. I really question Maurice at times.
Myers, Kulikov and Chariot all sub par as a whole and need to be upgraded.
