puck stoppa
Registered User
We should pick up Aaron dell and miles wood just so we can hear beyak say their full names all year.
Speculation: Jets General Rumour, Trade, Free Agent and Waiver Speculation 16-17 Part XVII
- Thread starter Romang67
- Start date
- Status
Even-strength save % (2016/17) - among goalies with 15+ games
Schneider 0.921 (#31 in NHL)
Hellebuyck 0.915 (#45 in NHL)
Hutchinson 0.914 (#47 in NHL)
Halak 0.905 (#52)
Neuvirth 0.901 (#56)
Schneider 0.921 (#31 in NHL)
Hellebuyck 0.915 (#45 in NHL)
Hutchinson 0.914 (#47 in NHL)
Halak 0.905 (#52)
Neuvirth 0.901 (#56)
Cam Talbot went for a 2nd and some other late picks. No reason we couldn't bundle a similar package for a similar goalie.
puck stoppa
Registered User
Gil Fisher
Registered User
I think league-wide this offseason will be crazy. LV really throws a wrench in team's plans and I can't wait to see what unfolds 
Especially excited to see who moves in the goalie market. There will be some big splashes.
Especially excited to see who moves in the goalie market. There will be some big splashes.
- Jun 10, 2014
- 57,582
- 29,494
I think league-wide this offseason will be crazy. LV really throws a wrench in team's plans and I can't wait to see what unfolds
Especially excited to see who moves in the goalie market. There will be some big splashes.
Could be interesting last minute before the XD - or more precisely before the lists have to be finalized.
Then again as soon as the dust from that settles. Expect LV to be very active.
Tom ServoMST3K
In search of a Steinbach Hero
The most spectacular Beyak moments are when Nino Niederreiter is having a strong shift.
garret9
AKA#VitoCorrelationi
I didn't distort anything since I included OBPM for this season with the career numbers.
Your reference to the points production of linemates over the past 7 seasons over-estimates the productivity of Lowry's most common linemates, giving the distorted impression that he's been playing with more offensively productive players than he has. That's what I was referring to. None of Armia, Matthias or Lowry look particularly good in the OBPM metric, do they?
You'd need to provide more information about the statistical methods used to "adjust for linemate quality" for me to comment further on the OBPM metric. Have you looked at Perreault's OBPM when he is playing with bottom-6 linemates? It is quite possible in models to combine different strata and try to smooth the relationship when there is actually quantitative interaction, or at least residual confounding, which undermines the validity of the model. If Perreault (or Ehlers) played exclusively with bottom-6 linemates like Armia, Thorbs, Burmi and Stafford, what would their OBPM look like?
A quick way of looking at this would be to compute directly adjusted points/60 and CF% to account for differences in linemates (or line strata), but I don't have the time.
My guess is that anyone who plays with the bottom-6 on the Jets over the past few seasons gets mired into a low productivity swamp. Some call it the "Thorburn effect" or the "Burmi effect".
Mud Turtle
Registered User
- Jul 26, 2013
- 8,205
- 18,737
garret9
AKA#VitoCorrelationi
If you are curious of models you can look them up to learn more.
I tire of this disingenuous argument and this board at times.
I tire of this disingenuous argument and this board at times.
Yeah he had talent - but I'd heard he didn't like being on the Moose. He figured he was too good for the AHL or something. Just thought maybe that's who Gil was remembering as not liking it here.
It's disingenuous to ask for more information about models? Statisticians do this all the time, but it usually doesn't lead to charges of disingeuity.
My question was quite straightforward. How does the OBPM adjust for linemate quality, and is the adjustment robust enough to account for large discrepancies in linemate mix. Also, what timeframe is used to parameterize linemate quality?
If you don't want to include that information here, a link to a methodological paper or two would be fine. I'm happy to read the source methods.
Models... (maybe fits better in another thread)
Did a bit of digging, and thought that some might be interested in some of the underlying structures of the statistical metrics.
First, for those who are a bit more "geeky", the parameters for the OBPM metric are determined from long-term data (to provide some statistical stability) and are not based on regression models but rather through various machine learning techniques. So there is a bit of a "black box" in terms of how the weights for the parameters are derived. Below is a summary of the weights used for OBPM metrics. You can find an article here (https://hockey-graphs.com/2016/10/26/introducing-box-plus-minus/).
These parameter weights might help interpret how different players are ranked. Some obvious features of the parameter mix:
1) Assists are weighted more heavily than goals - For defensemen (black bars) at both even strength and on the power play, a 2nd assist is weighted more positively than a 1st assist, which is weighted more heavily than a goal. For forwards (blue bars), at even strength 1st assists are weighted the most, followed by goals and second assists. On the power play, goals have very little value and 2nd assists are weighted more than 1st assists.
2) Quality of teammates matters a lot at even strength - This means that your OBPM is boosted if you are playing with high quality teammates, and downgraded if you are playing with low quality of teammates, especially if you are a forward.
3) Individual metrics don't matter much, except individual expected goals on the PP - This means that what an individual player's performance in terms of shot metrics doesn't really have much weight, except in terms of shots and shot quality on the PP.
Another key aspect of the OBPM metric is that it is tethered to a "prior distribution", which is formed by a player's performance in the prior season. This is done to stabilize metrics that vacillate excessively. So, for example, if a player had poor production in the season prior, that "tethers" his metrics for the current season, assuming that he is likely to produce like last season. Conversely, if a player had a good season the year prior, that will tend to bring up the metrics for this season.
These parameter weights help to explain comparisons between players, which might otherwise seem a bit puzzling. For example:
- Lowry's higher G60 (goals per 60 minutes) than Perreault's are more than offset because goals are worth less than assists in the OBPM model. Lowry's PP goals are basically worthless in the model, whereas Perreault's assists and shot attempts (from the half wall) are valued at more than twice the rate of goals.
- Perreault's OBPM is raised because he plays more with higher quality teammates (TOI.QoT), whereas Lowry's OBPM is depressed because he plays with lower quality teammates. So, the metric doesn't actually adjust for quality of teammates per se, but rather adjusts your OBPM upwards or downwards depending on whether you play with good teammates. If you only play with lower quality teammates, as Lowry does, your OBPM is reduced. Interestingly, the OBPM doesn't appear to adjust for poor linemate quality per se, so that if a player like Perreault played continuously with lower quality teammates and produced at a lower rate (goals and assists), his OBPM would be reduced.
This is based on a rather quick read of the methodology, and I expect that Garret can correct me if I've misinterpreted things.
However, based on my understanding of the OBPM model I think it's understandable why Lowry has such a poor metric. His greater production of goals instead of assists (especially on the PP) actually downgrades his OBPM, and the coaches decision to play him with poor linemates further pulls down his score, rather than adjusting his score.
Did a bit of digging, and thought that some might be interested in some of the underlying structures of the statistical metrics.
First, for those who are a bit more "geeky", the parameters for the OBPM metric are determined from long-term data (to provide some statistical stability) and are not based on regression models but rather through various machine learning techniques. So there is a bit of a "black box" in terms of how the weights for the parameters are derived. Below is a summary of the weights used for OBPM metrics. You can find an article here (https://hockey-graphs.com/2016/10/26/introducing-box-plus-minus/).
These parameter weights might help interpret how different players are ranked. Some obvious features of the parameter mix:
1) Assists are weighted more heavily than goals - For defensemen (black bars) at both even strength and on the power play, a 2nd assist is weighted more positively than a 1st assist, which is weighted more heavily than a goal. For forwards (blue bars), at even strength 1st assists are weighted the most, followed by goals and second assists. On the power play, goals have very little value and 2nd assists are weighted more than 1st assists.
2) Quality of teammates matters a lot at even strength - This means that your OBPM is boosted if you are playing with high quality teammates, and downgraded if you are playing with low quality of teammates, especially if you are a forward.
3) Individual metrics don't matter much, except individual expected goals on the PP - This means that what an individual player's performance in terms of shot metrics doesn't really have much weight, except in terms of shots and shot quality on the PP.
Another key aspect of the OBPM metric is that it is tethered to a "prior distribution", which is formed by a player's performance in the prior season. This is done to stabilize metrics that vacillate excessively. So, for example, if a player had poor production in the season prior, that "tethers" his metrics for the current season, assuming that he is likely to produce like last season. Conversely, if a player had a good season the year prior, that will tend to bring up the metrics for this season.
These parameter weights help to explain comparisons between players, which might otherwise seem a bit puzzling. For example:
- Lowry's higher G60 (goals per 60 minutes) than Perreault's are more than offset because goals are worth less than assists in the OBPM model. Lowry's PP goals are basically worthless in the model, whereas Perreault's assists and shot attempts (from the half wall) are valued at more than twice the rate of goals.
- Perreault's OBPM is raised because he plays more with higher quality teammates (TOI.QoT), whereas Lowry's OBPM is depressed because he plays with lower quality teammates. So, the metric doesn't actually adjust for quality of teammates per se, but rather adjusts your OBPM upwards or downwards depending on whether you play with good teammates. If you only play with lower quality teammates, as Lowry does, your OBPM is reduced. Interestingly, the OBPM doesn't appear to adjust for poor linemate quality per se, so that if a player like Perreault played continuously with lower quality teammates and produced at a lower rate (goals and assists), his OBPM would be reduced.
This is based on a rather quick read of the methodology, and I expect that Garret can correct me if I've misinterpreted things.
However, based on my understanding of the OBPM model I think it's understandable why Lowry has such a poor metric. His greater production of goals instead of assists (especially on the PP) actually downgrades his OBPM, and the coaches decision to play him with poor linemates further pulls down his score, rather than adjusting his score.
...another thing I learned
Chiarot was a defensive stalwart in 2015/16. No wonder the Jets penciled him into the top-6 for this season.
Chiarot was a defensive stalwart in 2015/16. No wonder the Jets penciled him into the top-6 for this season.
Chiarot was a defensive stalwart in 2015/16. No wonder the Jets penciled him into the top-6 for this season.
His PK numbers are pretty respectable this year. He needs a guy next to him that can make the tough passes.
If we play Chiarot with Buff and Entrom with Myers (instead of vice versa) we will be in good shape.
Can't see our team making a big free agent signing on D. You never know I guess, but yeah.
To be clear, I actually think Chiarot is pretty terrible, mostly because of his really bad decisions with the puck.
I was just observing that he did seem to measure out well in one metric (DBPM), which is based on hits for and against, blocks, and give-aways and take-aways (which are poorly measured).
Holden Caulfield
He's guilty
Those stats are all very poor stats as you mention. There's a reason most of the players on that list have trouble staying in the lineup with their respective teams. No shocker that multiple teams have multiple players, it's because they track the stats differently from other rinks.
I thought this was interesting:
You can sift through the Tweets explaining the Star rating but it's basically what defensemen get sent out to face the best players on the opposition teams. Enstrom at 20. Postma dead last (and Chiarot not much ahead of him in the "Bottom 20") of about 200-ish players.
- Status
Ad
Upcoming events
-
-
-
BET ON ONLY ONE HORSE WATCH LIVE - Jacques Cartier Stakes (Gr. 3) - WOODBINE RACE 5 - TorontoWagers: 1Staked: $100.00Event closes- Updated:
-
BET ON ONLY ONE HORSE WATCH LIVE - Belle Mahone Stakes (Gr. 3) - WOODBINE RACE 7 - TorontoWagers: 1Staked: $100.00Event closes
- Updated:
-