I'm still impressed that he refuses to accept that "is a player in front of the goalie" isn't a black&white issue. Is that really SO hard to understand?
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
NHL Off-Season 2014 |OT2| - Dan Cleary
- Thread starter Marvie_3
- Start date
I'm still impressed that he refuses to accept that "is a player in front of the goalie" isn't a black&white issue. Is that really SO hard to understand?
The only ones being obtuse is you guys.
I don't know how more clear i can be on this matter
It is when it undermines DopeyStats.
I'm gonna assume that this specific DopeyStat can just be summed up by the weight of a player. Much easier stat to track.f
It's quite black and white. Is there a player, from the shooters team.. that is instructed/standing in front of the goaltender with the intention of obstructing the goaltenders view, interfering with the goaltenders ability to save the puck or possibly wanting to commit suicide by rubber pelting as long as he is directly in front of the goaltender... no more than 3 feet away.
Players skating by don't count. Players staying slightly to the left or right (tip seekers) do not count. (as goaltenders have clear sight in those instances) Players that are laying on the ice do not count.
All we are doing is establishing general conditions for the shot, the moment the shot is taken so when we look at all the shots we can establish a shooting percentage across every one of those shots with similar conditions. They don't need to be exact or precise, just close. Tips and rebounds will be considered as separate events. Shots that miss the net or hit players don't count toward the shooting percentage, but are tallied under attempt on net quality using the shooting % of pucks that hit target.
SmackCrackleNPop
Member
I know who I'm not inviting to Selanne Steak Tavern, lol
The picture he drew was black and white.
Dubbedinenglish
Banned
I've ground up some Jesus and will be liberally applying it to all y'all of the last page.
wsoxfan1214
Banned
Jesus fucking christ dopey
Great article on the incoming analytical explosion regarding the NHL:
http://www.bostonglobe.com/sports/2...B1N81iptBsR1NQkslbGM/story.html?event=event25
http://www.bostonglobe.com/sports/2...B1N81iptBsR1NQkslbGM/story.html?event=event25
CrazedArabMan
Member
I don't even know what goes on in this thread anymore. However, is there a corsi rating for intangibles? I have a feeling Cleary would be the highest rated if they do.
:lol
This last page has been thoroughly entertaining. This is like the third time in the past week you guys manage to amuse me in this time of horrible hockey withdrawal.
That's a fucking impressive/useful stat, IMO.
Also, I have a friend who's exactly like Dopey. He doesn't give a shit about hockey but he could definitely assist with the believing his opinion is better/the only one and you're all morons for not seeing shit the same way part. Not that I'm saying I don't fully agree with Dopey. Or that I do. I'm not riding this train either which way. Please keep arguing, but remember...
We're not always safe even inside the bubble =(
That's a fucking impressive/useful stat, IMO.
Also, I have a friend who's exactly like Dopey. He doesn't give a shit about hockey but he could definitely assist with the believing his opinion is better/the only one and you're all morons for not seeing shit the same way part. Not that I'm saying I don't fully agree with Dopey. Or that I do. I'm not riding this train either which way. Please keep arguing, but remember...
We're not always safe even inside the bubble =(
Calamari41
41 > 38
I'm with Dopey. I am fully willing to jump on board with him and ride this innovative new analysis method straight into...
Down with Dopey!
Down with Dopey!
Yeah its called Clearsi
Bergevin got humbled, probably in a bar somewhere telling about the time he signed Patches for a bargain,
:lol
How does that feel like?
So maybe I didn't read through your stuff all thoroughly dopey but are you boiling screens down to a set distance, like 3 feet from the goalie? What about a screened shot from 7 feet out, a harmless looking player drifting/standing in a shooting lane can cause more harm then that close proximity screen, I've had easy saves on close screens and difficult saves on distanced screens, I mean at 6'7 at least on skates I see over some of those screens easily.
The fact Molson had to step in, is in itself uncomfortable.
Kuro Madoushi
Unconfirmed Member
Lol wtf
:lol :lol
Thank the Anti-Alfie that GMs can't be traded because Melnyk would definitely be on the phone with Montreal.
Simon Belmont
Banned
Jesus Christ dopey, you lost me at the point where you said it's completely impossible to rig stats. I hope you never actually took a stats / experimental design class, because if you did your institution should be taking their credit back.
It's far too difficult to make an assessment of a player screening from further out. So intead of judging from what a goalie can see, i'm just going to use the act of putting a player in front. Some shots may look like that and may originally inflate or deflate the data but after i get more and more data, it will be essentially noise. Or at least it will balance out.
One full season is going to be more than 50,000 shots. (100,000 attempts)
So the idea is when i am choosing what path my data is going to be using, i am going to be consistent
That's why i say it's not going to be perfect and there will be some anomalies, but the goal is to get the league average success rate for that shot.... and when the data is filled out... instead of being 1 shot... it will be a shot of % percentage.
The idea is... use current corsi/fenwick as a backbone and make offensive output differential against time on ice. Shot quality for actual offense (corsi), attempted shot quality for potential (fenwick)... this way we get ratings for defense instead of just offense as they are weighted figures. Then it makes the QoC information a lot more sensible. QoT is still going to be muddy as players like Bozak will still be inflated...
Basically I am hoping that when you add all shot qualities together and compare to actual goals scored... we should see both figures in the same ballpark. Track this across a season and we can see if the data is faulty or not.
This will get use
What was Bergevin thinking. Then moneybags steps in to save the day and very likely gives PK a bigger contract than would have been necessary.
CrazedArabMan
Member
Best rating system ever!
wsoxfan1214
Banned
Uh... Huh.
KernelPanic
Member
Corsi is a proxy stat for possession. All the articles I've read indicate that Corsi tracks possession time very closely. It's not exact but it's a fairly good indicator. Go ahead and track possession time, you'll find it comes awful close to the Corsi.
Over a large sample size yes it does you are more likely to win.
If a team is winning games despite being outshot every game you are winning on luck(ie. the Leafs). Most likely your PDO is above the norm. That is unsustainable. Sooner or later you will come crashing down. This is why some teams can make the playoffs in a lockout shortened season but implode over a full season.
I'm not saying Corsi is a perfect stat, there's a lot of context and other things that need to be considered which is why Fenwick and Close stats exist etc. A hockey game is full of "noise" and random events. Goals can amazingly deflect in off a player's ass, a goalie can make a miraculous save that he would never be able to replicate. Such events don't really mean shit in the grand scheme of things.
Hockey analytics is in its infancy, but so far teams with better puck possession tend to do better than those that are worse. And Corsi has been a pretty good indicator of that so far, but obviously there is more work being done in this area.
alexthekid
Member
I certainly think that a system that tries to value the quality of the shots is a very interesting thought. As soc has been saying it is indeed going to need a lots and lots of subjective decisions while recording the stat. However I dont think that it makes it useful per default. Even to it is subjective if the vision of the goalie is obstructed or not, I think that if you go into the stat collecting with an open mind (this might be hard when you start to recognize trends for some teams and maybe start reinforcing them yourself) i think that it is possible to separate shots with an obstructed view, from the others, on a good enough basis to actually make use it as a basis when valuing the quality of the shot, the same can be said for:
Direct shot or not.
Does the puck pass the centreline (line from goal to goal, couldnt come up with an english phrase) on the pass to the direct shot.
Is the player allowed to cross the centre line when making an attempt on the goal
It is really hard to be sure that it will actually add something of value, but if you can come up with theory that is reasonable enough and you at least try to be as honest as you can when recording the stats. You might be able to find something useful in your system.
However there are other problems that are way way bigger than having to come up with somewhat subjective theories. You are going to need a great number of statcollecting, something that I doubt you will be able to do on your own. If you based it on some stat that you could collect from other sources, then maybe, if you spent a lot of time, maybe you could evaluate your system and if it seems to work, try to use it for something beneficiary. But trying to evaluate a system based on stats that you have to get while watching games, seems implausible. The amount of situations you need to be sure that it is actually your variable that is decisive seems absurd.
Basically I think that a system that tries to value shot quality might be possible to find, and without a doubt an interesting investigation but it seems way to impractical to do while doing the stat collecting yourself. But I wish you the best of luck!
Direct shot or not.
Does the puck pass the centreline (line from goal to goal, couldnt come up with an english phrase) on the pass to the direct shot.
Is the player allowed to cross the centre line when making an attempt on the goal
It is really hard to be sure that it will actually add something of value, but if you can come up with theory that is reasonable enough and you at least try to be as honest as you can when recording the stats. You might be able to find something useful in your system.
However there are other problems that are way way bigger than having to come up with somewhat subjective theories. You are going to need a great number of statcollecting, something that I doubt you will be able to do on your own. If you based it on some stat that you could collect from other sources, then maybe, if you spent a lot of time, maybe you could evaluate your system and if it seems to work, try to use it for something beneficiary. But trying to evaluate a system based on stats that you have to get while watching games, seems implausible. The amount of situations you need to be sure that it is actually your variable that is decisive seems absurd.
Basically I think that a system that tries to value shot quality might be possible to find, and without a doubt an interesting investigation but it seems way to impractical to do while doing the stat collecting yourself. But I wish you the best of luck!
Lone_Prodigy
Member
If we ever make it to OT3 I nominate this as the thread title.
I glaze over reading these DopeyStats arguments, but I just wanted to say that I agree 100% with the part of the advanced stat community that thinks "shot quality" is impossible to measure and, more importantly, so close over time that stats on it are an illusion. Shots are shots, and other than separating SoG from attempts (blocks, deflections etc) attempts to proclaim some SoG as better than others isn't possible or even helpful.
CrazedArabMan
Member
I want another beautiful picture of our lord and savior (Cleary) as the only thing in the OP. Screw it, that should just be the title for the official first month of the season.
KernelPanic
Member
I think Dopey has the right idea, I just think it's just too subjective and impractical to pull off. Closest thing I heard is tracking shots from the slot where I believe statistically most goals are scored. That makes it less subjective(it's either in the slot or it's not).
Even then I suspect that teams that take the highest quality shots(if it were possible to determine this objectively) probably take the largest amount of shot attempts. So for the purpose of trends you are not gaining much.
Shot attempts are used for Corsi rather than SOG because more of them occur in a game. Relatively very few quality shots occur in a game so your sample size would be small making it very difficult to draw any conclusions.
Hate to break it to ya Dopey but I remember reading an article where teams are already adjusting SV% for shot quality so it sounds like internal analytics consultants/staff are already creating whatever miracle metric for shot quality you are thinking of.
And they definitely have more:
1) Data
2) Resources
3) Time
then you do :/
And they definitely have more:
1) Data
2) Resources
3) Time
then you do :/
bonesmccoy
Banned
Isn't there some new camera tech that will make it much easier for the NHL to track 'new' stats?
KernelPanic
Member
I'm willing to bet each team has it's own definition of shot quality. I guess it could work for them as they'd have the same person/people doing it each game.
CrazedArabMan
Member
So, new CoD MP looks actually good, may pick it up down the line on sale as my MP for this year will be Halo for competitive and Destiny for co-op.
Definitely teams aren't sharing anything and I'm willing to bet they make sure their consultants aren't allowed to share data/ideas with consultants employed by other companies.
I think that will ultimately slow down the progress of developing better stats.
CrazedArabMan
Member
99% of stats are stupid. All hale the Clearsi!
It's not really subjective or impractical. All that is being done is setting the conditions in general form.
It's a shooting percentage with parameters.
Quality of shot is dictated by the shots that have been sampled, so every shot has an assigned quality. A shot from 5 feet away with the goalie out of position is going to have a ridiculously high shot quality as opposed to a slap shot 40 feet away with no obstruction and goalie in position but in our current stats world a shot is a shot is a shot. But it really isn't. The offensive zone is broken down into multiple zones. Walls, point, upper slot, lower slot, directly in front of the net.
The old NHL shot quality was derived from shot distance, which doesn't help with angles... (they don't even track distance right anyways) which is far more important to a shots quality than anything. Then we take goalie position into account... if he's ready for shot and square to shooter, if he's transitioning (either because a quick pass or player shifting around the ozone) or out of position which means he has to scramble for a save when the shot is made. Then the type of shot (slap or wrist/snap)... then other variables like the player in front of the net, one timer...
So every game we have typically 40-60 shots to derive data from. But I've been saying end of season is when I can get preliminary data. Realistically i need about 3 seasons worth of data, about 150,000 shots or so...
Remember... im looking for league average. So if there's some bum like clarkson who is scoring a ton of garbage goals, stats should dictate he is burying a lot of these goals because of circumstance, not because of talent. His shot quality from NJD should be very high, but i suspect even with that i predict he'd have a negative variance (gf - tSONQ (total shot on net quality)) but a player like kessel would have a positive variance.
A team that is pouring tons of shots may have negative variance, or they may have positive variance depending on if the pucks go in. It should actually reinforce a lot of assumptions but bring more questions to the table.
Point is... it's to establish that a shot with similar conditions from similar position has an x chance to score based on previous outcomes from every type of player. It's to lower the value of perimeter, easy saves and increase the value of hard to stop shots... which in turn helps detail the cost of defensive screwups (inability to contain)
With the zone breakdown, we pull back the variables and look at team vs team. If sens pour more shots from one area of the ice with x average sonq or aonq but boston allows less shots with a lower sonq from those zones, it would dictate sens offense having a more difficult time scoring based on previous shots and how both teams make shots and attempts.
It's not just an analysis of shot quality, but an analysis of systems. It's an analysis of defense. It's an analysis of offense. Shot quality brings context to everything and is the biggest missing link to all hockey stats.
Important note, because my first post was short and vague: I don't think anyone argues that shot quality doesn't exist - obviously a tap-in to an empty net is different from a dump in from centre ice. But attempts to *measure* shot quality end up being less useful than you'd think, because of all the reasons mentioned already (how to define all the elements, how to gauge one shot from a nearly identical one etc) and a growing suspicion that shot quality for and against is pretty random. A few years ago the advanced stats crowd was going nuts for shot quality but after a lot of attempts I don't think anyone was able to come up with a system that showed a meaningful difference from team to team never mind player to player. And if there's no real difference (ie a difference beyond noise/luck) in a stat between good teams and bad then it's not worth tracking.
The other aspect of shot quality that has gone out of favour is the notion that good defensive teams force the other team to take 'bad' shots from the outside. While you can see a connection between good defensive teams and puck possession there isn't one with shot quality, which is to say that good defensive teams give up high quality shots in a seemingly random fashion, or at least the numbers between good and bad quality shots against are so close it may as well be random That's another knock against shot quality as a useful metric.
The other aspect of shot quality that has gone out of favour is the notion that good defensive teams force the other team to take 'bad' shots from the outside. While you can see a connection between good defensive teams and puck possession there isn't one with shot quality, which is to say that good defensive teams give up high quality shots in a seemingly random fashion, or at least the numbers between good and bad quality shots against are so close it may as well be random That's another knock against shot quality as a useful metric.