The Advanced Stats Thread Episode V: Rick Nash Camera Stares/60

[Machinehead]

These are the kind of bad interpretations that lead people to fall back on the "you've never played the game" argument.

You're assuming that Girardi caused the play to progress in such a way that led him to have to block a shot. That very well might be the case, but there are five other players on the ice (including one between the pipes who is notoriously bad at playing the puck) who could be at fault as well.

If you had concrete data that showed me that most of Girardi's blocks follow his own turnover and not one from his teammate, or a bad bounce, or a lost face off (you see where I'm going with this), I'd be interested.

I'd also be interested in seeing data about attempted/missed blocks. The other team will have the puck. The other team will shoot the puck. A defenseman' s job is to prevent those things from happening as much as practicable, and blocking shots is a good way to do that.

Yeah, I mean, if only we had ten years of evidence which shows that Girardi hurts the shot metrics of every teammate he's ever played with. Something like that would be really cool....

 

[Synergy27]

So lets assume you're right and Silverfish never played the game because he doesn't think blocked shots is a particularly great characteristic for a defenseman to have as his main attribute. Lets further assume that you're correct Girardi is covering for everyone's gaffs, especially Hank's, with excellent shot blocking ability.

Now that we're assuming both things are true isn't the following also true?

Good defense is also keeping the puck inside the other zone, having good gap control to prevent easy access to the neutral and defensive zones, preventing zone entries, separating the opponent from the puck and once getting the puck moving effectively by making a good first pass.

Well Girardi isn't good and or reliable at any of those things and the advanced numbers consistently reflect that.

Questioning blocked shots doesn't make me question whether another poster played the game or not because it really shouldn't matter. Regardless, there's a lot more to playing good defense than blocked shots.

The statement was that blocked shots occur because you put yourself in a position to have to block them. That's not (always) the case. I think discounting blocked shots as a stat is not a good idea. It's a very useful skill.

 

[Synergy27]

Yeah, I mean, if only we had ten years of evidence which shows that Girardi hurts the shot metrics of every teammate he's ever played with. Something like that would be really cool....

I'm arguing against the attempt to discount blocked shots as a valuable statistic, not against Girardi's decline.

 

[silverfish]

Jesus man had did you respond that fast?

I mean you can tell me a bit about how it's done if it's not that complicated. Kind of curious now that you mention it.

I'm on the computer

You'd need to download the context and relative files from Corsica.Hockey. Then, it's just matching up the two data pulls by player; relative zone starts and qual comp. From there, it's just a scatter plot.

These aren't in alphabetical order at all. I also removed everyone who played for two different teams.
http://imgur.com/a/Lg7yG

Under different circumstances, I'd make these prettier. Like, establish a TOI minimum that is higher than 50 so players who barely played don't mess up the axis. Or add a TOI variable, or a relSA% variable. In reality this is like two more lines of code, but Imgur is being very annoying and I don't want to go through the process of uploading 30 images again

Here's the R code:

library(ggplot2)
library(dplyr)
library(ggrepel)


relative <- read.csv("Corsica_Skater.Stats_19h50.csv")
context <- read.csv("Corsica_Skater.Stats_19h50 (1).csv")

vars1 <- c("Player", "Team", "Rel.ZSR")
vars2 <- c("Player", "xGF.QoC")

rel <- relative[vars1]
con <- context[vars2]

merge <- dplyr::left_join(rel, con, by = "Player")

merge$TeamClean <- substr(merge$Team, 1, 3)

teams <- unique(merge$TeamClean)

for(team in teams){
  plot <- subset(merge, merge$Team == team)
  
  plot2 <- ggplot(plot, aes(Rel.ZSR, xGF.QoC, label=Player)) + geom_label_repel() +
    labs(title = paste(team, "5v5 Usage"),
         x = "Relative Zone Starts",
         y = "xGF Competition",
         caption = "Data via Corsica.Hockey")
  print(plot2)
  ggsave(plot2, file = paste0(team, "_usage.png"))
}

 

[aufheben]

Oh this involves coding? Pass. Lol

 

[silverfish]

Oh this involves coding? Pass. Lol

Well, only the way I did it. Could just as easily be done in Excel.

That imgur link has zone starts and competition though. Not sure if you wanted something else?

 

[aufheben]

Well, only the way I did it. Could just as easily be done in Excel.

That imgur link has zone starts and competition though. Not sure if you wanted something else?

I just wanted something like what War-on-Ice had. Up-to-date player usage, regular season/playoff separation, variety of filters.

 

[aufheben]

Anyone have Micah Blake McCurdy's 2015-16 Playoff Predictions graph? hockeyviz @ineffectivemath

Edit: nvm found it

 

[Ola]

These are the kind of bad interpretations that lead people to fall back on the "you've never played the game" argument.

You're assuming that Girardi caused the play to progress in such a way that led him to have to block a shot. That very well might be the case, but there are five other players on the ice (including one between the pipes who is notoriously bad at playing the puck) who could be at fault as well.

If you had concrete data that showed me that most of Girardi's blocks follow his own turnover and not one from his teammate, or a bad bounce, or a lost face off (you see where I'm going with this), I'd be interested.

I'd also be interested in seeing data about attempted/missed blocks. The other team will have the puck. The other team will shoot the puck. A defenseman' s job is to prevent those things from happening as much as practicable, and blocking shots is a good way to do that.

Fair points, then someone might argue, why then are Girardi's teammates always screwing up when he is one the ice and not to the same extent when other teammates are one the ice. Which also is a fair point.

But in addition to this, a very big factor is the overhead creature that is a hockey game.

I do not have data to prove this -- but I can bet my right arm, house and family -- on that in terms of momentum a shift in hockey, after say a line change during a stoppage and that ends with another stoppage in play, does not start when the puck is dropped and stops when the ref call the whistle.

What I am sure the data would show is:
-If one team has taken a shot, this team is more likely to take the next shot even in light of a in between line change.
-Offense comes in bundles. It is not like the shots coming in sequences like this:
MTL (shot by MTL)-NYR (shot by NYR)-MTL-NYR-MTL-NYR-MTL-NYR-MTL-NYR-MTL-NYR-MTL-NYR-MTL-NYR-MTL-NYR-MTL-NYR.
Its much more like:
MTL-MTL-MTL-MTL-NYR-MTL-MTL-MTL-MTL-NYR-MTL-NYR-NYR-NYR-MTL-NYR-NYR-NYR-MTL-NYR-MTL-MTL-MTL-NYR-NYR-NYR

AND line changes in between these periods of momentum for either team does NOT have a digital affect. It without any single doubt carries over. It is not like Radulov's line putting 5 shots on net one shift, the next Lindberg's line puts 4 shots, and so forth. For several reasons. One being mental. But also many practical reasons. You get out of synch. You scramble to get all players off the ice, and when the next line is going some have been on 45 seconds when others been on 15, and you just dump it in and change again to get in phase again. You do not have the puck when you hit the ice. Its more pressure, after three horrible shifts, and a D gets the puck, he better not fumble it -- so it opts for the safe lift it out play instead of a small percentage play. And so forth, and so forth, and so forth and so forth.

Now to my point, during these swings in momentum during a game -- when does Beuke/AV put certain D pairings on the ice? Its definitely NOT "fair" in the sense that everyone get the same shifts that are tough and everyone get the same shifts where we are heading in the right direction. Skjei, Yandle before that, was thrown on the ice when we got going. McD-Girardi was thrown on the ice when we were in trouble.

If you look at like G2 in MTL, McD had a CF% of 29% and Skjei-Smith was north of 50%. Is this a "fair" reflection of McDs play in relation to Skjei-Smith? Of course not. Watch that game again. McD is double shifted during times we can't get out of our own end and so forth.

Executive Summary: So does this exonerate Girardi from having worthless stats? No, Girardi has worthless stats because he just throws the puck all over the place when he gets it. At least pretty often. Its that Torts style. But since he always gets the toughest shifts when he is dressed, that also has a cumulative effect. No doubt. And you cannot like compare Clendenning's numbers with Girardi's straight off.

And of course "we" in the metric community is aware of this. Hence why so many attempts are made to adjust for all kind of factors. Its easy to see that a number like CF% does not tell the full story, right? Everyone agree on that. But has these attempts to adjust for outside noise been effective on an individual level? Can you "scout" players by looking at CF% only? Nope, not even remotely to be honest.

 

[silverfish]

I just wanted something like what War-on-Ice had. Up-to-date player usage, regular season/playoff separation, variety of filters.

Yeah you're talking about a full-on website which is absolutely much less simple

 

[Off Sides]

Because Nash had a lot of good "loud" moments. Taking the puck to the net. Big goals/assists. It's a lot easier to remember those plays, and less the shifts where the Nash line was pinned in their own zone due to small errors here or there.

Still, we're talking about a six game sample size. The data we're playing with in the playoffs isn't the same as the data we get to play with in the regular season. This is what makes #fancystats in the NWHL so difficult to derive true insights from. Only 18 games

So Nash was not good?

I think I understand what you are saying there, but are the playoffs in the NHL not a short sample size, that should also probably be treated a little different than the regular season sample given it's series format with what I think is a different level of officiating?

Back to Nash, I really don't know, like if he does not score those goals, do the Rangers still win those games?

If Girardi does not make that block...


I guess it leads me back to thinking that there are certain game situations where it is far more important to make that specific play than it is to make the play that would result in better stats more often.

Like the weighting of what should be considered really important toward the outcome of that game, for the individual skater may just come down to one or two important opportunities per game for that player to make that play. The other stuff even if it happens way more often, and it's a negative, it is possible much of it is and should be treated as way less important?

 

[silverfish]

So Nash was not good?

I think I understand what you are saying there, but are the playoffs in the NHL not a short sample size, that should also probably be treated a little different than the regular season sample given it's series format with what I think is a different level of officiating?

Back to Nash, I really don't know, like if he does not score those goals, do the Rangers still win those games?

If Girardi does not make that block...


I guess it leads me back to thinking that there are certain game situations where it is far more important to make that specific play than it is to make the play that would result in better stats more often.

Like the weighting of what should be considered really important toward the outcome of that game, for the individual skater may just come down to one or two important opportunities per game for that player to make that play. The other stuff even if it happens way more often, and it's a negative, it is possible much of it is and should be treated as way less important?

I think Nash was good. He had that one brutal turnover in the OT of game 2 that led to the game winner, but, **** like that is going to happen when you try and make plays.

If you can off-set your bad shot attempt metrics by being on the ice when the puck goes in the other team's net... well, the only issue with that is the argument of sustainability. But, in a short playoff series, the effects of shot-attempts aren't as highlighted as they are in the regular season because of the small sample size of games. Even if you get decimated in some games by shot attempts, it's still only a max of 7 games, and if you have a goalie capable of stealing you a series (hi, Hank), then shot attempt metrics mean even less.

When someone argues for the use of shot attempts, they aren't looking at the game that just happened, they are looking at the game that's going to happen tomorrow.

Think about it this way... In their 6 games, the Rangers had a 5v5 SA% of 47.33% vs Ottawa's 51.37%. If we throw quality in the mix using xGF%, the Rangers are at 47% while Ottawa is at 49.97%. We can surmise, that if these numbers hold, that Ottawa has a slightly better chance at winning the series over NYR because they'll likely have more chances and more attempts.

Then again, we have Hank operating at a .9565 sv% right now during 5v5 play. That alone offsets everything shot attempts can tell us. Is that a sustainable save percentage for Hank?

Considering the amount of time in the series against Montreal that was spent tied or +/- 1 goal, if Hank instead is playing at his regular season form, where are the Rangers right now? Not winning that series. We would've spent yesterday reading breakup day quotes and arguing about AV's future.

 

[Off Sides]

I think Nash was good. He had that one brutal turnover in the OT of game 2 that led to the game winner, but, **** like that is going to happen when you try and make plays.

If you can off-set your bad shot attempt metrics by being on the ice when the puck goes in the other team's net... well, the only issue with that is the argument of sustainability. But, in a short playoff series, the effects of shot-attempts aren't as highlighted as they are in the regular season because of the small sample size of games. Even if you get decimated in some games by shot attempts, it's still only a max of 7 games, and if you have a goalie capable of stealing you a series (hi, Hank), then shot attempt metrics mean even less.

When someone argues for the use of shot attempts, they aren't looking at the game that just happened, they are looking at the game that's going to happen tomorrow.

Think about it this way... In their 6 games, the Rangers had a 5v5 SA% of 47.33% vs Ottawa's 51.37%. If we throw quality in the mix using xGF%, the Rangers are at 47% while Ottawa is at 49.97%. We can surmise, that if these numbers hold, that Ottawa has a slightly better chance at winning the series over NYR because they'll likely have more chances and more attempts.

Then again, we have Hank operating at a .9565 sv% right now during 5v5 play. That alone offsets everything shot attempts can tell us. Is that a sustainable save percentage for Hank?

Considering the amount of time in the series against Montreal that was spent tied or +/- 1 goal, if Hank instead is playing at his regular season form, where are the Rangers right now? Not winning that series. We would've spent yesterday reading breakup day quotes and arguing about AV's future.

Concerning the bolded, would shooting percentage also have to be considered?

I mean regular season Habs had 30 shot per game on goal, Rangers 29.7, yet the Rangers had 3.09 goal per game to the Habs 2.72

So could I not perhaps think the Rangers had better scorers and regardless of the shot attempt negative differential, the Rangers had a better chance at a goal for event happening?

I mean I agree about save percentage too, but if the Rangers should win the shooting % battle, they should likely win the save % one too, regardless of the SA differential?

 

[Trxjw]

I just wanted something like what War-on-Ice had. Up-to-date player usage, regular season/playoff separation, variety of filters.

You sound like every web client I've ever had.

 

[silverfish]

Concerning the bolded, would shooting percentage also have to be considered?

I mean regular season Habs had 30 shot per game on goal, Rangers 29.7, yet the Rangers had 3.09 goal per game to the Habs 2.72

So could I not perhaps think the Rangers had better scorers and regardless of the shot attempt negative differential, the Rangers had a better chance at a goal for event happening?

I mean I agree about save percentage too, but if the Rangers should win the shooting % battle, they should likely win the save % one too, regardless of the SA differential?

Well, yes, but this is the obvious hot take, right? The team with the better PDO (shooting % + save %) is going to win a single game every time.

The point is, you need to maximize your PDO by generating quality and quantity and limiting quality and quantity to the best of your ability.

 

[aufheben]

You sound like every web client I've ever had.

I just want it that exact way why can't I have it I don't understand!

I know man, I do graphic design. Saaaame thing haha.

 

[Off Sides]

Well, yes, but this is the obvious hot take, right? The team with the better PDO (shooting % + save %) is going to win a single game every time.

The point is, you need to maximize your PDO by generating quality and quantity and limiting quality and quantity to the best of your ability.

Right, I want the Rangers to have a super high PDO in the playoffs as long as it lasts all the way through them.

I don't think it's uncommon for the playoffs to have outliers. Some players/teams for whatever reason find a way to beat their career averages. Even long term the Rangers have been somewhat of an outlier compared to what their numbers say they should be.

Glass has 100% shooting percentage. Brassard, Williams, Couture.

When stuff like that happens I'm not so sure there are predictive powers to be found. In a way having a team capable of being an outlier, PDO, may be a good thing for the playoffs. Obviously not as good as having a team full of players who are just good all the time but perhaps it at least partially explains the difference between what the numbers would predict and who wins. Not sure I think of that as luck or not but whatever it takes.

 

[Synergy27]

But in addition to this, a very big factor is the overhead creature that is a hockey game.

Maybe I'm just naive, but I just can't get over the hump in feeling that this is consistently ignored.

I do not argue that Girardi is still a good defenseman, but to completely discount certain things (like blocks) that are by definition defensive (i.e., not offensive/positive) ignores the fact that there are two teams playing. You definitely do need guys that are good at doing things when they don't have the puck. Obviously you want to have the puck all the time, or at least guys that are good on both sides of it.

The former is obviously impossible. The latter, for whatever reason, seems to be unlikely due to supply and demand.

 

[silverfish]

Adapt or die.

EDIT - Explainer: This is basically a smoothed histogram that plots team CF% for each season since 07-08. You can see how in the last two seasons, the spread is thinning, and the middle is getting larger. Likely showing that teams are adapting to shot attempts as a major influence. This is why there were complaints about what David Johnson was trying to show with the Corsi ranked teams and how they are faring in the playoffs this year. The spread between teams is less than it's ever been.

 

[JimmyG89]

This would explain why shot quality is having a much greater impact this season than others. With more teams attempting roughly the same amount of shots, the team getting the higher quality chances is better off. The mid point team (50% SVA CF%) is Philly. They finished 18th.

Let's remove the top and bottom 5, as these are typically outliers in most research data. The difference in SVA CF% from Toronto to New Jersey is a little over 3%. That might be 4-5 shot attempts in a single game?

 

[Ola]

Well, yes, but this is the obvious hot take, right? The team with the better PDO (shooting % + save %) is going to win a single game every time.

The point is, you need to maximize your PDO by generating quality and quantity and limiting quality and quantity to the best of your ability.

Great point silverfish, I definitely agree.

And all games are different. Against Ottawa for example, I think it might be a bit dangerous to push as hard as we did against MTL. They will defend deep with 5 guys, it will be hard to score goals even if you get long shifts. Against them, it will be really important to play gutsy offensive hockey and really go for it instantly, and never be satisfied to just get possession. Because then you will end up in those situations when you put a lot of rubber on net, and Ottawa gets to play their sneak offensive game.

If you on the other hand like play Pittsburgh, Toronto as from this season, and speedy dangerous teams like that, you don't want to exchange chances too much with them. Same with CBJ, they will hammer away at you if you give them a shot at it. Against these teams I would like us to be a little bit more careful with the puck, not gamble too much on small marginal passes, maybe turn up at times instead of trying to beat a D 1 on 1.

 

[Ola]

This would explain why shot quality is having a much greater impact this season than others. With more teams attempting roughly the same amount of shots, the team getting the higher quality chances is better off. The mid point team (50% SVA CF%) is Philly. They finished 18th.

Let's remove the top and bottom 5, as these are typically outliers in most research data. The difference in SVA CF% from Toronto to New Jersey is a little over 3%. That might be 4-5 shot attempts in a single game?

I do think a major factor here also is that this season, teams have really let go of the safety precautions when they attack. Like it was not long ago we talked about having a "3rd guy high", right? Well, how often in the game right now don't we see a team only have ONE guy back on defense? It's not even a "2nd guy high". 3 forwards + 1 D push very hard to the net off rushes, of the rush, but also when circling the puck.

I think like Silverfish pointed out, the key here is to be able create those high danger situations offensively. If you do shooting% will go up A LOT. And those teams have the most success.

And this is the BIG hangup I have with Shattenkirk for example. Shatty is great at operating -- outside -- a defensive box. This is great, in an environment where you don't allow your Ds to go on the inside. If you do that -- you of course would prefer Ds who are great at joining the rush etc. Jumping into the play. Not that Shatty wouldn't help us, but it just don't sit right for me that we go from 1995 to 2017 without having a decent RHS PPQB, and then when we go out and get one for 7 years it's someone who in terms of style ALREADY is out of date...

BTW, anyone got a graph for spread on shooting% the last years? Has it been affected?

 

[silverfish]

BTW, anyone got a graph for spread on shooting% the last years? Has it been affected?

Have the code on my work computer. I'll do this tomorrow if I remember.

 

[Cassano]

 

[Ola]

Have the code on my work computer. I'll do this tomorrow if I remember.

Thanks!