Ongoing Stats and Analytical Discussion Thread: Battle of the Defense

[devilsblood]

So PP shots should carry more weight right?

So I can see why it's good to seperate them from 5v5. I think logic on that should be pretty clear.

But we don't want to exclude all these other situations.

So maybe the equation should look like 5v5 corsi x PDO + special team's= likelihood of success. Keep them as seperate values, but have them both in the equation?

 

[Emperoreddy]

I just got to glance through it real quick but it appears he’s taking the end of the season and saying goal differential explains standings better than shot differential does. I don’t think he will find anyone arguing against that. It seemed a lot of writing to confirm what was common thought, not a bad thing.

But if I was examining teams halfway through a season, would I see teams with good goal differentials and bad shot differentials keep their good goal differential all season or would they start drifting towards what their bad shot differential suggests. I believe it’s the latter but I haven’t run any analysis.

I think for me, and I mentioned it last night is that some people (not saying it is happening here) really try to push the shot based stats as the end all be all, and an excellent indicator of future performance.

That isn’t to say that it is a totally junk stat and doesn’t have a role in gauging a team’s performance either.

I think that annoys people and caused them to rather just ignore them completely, especially because those people (again not saying you are doing that) can be quite smug with their predictions of what teams will crater and what teams are going to bounce back.

 

[Devils731]

Wrong.

He conducted a linear regression analysis.

What this means is that the results are exactly the opposite of what you're suggesting. Namely, that yes, the GD held up and was more accurate in predicting future success than the Corsi or other similar but tweaked Shot counting models.

But where was he getting the goal differential numbers from, the end of season totals? I didn’t see where he said he got them from so that was my assumption and why I was asking.

 

[Devils731]

I think for me, and I mentioned it last night is that some people (not saying it is happening here) really try to push the shot based stats as the end all be all, and an excellent indicator of future performance.

That isn’t to say that it is a totally junk stat and doesn’t have a role in gauging a team’s performance either.

I think that annoys people and caused them to rather just ignore them completely, especially because those people (again not saying you are doing that) can be quite smug with their predictions of what teams will crater and what teams are going to bounce back.

I guess I just don’t see anyone here being that pushy...except maybe some people on the anti side of things.

A nice rational discussion is fun but right now it’s not a discussion looking for best answers for some participants.

 

[BenedictGomez]

But where was he getting the goal differential numbers from, the end of season totals? I didn’t see where he said he got them from so that was my assumption and why I was asking.

You're starting your analysis with a completed season, so you already have the complete and final data.

Then, you can start your analysis at any point you like, at 10 games, at 20 games, at 40 games, etc... with the known data at that point, and use that data set to see how tightly it correlated with the known 82 game completed data set. More correlation good, less correlation bad.

 

[devilsblood]

Rink Stats | Why popular advanced stats are bad at predicting who wins games

Another one saying thr same thing.

The conclusion in this one seems to be that he feels the flaws in Corsi and Fenwick is relying on missed and blocked shots because by definition a miss or a blocked shot will not directly change the game since it can never end up a goal.

Meh, do I think a better scoring chance is a guy in close who rings one off the post, or a soft shooting point man putting one into the goalies chest? The former easy.

 

[Emperoreddy]

I guess I just don’t see anyone here being that pushy...except maybe some people on the anti side of things.

A nice rational discussion is fun but right now it’s not a discussion looking for best answers for some participants.

I was thinking about it from a practical standpoint. The big picture I don’t think you can hold a stat up high as a predictive model without it showing high levels of success over a huge sample size, something I have really not seen with either shot based stat.

Thinking about it on ice level I look to Severson as an example. I love the kid, but even I can admit his shot kind of sucks. It is soft and he misses the net a lot.

So Severson shoots the puck towards the net. It misses badly. His and the team’s corsi ticks up, but in the context of the game the missed shot has done nothing to boost his team’s chances of winning. By definition a missed shot can’t change the outcome of a game, it has a zero percent chance of becoming a goal because it missed. Blocked shots are the same.

Go deeper and actual shots are also counted the same, but we know on the ice all shots are not created equal. A soft shot from the blue line into the logo has a dramatically less chance of going in the net as opposed to a bullet in tight towards the corners.

We see whole games where this happens. Team peppers shots, but they are all in the low danger areas. They give up fewer shots but the shots they gave up were much higher quality and they lose the game.

And that is just it for me. If you want to use a stat to predict the future, it must be held to a much higher standard then a stat simply looking to help analysis past performance.

Maybe there is a formula that does it better and factors in the quality of shots, somehow accounts for special teams, and everything else. I’m just not seeing corsi on its own at evens as doing that.

 

[devilsblood]

I was thinking about it from a practical standpoint. The big picture I don’t think you can hold a stat up high as a predictive model without it showing high levels of success over a huge sample size, something I have really not seen with either shot based stat.

Thinking about it on ice level I look to Severson as an example. I love the kid, but even I can admit his shot kind of sucks. It is soft and he misses the net a lot.

So Severson shoots the puck towards the net. It misses badly. His and the team’s corsi ticks up, but in the context of the game the missed shot has done nothing to boost his team’s chances of winning. By definition a missed shot can’t change the outcome of a game, it has a zero percent chance of becoming a goal because it missed. Blocked shots are the same.

Go deeper and actual shots are also counted the same, but we know on the ice all shots are not created equal. A soft shot from the blue line into the logo has a dramatically less chance of going in the net as opposed to a bullet in tight towards the corners.

We see whole games where this happens. Team peppers shots, but they are all in the low danger areas. They give up fewer shots but the shots they gave up were much higher quality and they lose the game.

And that is just it for me. If you want to use a stat to predict the future, it must be held to a much higher standard then a stat simply looking to help analysis past performance.

Maybe there is a formula that does it better and factors in the quality of shots, somehow accounts for special teams, and everything else. I’m just not seeing corsi on its own at evens as doing that.

1)From a practical standpoint 5v5 corsi is a good predictor of success.

2)Sev's impressive corsi stats are really his on-ice #'s(best on team), not so much his individual #'s, 10th on team(icf/60). Although to your point his on ice scoring chances are not that great, and he has been getting some pretty easy minutes of late.

But all these different stats are worth looking at. Corsi looks to be a good stat, but not the only stat.

 

[Emperoreddy]

1)From a practical standpoint 5v5 corsi is a good predictor of success.

2)Sev's impressive corsi stats are really his on-ice #'s(best on team), not so much his individual #'s, 10th on team(icf/60). Although to your point his on ice scoring chances are not that great, and he has been getting some pretty easy minutes of late.

But all these different stats are worth looking at. Corsi looks to be a good stat, but not the only stat.

My whole issue with this started because I read an article where the author rather smugly stated the Devils would be a bottom 5 team even after the start based on corsi alone.

I don’t have a problem with using a bunch of stats to try and paint the picture of what is going on.

I think it was you that showed that play style factors into this as well as teams start switching to a more counter attack style.

 

[billingtons ghost]

What is really necessary is better, more descriptive stats.

Ideally, you could create a GHP stat (good hockey play) that would give you a +1 when you make a good hockey play:
(blocks shot, deflects pass, frees up puck, ices the puck, steals puck, caught out of position etc...) and -1 when you don't and weight their importance based upon result (-2 if it results directly in a goal).

Because at the end of the day - Severson's Corsi will look fine, and Santini might get credit for blocked shots (10 ?! the other night) - but no one is counting the passes that Santini breaks up with a good stick - or the goals that happen because Severson fails to cut the pass across the crease.

I mean - we could put in the human effort to do exactly that - watch each period and count when players do good things and bad things, but ultimately it will suffer the same as shot totals do, because of subjective viewing.

Ideally - on a small scale - it would be easy to do (just assign someone to watch player X all night) or just do goalies to start and solidify their statistics - and ideally it would be great to make a pact with the rival team's fan - (so that I watch and tally Mitch Marner's stats while Leafs fan is tallying Adam Henrique, for example) for objectivity.

If we were to create a list of what constitutes good hockey plays - then you'd get a really good approximation of a player's impact on a game and how he compares to his peers and others.

I personally was thinking about whipping out some of my programming chops and seeing if I could create a commentary compiler that uses javascript to suck in transcripts of radio play-by-play calls and uses some AI parse the text with some LISP-based language. It's a pipedream, but at the minimum to start you could tally how many times players are mentioned in a game, and whether or not they are mentioned in a positive or a negative and award points on that.

 

[billingtons ghost]

Everyone is saying that stats need context - or that in any case it is difficult to draw accurate absolute conclusions from them. CF, CF+ and even Fenwick can describe play - but in terms of an actual player's game impact, they can be arguably a watered down derivative of +/-.

The most advanced stats of today are just too simplistic to make black and white statements about a player's impact.

 

[Emperoreddy]

Hey this thread

 

[BenedictGomez]

I was curious about the Devils Giveaways & Takeaways this season & decided to do some work beyond what exists. That stat, IMO, is only valuable when you look at a players differential (i.e. Takeaways minus Giveaways), which you can easily find & is something people do look at.

But I thought it might be more meaningful (or at least interesting) when you take a look at that data based on TOI, which is not something you can find. So I did the math.

It turns out Mueller at a push is the "best" on the Devils D, and Moore was the worst. Mueller's data of course is admittedly limited, so among those with at least 900 minutes of TOI, Lovejoy was the best. I think this will surprise people, but it really doesnt shock me given both Lovejoy & Mueller play a very safe & low-risk defensive game.

Takeaways / Giveaways (Per Time On Ice in Minutes)

That said, analyzing it on a Time on Ice basis, admittedly didn't change much from simply looking at the mathematical takeaway / giveaway differential as is commonly done. Mueller & Lovejoy are still at the top, and Severson & Moore are still at the bottom.

Takeaways / Giveaways (Differential)

Simply looking at the raw data on a per minute basis, Mueller, Lovejoy, and Severson have the highest Takeaways per minute, and Severson, Moore, and Butcher cough the puck up the most.

 

[MartyOwns]

i'm going to need a tour guide to find which thread to post things in. or one of those sherpas you can pay in skittles

 

[BenedictGomez]

Should have also mentioned, 3 of the top-10 in the entire NHL in Takeaway/Giveaway are Devils.

#3 Gibbons
#4 Coleman
#9 Zajac

Impressive.

 

[AfroThunder396]

Takeaways and giveaways are inherently limited because :

i.) They are not standardized and are recorded differently at every venue. Every arena has its own criteria for what constitutes a giveaway/takeaway, meaning theres variation depending on where each game takes place.

ii.) They are highly situational. By definition, you cannot have a takeaway unless the other team has the puck and you cannot have a giveaway unless you have the puck. Players who have the puck on their stick a lot tend to have higher Take/Give ratios simply because they have the puck more often, giving the opponent more opportunities to retrieve it and their team mate less time to get it back.

 

[My3Sons]

Why hasn’t someone tried NFL style grading over the course of a game? I’d be curious if that would provide new or different results. Sure there is a subjective component but that’s true of any stats that have an opinion element about them. High danger or high quality chances are opinions at some point in the process so why not track a player by executing not executing his assignments and see what develops?

 

[Zajacs Bowl Cut]

Why hasn’t someone tried NFL style grading over the course of a game? I’d be curious if that would provide new or different results. Sure there is a subjective component but that’s true of any stats that have an opinion element about them. High danger or high quality chances are opinions at some point in the process so why not track a player by executing not executing his assignments and see what develops?

There is something like that around

here is last night's as an example:

 

[My3Sons]

There is something like that around

here is last night's as an example:

That’s more like a combo stat to my eyes. I’m not seeing anything that tracks individual assignments over the course of a game. I could be misreading it.

 

[Devils731]

https://t.co/sWb6SboCb4?amp=1

Possibly interesting article in the Athletic. If you get the app they let you have 3 free articles so you can check it out if you want, even as a non-subscriber.

 

[njdevils19]

Hall has a much higher sports IQ than I ever thought when we got him. He is also pretty well-spoken and an incredible teammate. IDK if the trade triggered some of this but man did that get lost on me when he was in EDM.

 

[Devils Dominion]

Hall has a much higher sports IQ than I ever thought when we got him. He is also pretty well-spoken and an incredible teammate. IDK if the trade triggered some of this but man did that get lost on me when he was in EDM.

He's true captain material

I'm surprised that I didn't see more Hall jerseys at The Rock last season, I'll bet I see a lot more in October.

 

[Jets012]

Take that Corsi Haters! Hall's down for the cause

 

[devilsblood]

Take that Corsi Haters! Hall's down for the cause

I think he's noting Corsi is flawed while looking for a better advanced stat.

 

[njdevils19]

Yeah he was pointing out that Corsi thought he had a better 16-17 vs. 17-18.