Dave Nonis: [mod: Commentary on advanced stats] (Video)

[K19*]

Back in November of last year







This team is doomed

What an idiot.

Click to read more...

 

[SaskRinkRat]

"If we have the puck all the time, and our goalie can't stop it, we lose the game."

That's a really hard statement to get behind.

"If the other team never gets a shot on our net, because they never have the puck, and our goalie can't stop it, we lose the game." No you don't. At worst, you tie 0-0.

 

[BraveCanadian]

"If we have the puck all the time, and our goalie can't stop it, we lose the game."

That's a really hard statement to get behind.

"If the other team never gets a shot on our net, because they never have the puck, and our goalie can't stop it, we lose the game." No you don't. At worst, you tie 0-0.

You're just stating the other ludicrous extreme.

Obviously having the puck more often than the opposition is a good thing in general but I think what he was saying is that if I have the choice between good Corsi rating and a good goaltending performance I'll take the good goaltending every time.

I agree with that.. teams that buck the Corsi prediction trend tend to do so based on their goaltending.

 

[hatterson]

Obviously he's speaking in hyperbole with his "have the puck all the time" comment.

In the video he does say that obviously it's a good thing to have the puck as opposed to not have it. Anyone who has ever watched hockey will tell you that. I would rather that I have the puck than my opponent has it. That's not some huge analytics breakthrough.

His point was that having the puck, in and of itself, doesn't win a hockey game. It doesn't magically make you better.

He's right that corsi and fenwick, in and of themselves, are useless. Like all stats, they require context. Granted he goes further than I would in regards to how far to the side he pushes corsi, but it's not like he's just plugging his ears and yelling "LALALA" He says that there's likely advanced stats and analytics that will help, he just hasn't seen them yet.

 

[SaskRinkRat]

You're just stating the other ludicrous extreme.

Obviously having the puck more often than the opposition is a good thing in general but I think what he was saying is that if I have the choice between good Corsi rating and a good goaltending performance I'll take the good goaltending every time.

I agree with that.. teams that buck the Corsi prediction trend tend to do so based on their goaltending.

Hot goaltending has been shown to be unsustainable, whereas corsi dominance has been shown to be repeatable.

My original point still stands. The more you have the puck, the less you have to worry about your goaltending, because the area of opportunity for shot attempts against is smaller. Think about it this way:

Suppose that your opponent averages 1.0 shot attempts for every minute they have the puck. So if they had the puck all game they would get 60 shot attempts toward your net.

Now suppose you have a really solid goalie with a career .920 save percentage.

If 75% of the other team's shot attempts hit the net, they're going to get 45 shots on your net during their 60 minutes with the puck, and your goalie is going to allow 3.6 goals (over time). If your goalie gets hot for a stretch, and bumps his SV% up to .940, he's going to reduce his goals against to 2.7 on 45 shots.


Suppose you can reduce the amount of time the other team has the puck from the full 60 minutes to 45 minutes. You have the same goalie, but you're a team that only gives up 45 shot attempts and 34 of those (75%) hit the net. Your goalie is going to allow 2.72 of those when he's playing at his normal .920 SV% and only 2.04 when he's at .940.

Same goalie playing the exact same way allows substantially fewer goals when his team reduces the number of shot attempts he gets.


The other side of this equation often goes ignored - there is only one puck on the ice, so the more the other team has the puck, the less you have it. In our example above, you have the puck for 0 minutes in the first scenario, so you can generate 0 shots on the other net. You lose 3.6-0 when your goalie is performing to average, and you lose 2.7-0 when he's on fire. In the second scenario you actually have a chance to generate shot attempts toward the other team's net AND you're getting scored on less, even if your goalie plays to the exact same level.

 

[Finnish your Czech]

Can you please change the title. Dave Nonis never said "Advanced stats, like Corsi, are useless."

He said "I wouldn't say analytics are useless" and "Corsi and Fenwick are interesting, but that's all they are".

 

[tarheelhockey]

Listen to the entire clip... he explicitly states that advanced stats can be useful. He just doesn't think analyzing Corsi is helpful when it comes to the actual winning and losing of games.

Even as someone who likes an uses advanced stats, I agree with him. Yes, Corsi is a helpful forensic tool for analyzing outcomes, and has pretty good predictive value. But if you were the GM or coach of a hockey team, what good would Corsi do you? "Your team doesn't do well in the possession game"... well yeah, you know that already. That doesn't help you solve the problem. I can imagine why he gets tired of people presenting Corsi analysis as if it's going to change the way he looks at his team.

It's not so much an old-guard/new-guard conflict, as a hockey-ops/fanbase difference in perspective.

 

[hatterson]

Can you please change the title. Dave Nonis never said "Advanced stats, like Corsi, are useless."

He said "I wouldn't say analytics are useless" and "Corsi and Fenwick are interesting, but that's all they are".

I know he said something similar in an interview, however you're right, the clip doesn't contain that quote, so I've removed it from the title. Must have either been just before the clip or in another interview where he said something similar to the original quote.

 

[TheDevilMadeMe]

Hot goaltending has been shown to be unsustainable, whereas corsi dominance has been shown to be repeatable.

"Hot" goaltending is obviously unsustainable, or else it wouldn't be called "hot." Strong goaltending, however, has been and always will be, sustainable by strong goaltenders.

 

[SaskRinkRat]

Even as someone who likes an uses advanced stats, I agree with him. Yes, Corsi is a helpful forensic tool for analyzing outcomes, and has pretty good predictive value. But if you were the GM or coach of a hockey team, what good would Corsi do you? "Your team doesn't do well in the possession game"... well yeah, you know that already. That doesn't help you solve the problem.

There are a couple of obvious things he can do:

1) Stop signing and trading for players who don't drive puck possession.

2) Stop signing coaches whose tactics don't drive puck possession.

3) Get on the bus and start contributing to the exploration of advanced stats so we can better understand what sorts of team systems and individual skills drive puck possession.

 

[SaskRinkRat]

"Hot" goaltending is obviously unsustainable, or else it wouldn't be called "hot." Strong goaltending, however, has been and always will be, sustainable by strong goaltenders.

And strong goaltending will look even stronger when your opponent has the puck less.

 

[tarheelhockey]

There are a couple of obvious things he can do:

1) Stop signing and trading for players who don't drive puck possession.

2) Stop signing coaches whose tactics don't drive puck possession.

3) Get on the bus and start contributing to the exploration of advanced stats so we can better understand what sorts of team systems and individual skills drive puck possession.

What he's saying, though, is that he is already aware of the need to have better puck possession than the other team. He doesn't need extensive number crunching to confirm what common hockey sense would tell him.

Corsi might give him some window into exactly what players to pursue, but then again he also has paid professional scouts whose job it would be to have that kind of information on-hand anyway. Same with coaches -- he could probably tick off the best possession-game coaches in the league without needing to refer to a stat sheet. It's his job to know those figures and to pursue the best ones available.

As for #3, I'm not sure why we would expect an NHL manager to become an advanced stats pioneer. They seem pretty busy with other things.

 

[SaskRinkRat]

What he's saying, though, is that he is already aware of the need to have better puck possession than the other team. He doesn't need extensive number crunching to confirm what common hockey sense would tell him.

Corsi might give him some window into exactly what players to pursue, but then again he also has paid professional scouts whose job it would be to have that kind of information on-hand anyway. Same with coaches -- he could probably tick off the best possession-game coaches in the league without needing to refer to a stat sheet. It's his job to know those figures and to pursue the best ones available.

The whole point of analytics is to supplement the eye test. You can't see everything. It's also pretty clear that many of the "common hockey knowledge" beliefs are flat out wrong, when it comes to winning hockey games.

On your point about having a list of coaches who drive puck possession. If not puck possession stats, what would determine where a coach sits on such a list?

As for #3, I'm not sure why we would expect an NHL manager to become an advanced stats pioneer. They seem pretty busy with other things.

I would expect an NHL manager to become a pioneer in anything that has been proven to increase the chances of his team winning games.

 

[tarheelhockey]

The whole point of analytics is to supplement the eye test. You can't see everything. It's also pretty clear that many of the "common hockey knowledge" beliefs are flat out wrong, when it comes to winning hockey games.

I don't think he's claiming to see everything. He's saying that he spends all day, every day, reviewing game film and scouting reports. For someone to come along and give him a graph that essentially says "you didn't shoot enough", isn't helpful. He knows that already.

He isn't bashing advanced stats in general, or even targeting Corsi's analytical value. He's saying it isn't useful to him in preparing his team to win games -- primarily because it simply explains what the end-goal of possession should be rather than providing a road map for success which can be applied in a real-world situation.

On your point about having a list of coaches who drive puck possession. If not puck possession stats, what would determine where a coach sits on such a list?

Detailed knowledge of the offensive and defensive systems that each coach instills on his team? Some coaches instruct their teams to retain possession at all costs, others emphasize volume-based shooting attempts, others focus on generating a few quick but high-quality rush opportunities.

This was all known before Corsi -- and again, it's the GM's job to scout and understand the various systems in place around professional hockey. Corsi gives us a window into each team's results, but it isn't a necessary element in understanding what each team is trying to do.

I would expect an NHL manager to become a pioneer in anything that has been proven to increase the chances of his team winning games.

Corsi has been proven to have predictive value regarding a team's chances of winning. I'm not so sure you can say it has been proven to increase the chances of winning through some sort of application.

 

[SaskRinkRat]

I don't think he's claiming to see everything. He's saying that he spends all day, every day, reviewing game film and scouting reports. For someone to come along and give him a graph that essentially says "you didn't shoot enough", isn't helpful. He knows that already.

He isn't bashing advanced stats in general, or even targeting Corsi's analytical value. He's saying it isn't useful to him in preparing his team to win games -- primarily because it simply explains what the end-goal of possession should be rather than providing a road map for success which can be applied in a real-world situation.

It sounds to me like he lacks an understanding of how possession stats relate to other statistics (or other behaviours, like goaltending, that can be measured statistically). His comment that I quoted above, regarding puck possession and bad goaltending, is one example. He doesn't appear to understand basic concepts that relate puck possession to goaltending outcomes (i.e., more puck possession for your team = less puck possession for opponent = less shot attempts against = less shots on goal against = less goals for a given level of goaltending).

Detailed knowledge of the offensive and defensive systems that each coach instills on his team? Some coaches instruct their teams to retain possession at all costs, others emphasize volume-based shooting attempts, others focus on generating a few quick but high-quality rush opportunities.

The only way to understand whether these systems result in increased or decreased puck possession is to actually measure whether or not they do. Someone at some point along the information flow is going to actually have to sit down and determine how each of these systems impacts puck possession by counting something, and the proven (and intuitive) link between shot attempts and puck possession suggests that shot attempts are a good thing to count.

Corsi has been proven to have predictive value regarding a team's chances of winning. I'm not so sure you can say it has been proven to increase the chances of winning through some sort of application.

If there is evidence suggesting that teams that do better at A have historically resulted in more B, and you want more B, then you'd be pretty silly if you didn't at least consider doing better at A.

 

[tarheelhockey]

It sounds to me like he lacks an understanding of how possession stats relate to other statistics (or other behaviours, like goaltending, that can be measured statistically). His comment that I quoted above, regarding puck possession and bad goaltending, is one example. He doesn't appear to understand basic concepts that relate puck possession to goaltending outcomes (i.e., more puck possession for your team = less puck possession for opponent = less shot attempts against = less shots on goal against = less goals for a given level of goaltending).

I don't think he was saying anything like that, though. He was just throwing out an extreme example about how real-life circumstances can undermine statistical models. IE, even if you dominate possession, you can still lose the game because your goalie sucked. I doubt he meant anything more complex than that.

The only way to understand whether these systems result in increased or decreased puck possession is to actually measure whether or not they do.

Yes but you are talking about outcomes as opposed to inputs. When it comes to hiring a coach, the GM is dealing with the input side of the equation. The results are already known (ie, how good was this guy's team offensively) so it's far more helpful to spend time and resources analyzing that coach's inputs (the Xs and Os of what he instructed his players to do on the ice) rather than continue drilling down into outcome-oriented data just to make redundant conclusions about puck possession.

Again, this highlights the difference between the fan's interest in manipulating data for the sake of analysis or prediction, and the GM's interest in finding the right balance of inputs to generate a desirable outcome.

If there is evidence suggesting that teams that do better at A have historically resulted in more B, and you want more B, then you'd be pretty silly if you didn't at least consider doing better at A.

That's his entire point -- of course he wants to do better at A. He knows that already. Finding new ways to prove the value of A is irrelevant to him. And presuming he already has file cabinets full of information about which coaches and players have skillsets which could contribute more A to his team (which he should if he is doing his job correctly), then further analysis of which players produced the most A last week is simply going to be annoying to him.

 

[SaskRinkRat]

I don't think he was saying anything like that, though. He was just throwing out an extreme example about how real-life circumstances can undermine statistical models. IE, even if you dominate possession, you can still lose the game because your goalie sucked. I doubt he meant anything more complex than that.

If this is the point he's trying to make, it's not a legitimate point. If your goalie sucks, you can only LESSEN the impact of his lack of skill by improving your possession game.

Yes but you are talking about outcomes as opposed to inputs. When it comes to hiring a coach, the GM is dealing with the input side of the equation. The results are already known (ie, how good was this guy's team offensively) so it's far more helpful to spend time and resources analyzing that coach's inputs (the Xs and Os of what he instructed his players to do on the ice) rather than continue drilling down into outcome-oriented data just to make redundant conclusions about puck possession.

The problem is without analyzing the outputs, you have no way of evaluating the inputs. Determining that Coach Bob using this sort of X and O's structure is completely useless if you don't bother to determine whether that X and O's structure results in having the puck more often than the other team.

That's his entire point -- of course he wants to do better at A. He knows that already. Finding new ways to prove the value of A is irrelevant to him. And presuming he already has file cabinets full of information about which coaches and players have skillsets which could contribute more A to his team (which he should if he is doing his job correctly), then further analysis of which players produced the most A last week is simply going to be annoying to him.

It seems to me like you're missing a key link in the equation. In order to get information, you have to be willing to actually collect the information. Your argument, in a financial context, would boil down to something like this: "I want to know how much money I have, but I don't really think it's important that anyone bothers to count my money." Or, alternatively "I want to have more money, but I don't think it's important to know how much money I have right now or what impact my decisions have on whether I make more money or not."

I want to improve puck possession, but it's not useful to know how often I have the puck, or how well potential new players drive puck possession.

 

[BraveCanadian]

It sounds to me like he lacks an understanding of how possession stats relate to other statistics (or other behaviours, like goaltending, that can be measured statistically). His comment that I quoted above, regarding puck possession and bad goaltending, is one example. He doesn't appear to understand basic concepts that relate puck possession to goaltending outcomes (i.e., more puck possession for your team = less puck possession for opponent = less shot attempts against = less shots on goal against = less goals for a given level of goaltending).

There is no one involved in the game of hockey that doesn't understand this...

 

[SaskRinkRat]

There is no one involved in the game of hockey that doesn't understand this...

Yet there are oodles of people dismissing the importance of statistics that tell us about puck possession.

 

[BraveCanadian]

Yet there are oodles of people dismissing the importance of statistics that tell us about puck possession.

That is only because they don't really tell anyone something they don't already know from watching the game or even the old fashioned shot totals.

All Corsi does is give you a little bit better a proxy for the same information because there are more events.

 

[SaskRinkRat]

Let's inject some actual data into the discussion.

Leafs 2013-14 5on5 shot attempts against = 5508
Leafs 2013-14 5on5 shots against = 2945
Leafs shots against per shot attempt against = 0.535

League average 2013-14 5on5 shot attempts against = 4584
League average 2013-14 5on5 shots against = 2464
League average shots against per shot attempt against = 0.538

So there is extremely little difference between the Leafs and the league average when it comes to how many shot attempts toward their own net actually end up hitting the net.


Leafs 2013-14 5on5 SV% = .914
League average 2013-14 5on5 SV% = 0.911

So the Leafs actually had better than average goaltending.


Leafs 2013-14 Goals Against = 252
League average 2013-14 Goals Against = 219

...yet allowed drastically more goals against than the league average.


I wonder how many fewer goals the Leafs would have allowed if they'd had the puck more (and allowed the league average shot attempts against)?

4584 shot attempts against
x 0.535 shots on goal against per attempt against (Leafs actual rate)
= 2452 shots against
x .086 goals against per shot against (1.000 - .914 save percentage - Leafs actual)
= 211 goals against

So average puck possession with the exact same goaltending would have reduced the Leafs goals against by 41. They would have given up exactly 1 less goal every 2 games on average.

So yes, it does matter if your team has the puck more even if your goalie can't stop it.

 

[BraveCanadian]

Let's inject some actual data into the discussion.

Leafs 2013-14 5on5 shot attempts against = 5508
Leafs 2013-14 5on5 shots against = 2945
Leafs shots against per shot attempt against = 0.535

League average 2013-14 5on5 shot attempts against = 4584
League average 2013-14 5on5 shots against = 2464
League average shots against per shot attempt against = 0.538

So there is extremely little difference between the Leafs and the league average when it comes to how many shot attempts toward their own net actually end up hitting the net.


Leafs 2013-14 5on5 SV% = .914
League average 2013-14 5on5 SV% = 0.911

So the Leafs actually had better than average goaltending.


Leafs 2013-14 Goals Against = 252
League average 2013-14 Goals Against = 219

...yet allowed drastically more goals against than the league average.


I wonder how many fewer goals the Leafs would have allowed if they'd had the puck more (and allowed the league average shot attempts against)?

4584 shot attempts against
x 0.535 shots on goal against per attempt against (Leafs actual rate)
= 2452 shots against
x .086 goals against per shot against (1.000 - .914 save percentage - Leafs actual)
= 211 goals against

So average puck possession with the exact same goaltending would have reduced the Leafs goals against by 41. They would have given up exactly 1 less goal every 2 games on average.

So yes, it does matter if your team has the puck more even if your goalie can't stop it.

That is a lot of verbiage to say that, all other things being equal, if you give up less shots you'll get scored on less.. hardly breaking news.

 

[SaskRinkRat]

That is a lot of verbiage to say that, all other things being equal, if you give up less shots you'll get scored on less.. hardly breaking news.

So tell me again why it's unimportant to understand how many shots you're giving up?

Also - and I think importantly - if you give up fewer shots, even if all other things AREN'T necessarily equal (i.e., goaltending), you still might get scored on less.

If the Leafs gave up league average shot attempts against they could have had the Oilers goaltending and still given up 9 fewer goals than they did.

The analytics tell us that it's possible to sustain shot attempt differential long term, while it's not possible to sustain significant deviation from mean save percentage long term. In other words, instead of trying to improve your team by improving your already-better-than-average goaltending, you might as well try to do it by improving your shot share.

 

[TKB]

The whole point of analytics is to supplement the eye test. You can't see everything. It's also pretty clear that many of the "common hockey knowledge" beliefs are flat out wrong, when it comes to winning hockey games.

On your point about having a list of coaches who drive puck possession. If not puck possession stats, what would determine where a coach sits on such a list?


I would expect an NHL manager to become a pioneer in anything that has been proven to increase the chances of his team winning games.

Such as?

 

[SaskRinkRat]

Such as?

1) Hitting more leads to winning more.
2) Blocking more shots leads to winning more.
3) Guys who turn the puck over more are worse players.
4) Dumping and chasing is better than carrying the puck in.

and the list goes on.