Toggle sidebar

Is our defence now dare I say it good?

  • Xenforo Cloud has upgraded us to version 2.3.6. Please report any issues you experience.
Now that analytics are judged unworthy, we shall now go back to using the basic all situations shots against to decide who the best defensive teams are.

"Looks at numbers"

Ok, Toronto has the 5th best D in the NHL and the best in the North!

I'd hate to be one of those botton 5 defensive trainwrecks like Columbus or Dallas though. It is early. Only takes a few games and we can slip down to Minnesota/Islanders territory in the middle of the pack. If we get lucky though, we can catch the stingy Pens for 4th soon.

:sarcasm:
 
  • Like
Reactions: Menzinger, Critical91, Dekes For Days and 1 other person
PromisedLand said:
My personal opinion on the so called advnaced stats is that they are not that "advanced"

When i look at the formula for xGF there is plenty wrong with it, for example,

  • Shot quality
  • was the shot deflected/high-wide
  • was the shot well defended
regardless of the outcome all the above three are treated equally in xGF and so it is a meaningless statistic

expected value does not mean "actual occurrence"

Also, facing Tampa/Boston/Pens/Caps/Florida/NYI/ heck even sabres is more challenging than facing flames/canucks/oilers

if flames/oilers/jets/canucks and ottawa/montreal have gotten significantly better than what they were last season then one can make an argument that northern division is more competitive than playing more against the atlantic division and metropolitan division teams
Click to expand...

"Advanced" doesn't necessarily mean "complicated". Advanced stats are pretty much anything that is not production based (i.e. goals, assists, points, etc.).

It is also certainly not a useless stat, unless you don't give a crap about how teams are actually formulating the opportunities to score. When you, as a team or coach, can't afford to be bipolar from game-to-game, they have to care about this data.
 
  • Like
Reactions: Dekes For Days
Zybalto said:
Now that analytics are judged unworthy, we shall now go back to using the basic all situations shots against to decide who the best defensive teams are.

"Looks at numbers"

Ok, Toronto has the 5th best D in the NHL and the best in the North!

I'd hate to be one of those botton 5 defensive trainwrecks like Columbus or Dallas though. It is early. Only takes a few games and we can slip down to Minnesota/Islanders territory in the middle of the pack. If we get lucky though, we can catch the stingy Pens for 4th soon.
Click to expand...

Way too early. The data sets are not big enough to draw any conclusions yet.

However there are some signs of encouragement. The fact that Dermott and Bogosian haven't imploded yet is a good sign.

Brodie is gradually adapting to playing with a rover like Rielly by staying back.

The mere fact they've been able to hold a lead is a positive step. Last season it was basically a train wreck when under pressure. The endless cycling of the opponents in the leafs end became the norm. While not flawless there's been less of that this season.

How would they fare against a team like Tampa? That remains to be seen.
 
SeaOfBlue said:
"Advanced" doesn't necessarily mean "complicated". Advanced stats are pretty much anything that is not production based (i.e. goals, assists, points, etc.).

It is also certainly not a useless stat, unless you don't give a crap about how teams are actually formulating the opportunities to score. When you, as a team or coach, can't afford to be bipolar from game-to-game, they have to care about this data.
Click to expand...

I have already laid out many times why the way the stats are being used on these boards to make "claims" are not correct.

you are welcome to go through my post histroy dig it up and respond to it
 
  • Like
Reactions: Cobra777 and NikoEhlers
zeke said:
Me i have hellebyuck up top, holtby at the bottom, and everyone else together in the middle.
Click to expand...
I'd have Price a tier above that middle just because he always seems to play decently when the pressure is on.

Murray has been hot/cold under pressure, we know Fred and the others are unknown/similar to Murray. At least Price offers some level of stability in key spots, from what I have off the top of my head.
 
PromisedLand said:
I have already laid out many times why the way the stats are being used on these boards to make "claims" are not correct.

you are welcome to go through my post histroy dig it up and respond to it
Click to expand...
I think you should take them with a grain of salt. They aren't the definite answer to anything, but really no stat or interpretation is.

Here's my question for those that dismiss them though: What's your superior alternative?

I've gone from dismissing to considering because I don't have a better alternative and I've tried to take them within their context instead of their naming convention (which can be misleading IMO)
 
Gallagbi said:
I think you should take them with a grain of salt. They aren't the definite answer to anything, but really no stat or interpretation is.

Here's my question for those that dismiss them though: What's your superior alternative?

I've gone from dismissing to considering because I don't have a better alternative and I've tried to take them within their context instead of their naming convention (which can be misleading IMO)
Click to expand...

any stat without "context" is really misleading. if you read any good scientific or mathematical economic papers there are assumptions, limitations and then significant adjustments to the data before being used in the model for predition given scenarios.

the hockey stats do not do it or if they do they are not fully transparent in how they are doing it on any website that publishes their data. When they are not transparent then they should be avoided.

The alternative depends on what do we want to use the data for? Analyze what exactly? and how do you probe it?

Using data to form an opinion is one thing; using the data to lay a claim that "expectations" mean "reality" like the xgf stats do without any quality control is meaningless.

Its like the corsi stats that says "shot attempts directed towards the net blocked or unblocked"; what kind of shot? perimeter shots?

but folks here use like corsi and xgf is the be all and end all statistics and it pisses me off.
 
  • Like
Reactions: NikoEhlers
PromisedLand said:
in a high event enviornment where data is very random it is not easy to "probe" information. However, if one understands the "reason" for the high event enviornment and "quality of agents in that system" then one can design a model to probe it.

Examples:

- Shots against (group by categories: shot quality, distance from the net, distance of the "defender" from the person taking the shot)

- time spent
(time spent in the defensive zone Vs offensive zone Vs neutral zone)

- Isolation of contribution to success based on cap%
isolate how players by their roles (center, defense, etc) contribute to team sucess in seprate categories 1) offense and 2) defense and 3) overall impact based on their cap% of the team standardizing the analytics
i.e. if you make more and do more there is nothing special; because you are paid like that. however, if you get paid more but do better than your peers who also get paid more and equal footing in TOI/linemates

- weight it appropirately to come up with 1 metric that is a weighted average of all of the above + other "reasons" one may have identified before modelling the data.

As you can see things get complicated very fast in a high-event- random event like hockey. And sophisticated mathematical models designed by people who know both hockey + math + statistics + data science should be doing that not some twitter warrior
Click to expand...

Id certainly love to see some sort of unified grand theory at some point but all the models I look at were developed through thousands of hours of hard work by folks with exactly that backgrond. There are different models out there with some variance but they are far more thorough than you are giving them credit for.

Here's are some of the weighted items Moneypucks overall Xg shot model takes into account:

  1. Shot Distance From Net
  2. Time Since Last Game Event
  3. Shot Type (Slap, Wrist, Backhand, etc)
  4. Speed From Previous Event
  5. Shot Angle
  6. East-West Location on Ice of Last Event Before the Shot
  7. If Rebound, difference in shot angle divided by time since last shot
  8. Last Event That Happened Before the Shot (Faceoff, Hit, etc)
  9. Other team’s # of skaters on ice
  10. East-West Location on Ice of Shot
  11. Man Advantage Situation
  12. Time since current Powerplay started
  13. Distance From Previous Event
  14. North-South Location on Ice of Shot
  15. Shooting on Empty Net


They also test the hell out of these things and try to improve them. There professional credibility on the line. Its not just some twitter kids. It's some guys passion and job too.
 
  • Like
Reactions: Menzinger, Dekes For Days and zeke
PromisedLand said:
any stat without "context" is really misleading. if you read any good scientific or mathematical economic papers there are assumptions, limitations and then significant adjustments to the data before being used in the model for predition given scenarios.

the hockey stats do not do it or if they do they are not fully transparent in how they are doing it on any website that publishes their data. When they are not transparent then they should be avoided.

The alternative depends on what do we want to use the data for? Analyze what exactly? and how do you probe it?

Using data to form an opinion is one thing; using the data to lay a claim that "expectations" mean "reality" like the xgf stats do without any quality control is meaningless.

Its like the corsi stats that says "shot attempts directed towards the net blocked or unblocked"; what kind of shot? perimeter shots?

but folks here use like corsi and xgf is the be all and end all statistics and it pisses me off.
Click to expand...
So again, what do you use as the alternative?

Corsi is a great example of my last point. I don't like it as "possession", but I will take it as shot attempts. If you want shot quality, you start considering Corsi in addition to the others, which you piece together for a more comprehensive picture.

I find most of the stats are used to dismissed anecdotal assumptions based on extremely small (and bias) samples or in an attempt to enlighten.
 
DarkKnight said:
Brodie is such a big upgrade, particularly when he replaces a defensive stiff like Barrie. I don’t mind our D now at all, particularly the depth, wherein a Sandin is still an intriguing piece moving forward.
Click to expand...

Ive especially noticed Brodie on the PK. He seems to make really good decisions and has a good stick. He’s settled in nicely after looking a little wild in the first two games.
 
  • Like
Reactions: DarkKnight
Zybalto said:
Not too far off what my ranking would be, the only difference being that I would have Markstrom solidly in third ahead of Andersen/Murray but all this can change of course as 3 of those keepers are on new teams.

Im not sure too many are aware of just how bad Vancouvers D was last year and how well Markstrom covered up for it. I'm talking maybe worst in the league bad.
Click to expand...
Hellebuyck needs no explanation. Price imo is still the best goalie in the world when he is on. I just think he needs to get away from that team but he's off to a good start. I would have Markstrom and Freddy as equal as I think both are top 10 goalies when on but not always consistent. As it stands right now I have those 2 grouped with Murray to start the year. Holtby is past his prime but I think I would still take him over Koskinen as Koskinen hasn't proven anything yet.
zeke said:
Me i have hellebyuck up top, holtby at the bottom, and everyone else together in the middle.
Click to expand...
 
Gallagbi said:
So again, what do you use as the alternative?

Corsi is a great example of my last point. I don't like it as "possession", but I will take it as shot attempts. If you want shot quality, you start considering Corsi in addition to the others, which you piece together for a more comprehensive picture.

I find most of the stats are used to dismissed anecdotal assumptions based on extremely small (and bias) samples or in an attempt to enlighten.
Click to expand...

Alternative: at present no alternatives exist.

Second, it depends on the want of an "alternative" and the kind of "alternative" one is seeking.

Example; If I am looking at the hamburgers consumed in Australia versus goals scored in NHL and I find a correlation; does that mean if people in Australia stopped eating hamburgers NHL goals would dry up? This statistic is meaningless. But if I cannot find the alternative I am suppose to just rely on meaningless statistic is an argument I am not in favour.

Using a statistic to probe few things is one thing; using a statistic to lay a claim that is the be all and end all like it is done on this board is another thing. The latter is dangerous.

Past events being a predictor of future events works relatively well in sciences because laws of nature more often than not are fixed and obey a trend i.e. not random.

hockey is very random. If Matthews scores a goal from a given position - or a team scores a goal from a given position xgf will always count it as "expected goal"; using it to lay a claim that yes xgf was 80% but the actual event might be 20% is a difference between chalk and cheese. CBJ series last year is a great example.

using any advanced metric would have shown that leafs were better; but in reality if one watched the series it would have shown that clogging neutral zone and not giving any time and space screwed up leafs offense. A good example from that series is Matthews trying to take the shot from slot but was always well defended; in the metric it will be counted as HDCF% and xgf%.

The system that CBJ employed worked to their advantage although hockey metrics would say that leafs were unlucky to lose that series. that however, is not the truth.
 
Zybalto said:
Id certainly love to see some sort of unified grand theory at some point but all the models I look at were developed through thousands of hours of hard work by folks with exactly that backgrond. There are different models out there with some variance but they are far more thorough than you are giving them credit for.

Here's are some of the weighted items Moneypucks overall Xg shot model takes into account:

  1. Shot Distance From Net
  2. Time Since Last Game Event
  3. Shot Type (Slap, Wrist, Backhand, etc)
  4. Speed From Previous Event
  5. Shot Angle
  6. East-West Location on Ice of Last Event Before the Shot
  7. If Rebound, difference in shot angle divided by time since last shot
  8. Last Event That Happened Before the Shot (Faceoff, Hit, etc)
  9. Other team’s # of skaters on ice
  10. East-West Location on Ice of Shot
  11. Man Advantage Situation
  12. Time since current Powerplay started
  13. Distance From Previous Event
  14. North-South Location on Ice of Shot
  15. Shooting on Empty Net


They also test the hell out of these things and try to improve them. There professional credibility on the line. Its not just some twitter kids. It's some guys passion and job too.
Click to expand...

It's almost entirely weighted toward the shot distance from the net.

And because of that, "Expected" goals don't really measure what's expected at all - but rather where a teams' shots are coming from on the ice.

No measure of quality is given.

A flubbed shot with a rolling puck from Zach Bogosian taken from the same location as a shot from Auston Matthews is weighted the same.

The model simply can't recognize the difference between the two shots and most importantly, the players taking them.

So yeah, while I don't totally dismiss these stats, I just don't find an "average distance from the net" statistic to be all that meaningful or insightful. Teams and coaches vary across the board in terms of strategizing around quality vs. quantity of shots of net.
 
Zybalto said:
Id certainly love to see some sort of unified grand theory at some point but all the models I look at were developed through thousands of hours of hard work by folks with exactly that backgrond. There are different models out there with some variance but they are far more thorough than you are giving them credit for.

Here's are some of the weighted items Moneypucks overall Xg shot model takes into account:

  1. Shot Distance From Net
  2. Time Since Last Game Event
  3. Shot Type (Slap, Wrist, Backhand, etc)
  4. Speed From Previous Event
  5. Shot Angle
  6. East-West Location on Ice of Last Event Before the Shot
  7. If Rebound, difference in shot angle divided by time since last shot
  8. Last Event That Happened Before the Shot (Faceoff, Hit, etc)
  9. Other team’s # of skaters on ice
  10. East-West Location on Ice of Shot
  11. Man Advantage Situation
  12. Time since current Powerplay started
  13. Distance From Previous Event
  14. North-South Location on Ice of Shot
  15. Shooting on Empty Net


They also test the hell out of these things and try to improve them. There professional credibility on the line. Its not just some twitter kids. It's some guys passion and job too.
Click to expand...

dude if the prediction from xgf is significantly different than actual gf than that model is not useful
 
PromisedLand said:
A paper that does not has a section on limitation is not a paper that one should rely upon.

If this was released in an academic journal this would be tossed out.

Here is a question; how well does xgf track actual gf? do you have a number?
Click to expand...

You didn't even read it.
 
  • Like
Reactions: Dekes For Days
Dekes For Days said:
If only we had a stat that could help answer and adjust for that question... we could name it something like... expected goals.
Click to expand...

if only people stopped peddeling other people's work without actually understanding the mathematical and conceptual designs and claiming it to be true as if they did it themselves then we would name it something like .... "competent arguments"
 
Bomber0104 said:
It's almost entirely weighted toward the shot distance from the net.

And because of that, "Expected" goals don't really measure what's expected at all - but rather where a teams' shots are coming from on the ice.

No measure of quality is given.

A flubbed shot from Zach Bogosian taken from the same location as Auston Matthews is weighted the same.

The model simply can't recognize the difference between the two shots and most importantly, the players taking them.
Click to expand...

Actually as you can see from his list there are many factors it uses to determine shot quality other than shot distance.

And many xgf models to incorporate shooting talent (i.e. who is doing the shooting) too.
 
You must log in or register to reply here.

Latest posts

Ad

Upcoming events

Ad

Top Bottom