#28 | The Utah HC vs Flyers | Sunday, December 8, 2024 | 7:30 PM | NBCSP, 97.5 FM

xGF and HDCF only ones that matter to me.
xGF is total quality of shots, HDCF is both opportunities created and "mistakes" made.

CF got inflated when teams started shooting from everywhere, same with Fenwick.
xGF corrects CF by accounting for where you're shooting from (actually Fenwick b/c it only includes shots on goal, I think).
To be fair, CF has some value with a team like the Flyers that block a lot of shots, since it gives an idea of possession.

A reminder that xGF doesn't measure true quality of shot because it doesn't factor in context. Like nature of set-up passing.

The Flyers don't do much high-quality playmaking of the sort that doubles scoring chance, so in a game where xGF is near even, the proper assumption should be that they're actually lagging in reality.
 
I think one reason Andrae is getting so much PT and they're sticking with Drysdale is they want at least one D-man on every pair (York and Risto are also carrying the puck more) who can skate the puck out of the D-zone when teams take away exit pass lanes.

The "safe" exit pass is no longer safe as NHL teams learn to pinch and cut off that pass.
The pass up the middle has always been dangerous.
So taking a few strides with your head up opens up more options (changes the angle on exit passes).
 
  • Wow
Reactions: Beef Invictus
A reminder that xGF doesn't measure true quality of shot because it doesn't factor in context. Like nature of set-up passing.

The Flyers don't do much high-quality playmaking of the sort that doubles scoring chance, so in a game where xGF is near even, the proper assumption should be that they're actually lagging in reality.
They try to play make, they're just not good at it right now, even Michkov has the majority of his setup passes blocked or intercepted. It's a learning curve for young players, they develop bad habits with open ice at lower levels and have to see and react faster in the NHL.

With a veteran like Couts, he's just not that good of a playmaker, he thrived when he had Giroux on his LW. Which is why I see him eventually moving to LW with a playmaker at center.

Cates isn't good at playmaking, but he has Brink, Frost isn't really that good either, but he has Michkov. Farabee and TK are more scorers than playmakers.
 
They try to play make, they're just not good at it right now, even Michkov has the majority of his setup passes blocked or intercepted. It's a learning curve for young players, they develop bad habits with open ice at lower levels and have to see and react faster in the NHL.

With a veteran like Couts, he's just not that good of a playmaker, he thrived when he had Giroux on his LW. Which is why I see him eventually moving to LW with a playmaker at center.

Cates isn't good at playmaking, but he has Brink, Frost isn't really that good either, but he has Michkov. Farabee and TK are more scorers than playmakers.

Their playmaking rarely crosses the ice and makes the goalie work to keep his angle. It's all north/South in the same area of ice or short range. That's the "low quality" aspect and we know that's by design because Tortorella has told us from day one it is his goal, and it is how all his teams have played.

The xGF for a team that makes a goalie do more than shuffle can be the same as a team that lets them stay close to set, but the former team will have had the better chances and higher probability of actually scoring. xGF doesn't cover that very well. The stat isn't designed to measure an outlier coach who discourages the most effective playmaking. It's treated the same as a coach who encourages it

Similar to how Corsi was useless to measure anything for Hakstol since a huge chunk of them were bad-idea shots from orbit, or individual Corsi stats for Carle were even less useful because his shots were so bad that it was arguable a higher Corsi was a negative. The more he shot the more we lost possession.

All these hockey stats are built around the average experience/expectation, and Tortorella in a lot of ways lies outside of that. The Flyers seem to promote or gravitate towards this in coaches, and unfortunately it isn't on the "new and innovative" side of the outlier spectrum.
 
Last edited:
Their playmaking rarely crosses the ice and makes the goalie work to keep his angle. It's all north/South in the same area of ice or short range. That's the "low quality" aspect and we know that's by design because Tortorella has told us from day one it is his goal, and it is how all his teams have played.
And every other team in the league knows this. They play the Flyers accordingly which accounts for boring games that end up in wipe outs against the better clubs. The lack of innovation on the PP is not the only predictable thing about this club. It's crying for innovation and that's not coming.
 
And every other team in the league knows this. They play the Flyers accordingly which accounts for boring games that end up in wipe outs against the better clubs. The lack of innovation on the PP is not the only predictable thing about this club. It's crying for innovation and that's not coming.
the Flyers idea of innovation is to rebuild the 1970s BSB.
 
xGF and HDCF only ones that matter to me.
I'd somewhat agree that xGF and HDCF are the 'most important' but as with everything there's nuance. You're can't ignore other parts of the dataset to build a model that suits a specific narrative. In this case, the narrative is 'they dominated the metrics battle' which isn't true. Even if you wanted to say they dominated the xG/HDCF battle, that's not even true. They won it, sure.

xGF is total quality of shots, HDCF is both opportunities created and "mistakes" made.
Not exactly, xGF is total amount of expected goals based on fenwick attempts. It has some issues, because if you take 80 fenwick attempts, but they're all from the point, you're still going to generate a reasonably high xGF. But' you'll also generate a high xG if you only take 10 shots from high danger shots. I think a better option for shot quality is 'shot danger' --basically fenwick attempts divided by the xG. Then on average, you can see how dangerous each shot was as described by that specific xG model.

HDCF is a specific metric to NST. And it's somewhat ambiguous for a host of reasons. Basically each corsi is given a value between 0-3+. Any shot with a 2+ rating or higher is a scoring chance (SCF/SCA), and any corsi with a 3 or higher is a HDCF. There are some drawbacks with this, it misses some nuance IMO, but it's a good 'back of napkin' type analysis.
CF got inflated when teams started shooting from everywhere, same with Fenwick.
xGF corrects CF by accounting for where you're shooting from (actually Fenwick b/c it only includes shots on goal, I think).
To be fair, CF has some value with a team like the Flyers that block a lot of shots, since it gives an idea of possession.
You've right, CF is a proxy for possession and zone time. You can't take lots of shots without having the puck in the offensive zone for a long time.

xG does not use CF numbers to generate the value, only Fenwick numbers, because as I said above**, the RTSS dataset does not provide shot location values for blocked shots, only a location value for where the block occurred.


**I did not say that above, my mistake**
 
Last edited:
People say a referee wistled right to his ear so he grabbed the wistle and was sent off. He can miss a game or two for this. Wouldn't be the worst thing actually...
Yeah, the ref was constantly blowing the whistle in his ear trying to get him to stop throwing jabs, and I guess he had enough and took swipe at his arm/whistle
 
I'd somewhat agree that xGF and HDCF are the 'most important' but as with everything there's nuance. You're can't ignore other parts of the dataset to build a model that suits a specific narrative. In this case, the narrative is 'they dominated the metrics battle' which isn't true. Even if you wanted to say they dominated the xG/HDCF battle, that's not even true. They won it, sure.


Not exactly, xGF is total amount of expected goals based on fenwick attempts. It has some issues, because if you take 80 fenwick attempts, but they're all from the point, you're still going to generate a reasonably high xGF. But' you'll also generate a high xG if you only take 10 shots from high danger shots. I think a better option for shot quality is 'shot danger' --basically fenwick attempts divided by the xG. Then on average, you can see how dangerous each shot was as described by that specific xG model.

HDCF is a specific metric to NST. And it's somewhat ambiguous for a host of reasons. Basically each corsi is given a value between 0-3+. Any shot with a 2+ rating or higher is a scoring chance (SCF/SCA), and any corsi with a 3 or higher is a HDCF. There are some drawbacks with this, it misses some nuance IMO, but it's a good 'back of napkin' type analysis.

You've right, CF is a proxy for possession and zone time. You can't take lots of shots without having the puck in the offensive zone for a long time.

xG does not use CF numbers to generate the value, only Fenwick numbers, because as I said above, the RTSS dataset does not provide shot location values for blocked shots, only a location value for where the block occurred.
It's even more complex.

Take those 80 shots from the point, unscreened they probably have a 1% chance of scoring.

But if you have traffic in front of the net, while rebounds and put backs will be accounted for, the increase in scoring probability from screening a goalie is not accounted for, making those point shots more valuable than raw stats would suggest.

In this case, blocking shots have more value when opposing teams put big bodies in front of the net, since it reduces the number of shots that get to the net, avoiding both screened shots that go in and shots that are deflected or cause rebounds.

I wonder if this is the kind of detail that team proprietary databases explore - it would require a couple interns to detail outcomes from every shot attempt, etc. So amateurs simply don't have the resources to do this kind of detailed analysis.
 
All xGF models are not created the same. If you really want to have this conversation, you have to dig into each model and what it does differently. For example, some use flat variable modifiers for rebounds while at least one uses a scale based on time gaps. Some also change those factors based upon the angle change of the resulting rebound. You can get significant swings.
 
Last edited:
It's even more complex.

Take those 80 shots from the point, unscreened they probably have a 1% chance of scoring.

But if you have traffic in front of the net, while rebounds and put backs will be accounted for, the increase in scoring probability from screening a goalie is not accounted for, making those point shots more valuable than raw stats would suggest.

In this case, blocking shots have more value when opposing teams put big bodies in front of the net, since it reduces the number of shots that get to the net, avoiding both screened shots that go in and shots that are deflected or cause rebounds.

I wonder if this is the kind of detail that team proprietary databases explore - it would require a couple interns to detail outcomes from every shot attempt, etc. So amateurs simply don't have the resources to do this kind of detailed analysis.
For sure, super complex based on how hockey is played. Not one shot is the same as any other shot. You have to work on the law of averages so much more than other sports.

I don't think other teams have that level of detail -- I don't know for sure tho. I remember JFresh did a comparison of xG model based on the RTSS dataset and one based on proprietary databases. I don't recall the specifics, but I recall thinking there wasn't a huge difference in conclusions. There may have been, likely were, some interesting difference in the minutia.

I would love for the NHL to adopt a puck with the exact same weight and specs that has a tracking device in it. I'd also love for every single player to have GPS units/tracking systems (maybe they do and I'm not aware of it). Obviously new-age cameras can do that as well. If you were able to generate those two datasets for comparison, the level of detailed analysis that you could get would be amazing.
 
  • Wow
Reactions: Beef Invictus
For sure, super complex based on how hockey is played. Not one shot is the same as any other shot. You have to work on the law of averages so much more than other sports.

I don't think other teams have that level of detail -- I don't know for sure tho. I remember JFresh did a comparison of xG model based on the RTSS dataset and one based on proprietary databases. I don't recall the specifics, but I recall thinking there wasn't a huge difference in conclusions. There may have been, likely were, some interesting difference in the minutia.

I would love for the NHL to adopt a puck with the exact same weight and specs that has a tracking device in it. I'd also love for every single player to have GPS units/tracking systems (maybe they do and I'm not aware of it). Obviously new-age cameras can do that as well. If you were able to generate those two datasets for comparison, the level of detailed analysis that you could get would be amazing.
While I think AI is grossly overrated, creating and analyzing these kind of data sets is where it will be useful - imagine trying to organize and categorize all these data - eyes would soon glaze over.
 
  • Like
Reactions: VladDrag
skipped through the game. As the last 2. The flyers are boring, slow, without offenisve instincts, pls trade frost, pls relocate the franchise!

THX
 
While I think AI is grossly overrated, creating and analyzing these kind of data sets is where it will be useful - imagine trying to organize and categorize all these data - eyes would soon glaze over.
I use AI to help in organizing data in my job routinely, and it’s a huge help.

It’s great because more time can be spent on analyzing, and not setting up the data.
 

Ad

Ad