Sperss1997
Registered User
Thought last year was bad after christmas - but this is crazy
GF% allows you to win hockey games, xGF% allows you to still b**** about the team even after you win the gameI keep forgetting what that stat means. @Whileee or someone can please explain to me again?(i feel so stupid )
GF% allows you to win hockey games, xGF% allows you to still b**** about the team even after you win the game
Obviously. But if we look at all games in 2019 its pretty scary. This season is just started, I know, but a xG% at 41 is really ugly!There’s a big difference between having a bad stretch of 15 games vs carrying these levels over a whole season...
It means that those numbers are more easily skewed by outliers. Take out the San Jose game and we probably aren't even on that list.Obviously. But if we look at all games in 2019 its pretty scary. This season is just started, I know, but a xG% at 41 is really ugly!
@Duke749
That means that every time WPG create chances for 41 goals - opponents has created chances for 59 goals
This was really great, WhileeCorsi is a shot attempt, from wherever, regardless of the danger of scoring from that attempt.
Expected goal metrics take shot attempts and estimate the probability of a goal for each shot attempt. This is based on the type and location of shot attempts, so a tap in opportunity has a much higher expected goal (xG) value than a shot from the point.
The Corsi for percentage (CF%) is the percentage of all shot attempts by a given team. So if each team has the same number the CF% is 50%. The Jets have had about 49% of the Corsi attempts in their games, with the opposition having 51%.
The expected goals for percentage xGF% estimates the total expected goals for and against a team and then converts that into a percentage for the team in its games. The Jets have generated about 42% of all of the expected goals this season, whereas the opposition has generated 58% summed across the Jets games.
So far this season, the Jets have been mediocre at CF%, but really low at xGF%. That suggests that they are close to even with their opponents in overall shot attempts, but much lower in expected goals because the shot attempts they take are lower danger and the shots they give up are higher danger than opponents.
The CF% and xGF% metrics are valued because they have been shown to be reasonably good at predicting future goal differentials (i.e. over the rest of the season), which translates into winning percentage.
But there are some caveats that should be considered in this case. First, it is very unusual for the CF% and xGF% to deviate as much as they have for the Jets this season. It seems likely that they will converge with a bigger sample size and reduced statistical error. Also, the Jets will almost certainly adjust, to translate more of their puck possession and shot attempts into higher danger shots, and higher expected goals.
While the current status of the Jets CF and xG metrics are worrisome in terms of predicting future results, it's useful to remember that the predictive performance of xGF% isn't that much better that CF%, and in fact CF% appears to be a better predictor in the first 20 games. However, the Jets performance hasn't been good enough and they'll need to start turning it around to have confidence that they won't need to rely too much on luck or goaltending.
View attachment 275565
Absolutely. If you look at that list, the teams with 82 games samples have little difference between CF% and xGF%, but the Jets and Sharks 2019-2020 teams have a CF% 7 points higher.There’s a big difference between having a bad stretch of 15 games vs carrying these levels over a whole season...
My work involves a lot of statistical analyses, usually much more complex than hockey "advanced stats". I am a strong advocate of statistical analysis, but I'm very wary about overinterpretation and like to see a range of quantitative and qualitative observations to understand complex processes. Hockey is fascinating because it involves a lot of relatively rare discrete events (goals) determining outcomes. I see shot metrics as an important signpost, but professional coaching staffs spend thousands of hours breaking down video to try to understand how possession, shot attempts, scoring chances, and goals occur. In the end, I'm not that persuaded that coaches differ that much. Some are clueless or lack basic social intelligence, but most know what they are doing at least as much as other coaches. I think the pendulum has swung too far in the direction of over-emphasizing basis shot metrics, including xG, especially at the individual level at the exclusion of macro level context. Berube didn't become a bad coach in the off-season, but the Blues' shot metrics are way below the latter half of last season.This was really great, Whilee
In my line of work, stats are EVERYTHING. We rely on raw data metrics almost exclusively, because they don't lie, (of course with sample size being extremely important). Raw metrics are very useful as long as processes for dissemination are sound.
My main concern with advanced stats in hockey is that data is often subjective, 'fiddled with' and using theories and formulas that are not necessarily vetted by scientific methodology or enough real life application to truly measure accuracy.
For example, one shot from the slot is not like the other:
- How hard was the shot?
- How quick was the release?
- How accurate was it?
- Was there a screen?
- Was there an open player at the side of the net who the goalie thought might get a slap pass or tip, splitting the goalie's focus?
- Who is determining how dangerous that shot was? How experienced are they? How would one analyst rate the shot vs. another?
All of these impact the purity of the data and call into question it's accuracy.
That's why I always take these stats with a grain of salt and defer to my eyes when things don't add up.
Again, I do think they're useful when used and interpreted correctly.
My main issue with advanced stats in hockey is how they're used. The good news is, the people who REALLY need them (hockey coaches and management) seem to understand where they fall in terms of weight when making decisions.
Awesome! Also, I always forget your third 'e'My work involves a lot of statistical analyses, usually much more complex than hockey "advanced stats". I am a strong advocate of statistical analysis, but I'm very wary about overinterpretation and like to see a range of quantitative and qualitative observations to understand complex processes. Hockey is fascinating because it involves a lot of relatively rare discrete events (goals) determining outcomes. I see shot metrics as an important signpost, but professional coaching staffs spend thousands of hours breaking down video to try to understand how possession, shot attempts, scoring chances, and goals occur. In the end, I'm not that persuaded that coaches differ that much. Some are clueless or lack basic social intelligence, but most know what they are doing at least as much as other coaches. I think the pendulum has swung too far in the direction of over-emphasizing basis shot metrics, including xG, especially at the individual level at the exclusion of macro level context. Berube didn't become a bad coach in the off-season, but the Blues' shot metrics are way below the latter half of last season.
In the end, shot metrics describe things, but don't do a good job of explaining them. They are also predictive, but not as predictive as is sometimes implied. They are often misused or misunderstood to make judgements on players that are overly simplistic.
This was really great, Whilee
In my line of work, stats are EVERYTHING. We rely on raw data metrics almost exclusively, because they don't lie, (of course with sample size being extremely important). Raw metrics are very useful as long as processes for dissemination are sound.
My main concern with advanced stats in hockey is that data is often subjective, 'fiddled with' and using theories and formulas that are not necessarily vetted by scientific methodology or enough real life application to truly measure accuracy.
For example, one shot from the slot is not like the other:
- How hard was the shot?
- How quick was the release?
- How accurate was it?
- Was there a screen?
- Was there an open player at the side of the net who the goalie thought might get a slap pass or tip, splitting the goalie's focus?
- Who is determining how dangerous that shot was? How experienced are they? How would one analyst rate the shot vs. another?
All of these impact the purity of the data and call into question it's accuracy.
That's why I always take these stats with a grain of salt and defer to my eyes when things don't add up.
Again, I do think they're useful when used and interpreted correctly.
My main issue with advanced stats in hockey is how they're used. The good news is, the people who REALLY need them (hockey coaches and management) seem to understand where they fall in terms of weight when making decisions.
If ANYONE said to you Stu was great they had to be drunk , BIG TIME .Nice post and your points are well taken. These things are definitely a work in progress.
I am not singling you out but one of the problems I have with the traditional “eye test” is that I find it even more flawed.
full confession I am guilty of some if not all of the symptoms listed below:
confirmation bias
Style preference bias
Big event impact
Anchoring effect
Only able to gather micro sample sizes especially on visiting teams.
****ty stupid eyes.
Ok I am having fun . IMO there are some great eye test fans on our site and when they break down games I have watched they “seem” spot on to what I have witnessed (not that that is a ringing endorsement) but when I talk to the average fan on the street their eye test hot takes are frightening.
In summary I don’t accept all stats as gospel especially in small sample sizes and I think it’s all a work in progress. However “on average” I am much more sceptical of the average hockey fans eye test.
If I had a dime for every “I played hockey at a high level” Pegger who told me Stu was great (fill in all the Don Cherry reasons) and Toby sucked I would be a much richer man.
You lost me after Corsi is . But great post at trying to get a dumbass like me to try and understand it .Corsi is a shot attempt, from wherever, regardless of the danger of scoring from that attempt.
Expected goal metrics take shot attempts and estimate the probability of a goal for each shot attempt. This is based on the type and location of shot attempts, so a tap in opportunity has a much higher expected goal (xG) value than a shot from the point.
The Corsi for percentage (CF%) is the percentage of all shot attempts by a given team. So if each team has the same number the CF% is 50%. The Jets have had about 49% of the Corsi attempts in their games, with the opposition having 51%.
The expected goals for percentage xGF% estimates the total expected goals for and against a team and then converts that into a percentage for the team in its games. The Jets have generated about 42% of all of the expected goals this season, whereas the opposition has generated 58% summed across the Jets games.
So far this season, the Jets have been mediocre at CF%, but really low at xGF%. That suggests that they are close to even with their opponents in overall shot attempts, but much lower in expected goals because the shot attempts they take are lower danger and the shots they give up are higher danger than opponents.
The CF% and xGF% metrics are valued because they have been shown to be reasonably good at predicting future goal differentials (i.e. over the rest of the season), which translates into winning percentage.
But there are some caveats that should be considered in this case. First, it is very unusual for the CF% and xGF% to deviate as much as they have for the Jets this season. It seems likely that they will converge with a bigger sample size and reduced statistical error. Also, the Jets will almost certainly adjust, to translate more of their puck possession and shot attempts into higher danger shots, and higher expected goals.
While the current status of the Jets CF and xG metrics are worrisome in terms of predicting future results, it's useful to remember that the predictive performance of xGF% isn't that much better that CF%, and in fact CF% appears to be a better predictor in the first 20 games. However, the Jets performance hasn't been good enough and they'll need to start turning it around to have confidence that they won't need to rely too much on luck or goaltending.
View attachment 275565
I am not good with these stats. Does this mean we are bad and shouldn't be winning? Or are we winning and once these stats get better or worse it will level us out?
Or something different I dont know lol thanks in advance
Obviously. But if we look at all games in 2019 its pretty scary. This season is just started, I know, but a xG% at 41 is really ugly!
@Duke749
That means that every time WPG create chances for 41 goals - opponents has created chances for 59 goals
This is a very good post from a statistically savvy fan, and should be essential reading for people on both sides of the debate.My work involves a lot of statistical analyses, usually much more complex than hockey "advanced stats". I am a strong advocate of statistical analysis, but I'm very wary about overinterpretation and like to see a range of quantitative and qualitative observations to understand complex processes. Hockey is fascinating because it involves a lot of relatively rare discrete events (goals) determining outcomes. I see shot metrics as an important signpost, but professional coaching staffs spend thousands of hours breaking down video to try to understand how possession, shot attempts, scoring chances, and goals occur. In the end, I'm not that persuaded that coaches differ that much. Some are clueless or lack basic social intelligence, but most know what they are doing at least as much as other coaches. I think the pendulum has swung too far in the direction of over-emphasizing basis shot metrics, including xG, especially at the individual level at the exclusion of macro level context. Berube didn't become a bad coach in the off-season, but the Blues' shot metrics are way below the latter half of last season.
In the end, shot metrics describe things, but don't do a good job of explaining them. They are also predictive, but not as predictive as is sometimes implied. They are often misused or misunderstood to make judgements on players that are overly simplistic.
It means that those numbers are more easily skewed by outliers. Take out the San Jose game and we probably aren't even on that list.
Bingo.
That one game was super awful. And, that game pushes the stat that makes this article somewhat compelling.
https://www.tsn.ca/connor-hellebuyck-keeping-the-winnipeg-jets-afloat-1.1396015
Has Helle been great? Yes, undeniably. Do we need improvements to our D? Yes, undeniably.
+++++++++++
Now, does this conundrum constitute a lightning bolt of revelation?
No. It simply means that Dostoevsky has found a new host in Travis Yost.
Same thing goes for the opponents xG share - has very very little influence, especially as the sample size increasesWhich, by the way, doesn’t factor the skill of the player. Whether Laine takes the shot or Lowry, WHO CARES (according to this metric).