[Game Takes] Flames 2 Predators 0

[DeluxeMoustache]

I think xGF / xGA is a very useful metric for evaluating skaters, especially over a larger body of work. It has bias towards total shot differential, refined by where chances are coming from. It decently reflects the overall balance of play

xGA is less useful imo for evaluating goalies at face value (without deliberation) because of what it doesn’t measure

[Enoch Root]

Sure, it's better than some other stats, and with larger sample sizes (like a whole season), it isn't terrible.

But all the criticisms of it are accurate. And what I really hate about it is how people interpret it. Especially for a single game. Because expected goals for and against, for most players, in any given game, is only about 1 or 2. So the percentages become stupid. For example:

Let's say a line has been given defensive responsibilities to shut the other team's top line down, and they do a great job, making the whole night a low event game, where both sides had an expected goals for of 1. That would mean everyone's percentages would be 50% - no problem. Now, let's imagine that at some point during the game (while this line is on the ice), a defenseman coughs up the puck and gives the other team a breakaway. That would add about .4 to the expected goals for for the other team. Suddenly, because of one play that had nothing to do with anyone else, the expected goals against go from 1 to 1.4. That would make the %GF/A drop from 50% to 41.6%. One random play.

And then people would look at that and say player X had a %GF/A of 41% - their line got caved in.

There isn't enough data from a single game, to make the percentages meaningful. And single events impact them far too much. But because they all get converted to percentages, people look at the percentages for the whole team and think: this guy had a great game, and that guy had a terrible game. In reality, this stat is completely incapable of telling us that with any useful accuracy, using small sample sizes - but everyone does it anyway, and treat it like gospel.

[edslunch]

Originally Posted by Enoch Root

Sure, it's better than some other stats, and with larger sample sizes (like a whole season), it isn't terrible.

But all the criticisms of it are accurate. And what I really hate about it is how people interpret it. Especially for a single game. Because expected goals for and against, for most players, in any given game, is only about 1 or 2. So the percentages become stupid. For example:

Let's say a line has been given defensive responsibilities to shut the other team's top line down, and they do a great job, making the whole night a low event game, where both sides had an expected goals for of 1. That would mean everyone's percentages would be 50% - no problem. Now, let's imagine that at some point during the game (while this line is on the ice), a defenseman coughs up the puck and gives the other team a breakaway. That would add about .4 to the expected goals for for the other team. Suddenly, because of one play that had nothing to do with anyone else, the expected goals against go from 1 to 1.4. That would make the %GF/A drop from 50% to 41.6%. One random play.

And then people would look at that and say player X had a %GF/A of 41% - their line got caved in.

There isn't enough data from a single game, to make the percentages meaningful. And single events impact them far too much. But because they all get converted to percentages, people look at the percentages for the whole team and think: this guy had a great game, and that guy had a terrible game. In reality, this stat is completely incapable of telling us that with any useful accuracy, using small sample sizes - but everyone does it anyway, and treat it like gospel.

Pretty much like plus/minus. Players need to be consistently good or bad to get large numbers but in the middle it doesn't mean much.

Edit: I do like it more than on-ice PDO. At a player can have a direct impact on xGA, but much less on GA which is greatly dependent on the goalie. PDO should be a team-only stat IMO.

[Bingo]

Originally Posted by Enoch Root

Sure, it's better than some other stats, and with larger sample sizes (like a whole season), it isn't terrible.

But all the criticisms of it are accurate. And what I really hate about it is how people interpret it. Especially for a single game. Because expected goals for and against, for most players, in any given game, is only about 1 or 2. So the percentages become stupid. For example:

Let's say a line has been given defensive responsibilities to shut the other team's top line down, and they do a great job, making the whole night a low event game, where both sides had an expected goals for of 1. That would mean everyone's percentages would be 50% - no problem. Now, let's imagine that at some point during the game (while this line is on the ice), a defenseman coughs up the puck and gives the other team a breakaway. That would add about .4 to the expected goals for for the other team. Suddenly, because of one play that had nothing to do with anyone else, the expected goals against go from 1 to 1.4. That would make the %GF/A drop from 50% to 41.6%. One random play.

And then people would look at that and say player X had a %GF/A of 41% - their line got caved in.

There isn't enough data from a single game, to make the percentages meaningful. And single events impact them far too much. But because they all get converted to percentages, people look at the percentages for the whole team and think: this guy had a great game, and that guy had a terrible game. In reality, this stat is completely incapable of telling us that with any useful accuracy, using small sample sizes - but everyone does it anyway, and treat it like gospel.

They can improve for sure.

But expected goals are built over a game by events. If you have 10 chances within x feet of the net you get x% of a goal for added to your total.

So in a single game if you create 14 good things and 11 bad things you may end up at 1.11 vs 0.74 but that doesn't change that you had 14 of 25 things go your way.

One single game is one game though. So no one game doesn't mean good player or bad player. But it does mean good split in events game vs bad split in events game.

[Strange Brew]

Originally Posted by Bingo

Looked it up ... Backlund is 23rd in the last three years in the NHL if xGF%, Coleman at 28th.

Yeah the bigger the sample size the more you can see that this stat is measuring something impactful.

For the same reason, I don't hate plus/minus with large sample sizes. Bobby Orr was +124 in his best season. I don't think it was luck that was the ice so tilted in the Bruins favor when he was on the ice.