[Game Takes] Flames 4 Predators 3

[Bingo]

Originally Posted by Enoch Root

Well, the standings suggest some teams are bad consistently.

When we see bad records with high goals against, but the stats say their goals against shouldn't have been that high, there are two possible reasons: one, it's just been luck so far, and it should even out, or two, the stat isn't capturing what's happening.

Both are possible, but to suggest the latter isn't likely, is where we are going to disagree.

Most of the time standings are very similar to expected goal sorts. There are always teams that beat the odds in both directions; which you'd expect in a game like hockey, but they certainly seem indicative.

If you rank points percentage and xGF% and then do the differential this morning there are only 5 major outliers in the rankings.

Ottawa and Pittsburgh have better underlying numbers than their standing. Toronto, Winnipeg and Vegas have better results than it appears they deserve.

Toronto and Winnipeg have top 5 goaltending. Pittsburgh is bottom 3.

In your example I think it's three things ...

1) the stat can always improve
2) the team probably has ####ty goaltending
3) the team may be on the wrong side of luck

[DeluxeMoustache]

The main thing for me is giving a goalie a pass or fail on a single game based on xGA.

There are simply not enough shots taken in a game for the stat to be statistically significant

And I would say again, the biggest differentiating determinant of whether a shot goes in is where it is placed, and we can agree that just simply isn’t factored in.

Say a puck that crossed the royal road and is a wrist shot has ~33 percent chance of going in (by contrast, a point shot has a ~2 percent chance of going in)

Say in a game you have 3 shots that meet that description - cross the royal road, wrist shot (and that’s actually a lot more specific than most xGA models. More like time since previous event (a logged data point) which indicates it was passed). You expect 1 goal

It is very easy for all 3 shots to be drilled in to the goalie’s pads or logo. It is also pretty easy to have 3 well placed shots taken on any given night

That is a very small sample size of shots of that type. Sure, the model is good because it is based on 3000 shots of that type.. 2000 were saved, 1000 went in, so we are confident, right? Big deal. In a particular game, there were 3 shots of that type taken, the xGA is 1, and the potential outcomes for goals are either 0, 1, 2 or 3.

So two or more out of 3 are accurate / well placed shots and the goalie gets a ‘failing grade’, whereas the goalie who faces 3 shots into his logo ‘passes the test’

As for ‘quality of shooter’, that is of limited usefulness. Consider Pachal’s goal last game. How many shots does he get? Not a lot. How many are like that? Not a lot.

I agree in a sense with the idea that at this point, you have to use the eye test. It is fun to inform yourself as to what a model predicts, but more importantly use your head and do a sanity check, because there are insufficient data points to make a single game statistically significant

[Bingo]

Originally Posted by DeluxeMoustache

The main thing for me is giving a goalie a pass or fail on a single game based on xGA.

There are simply not enough shots taken in a game for the stat to be statistically significant

And I would say again, the biggest differentiating determinant of whether a shot goes in is where it is placed, and we can agree that just simply isn’t factored in.

Say a puck that crossed the royal road and is a wrist shot has ~33 percent chance of going in (by contrast, a point shot has a ~2 percent chance of going in)

Say in a game you have 3 shots that meet that description - cross the royal road, wrist shot (and that’s actually a lot more specific than most xGA models. More like time since previous event (a logged data point) which indicates it was passed). You expect 1 goal

It is very easy for all 3 shots to be drilled in to the goalie’s pads or logo. It is also pretty easy to have 3 well placed shots taken on any given night

That is a very small sample size of shots of that type. Sure, the model is good because it is based on 3000 shots of that type.. 2000 were saved, 1000 went in, so we are confident, right? Big deal. In a particular game, there were 3 shots of that type taken, the xGA is 1, and the potential outcomes for goals are either 0, 1, 2 or 3.

So two or more out of 3 are accurate / well placed shots and the goalie gets a ‘failing grade’, whereas the goalie who faces 3 shots into his logo ‘passes the test’

As for ‘quality of shooter’, that is of limited usefulness. Consider Pachal’s goal last game. How many shots does he get? Not a lot. How many are like that? Not a lot.

I agree in a sense with the idea that at this point, you have to use the eye test. It is fun to inform yourself as to what a model predicts, but more importantly use your head and do a sanity check, because there are insufficient data points to make a single game statistically significant

I'd agree with most of that.

The only thing I'd add is the danger of ignoring stats and applying an eye test if you have a bias.

Take a Huberdeau hater that is shown the player was #1 on the team in both expected goal and high danger splits in a hockey game, but then tries to play down those summaries by dismissing the level of danger.

Starts to get foolish.

[GioforPM]

Originally Posted by Bingo

I'd agree with most of that.

The only thing I'd add is the danger of ignoring stats and applying an eye test if you have a bias.

Take a Huberdeau hater that is shown the player was #1 on the team in both expected goal and high danger splits in a hockey game, but then tries to play down those summaries by dismissing the level of danger.

Starts to get foolish.

True, except sometimes it works in reverse. For example, in Nashville I thought Huberdeau had a decent game. Yet he was victimized by mistakes by Kadri and Andersson while he was on the ice in proper position. The stats will mark him down for that, merely because he shared the ice with players who made mistakes. And a Huberdeau hater will say he got caved in, assuming the stats support it (I don’t know what they were overall).

[Bingo]

Originally Posted by GioforPM

True, except sometimes it works in reverse. For example, in Nashville I thought Huberdeau had a decent game. Yet he was victimized by mistakes by Kadri and Andersson while he was on the ice in proper position. The stats will mark him down for that, merely because he shared the ice with players who made mistakes. And a Huberdeau hater will say he got caved in, assuming the stats support it (I don’t know what they were overall).

100%

If a player consistently plays with terrible defensive players his numbers will get dragged down.

Usually if that's happening the coach changes the lines up though, and you get back to better numbers in a larger sample size.