5on5, Almost all top 10 teams in xGF% score less actual goals than their xGF. Yet, 5 of the bottom 10 teams in xGF% score more goals than their xGF.

Bjornar Moxnes

Stem Rødt og Felix Unger Sörum
Oct 16, 2016
11,756
5,769
Troms og Finnmark
1743470229105.png

Only Winnipeg outscores their xGF, but by .51, so essentially they score the exact same. LA is in the same boat, scores at .74 less, so essentially the same. Many teams like Edmonton, Florida, and Carolina score WELL BELOW their xGF.

1743470370366.png

Conversely, Chicago, Seattle, Buffalo, New York Rangers, and Toronto all score above their xGF. Buffalo has scored a staggering 24.79 goals above their expected, with many 4th liners and defensive dman scoring more than their xGF.

What exactly is the reason for how the top teams in xGF% perform far worse in actual goals scored, and vice versa?
 
What are xGF% and xGF?
Stats that don't mean what people think, yet are cited as evidence anyway.

Stats are just an interesting set of collected digits without context. People want them to have predictive relevance, but they don't. In a sport with as much variance as hockey, advanced stats are anecdotal at best, more commonly misleading.
 
nvm misread but still should filter by 5v5 close

Top teams getting sloppy holding a 3 goal lead and basement teams finding a breakaway in that situation isn’t really indicative of a trend
 
Last edited:
expected goals for percentage and expected goals for. xGF% is xGF and xGA (goals against) combined and then divided.
So work with me here because I'd like to understand. So xGF is a value based on a range of probabilities assigned from a certain shot type, from a certain location on the ice, targeted to a certain area of the net? xGF% then is a ratio derived from the summation of quality shots taken on offense to the difference in the summation of quality shots given up on defense? So if a team has a worse xGF%, couldn't that mean they just give up more xGA than xGF?
 
Last edited:
5 of the bottom ten teams on each side doesn't really support your argument. I think there are certain teams that take more shots from locations that xG overvalue. Carolina is a team that always seems to be massively underperforming their xGF.
 
  • Like
Reactions: Elysian
The bottom teams being 5 out of 10 is what you would expect. The top teams being 1 out of 10 is a bit of a surprise. But you’ve run your experiment once. Run it for the last 10 years and see how it does. You could also include the middle 12 teams as a third category.
 
5 of the bottom ten teams on each side doesn't really support your argument. I think there are certain teams that take more shots from locations that xG overvalue. Carolina is a team that always seems to be massively underperforming their xGF.
Huh, I'm not making any arguments. I'm literally why 5 of the bottom 10 teams are outperforming, including teams like Chicago and Buffalo, yet every team in the top 10 in generation are underperforming by a lot or essentially dead even.
 
Huh, I'm not making any arguments. I'm literally why 5 of the bottom 10 teams are outperforming, including teams like Chicago and Buffalo, yet every team in the top 10 in generation are underperforming by a lot or essentially dead even.

How often are the two groups trailing vs leading, if the leading team starts taking it easy and rolling their bottom lines and pairs sitting on a 5-2 lead in the third, who is more likely to score a pair on a couple breakaways that don’t drive xGF as much as a full period of aggressive cycling and tight systemic play?
 
Isn't that exactly what you'd expect? As far as I know, expected goals are not a stat like PDO that's expected to normalize over time. Expected goals are just a rough measure of the quantity and quality of scoring chances your team is generating. I imagine it would be totally possible, and even likely, for a team to consistently outperform xG data with actual goals, or the reverse. With great goaltending, or really gifted finishers, for example. And I would expect that outliers for that stat (for whom it's particularly low or high) employ tactics that skew the stat one way or the other; isn't that what we've clearly observed with the Canes for years? And I'm sure there's a bunch of other factors that could systematically affect xG.

Then again, I'm not an expert on advanced stats, so maybe I'm completely off.
 
Last edited:
I've always found advanced stats to be shorthand for:
"Team X is in the bottom of the league, but is actually the best team in the league"
Good advanced stats and bad results are not always indication that team is good (sometimes crappy sts and goaltending are really crappy sts and goaltending, not a proof that team is " snakebitten", or "unlucky"), but if it's other way around...

If a team has crappy advanced stats and good results despite it, its almost always a major red flag.
 
Assume that xGF is affected more strongly by the team's skill than GF. As such, the top teams having higher xGF than GF, and bottom teams having lower xGF than GF, is only logical because xGF has less noise to it than GF.

Here, xGoals, of course, have calibration issues(total expected goals are different from total goals), so comparing them one-to-one isn't very fruitful. But this is exactly how the trends should be, if you look at how the stats are meant to work.
 
Stats that don't mean what people think, yet are cited as evidence anyway.

Stats are just an interesting set of collected digits without context. People want them to have predictive relevance, but they don't. In a sport with as much variance as hockey, advanced stats are anecdotal at best, more commonly misleading.
What's this nonsense? They absolutely do have predictive relevance. Heck, you could build your own machine learning model right now and verify that very thing.
 
It just shows the stat is still new and faulty. Great premise - but not super useful. The most frustrating part of it is how it’s become THE go to stat for a lot of analysts. It’s still guesswork.

As a Jets fan, I hate how often expected goals saved is quoted. Like - the actual stats tell a good enough story - why are we starting with expected goals saved?

Really like the premise. Don’t like how it’s being aired almost as fact. And it needs some work in how it’s calculated IMO.
 
  • Like
Reactions: Raccoon Jesus
I think it just shows that "expected" is developing and will ALWAYS lag what is actually happening currently. It is extremely useful and still is 70-90% predictive. I'm not sure if they have introduced any additional variables despite mainly being location based shooting probabilities with some modifiers like "rush y/n".

You are trying to predict/tell the future with these things and it will never be perfect (see laws of the universe). Only morons expect it to be, so don't be one of those. It is also going to have a normal distribution (ie bell curve) shape to it with close to half the teams above and below that expected GF amount.

To me, the main explanation of the variation (biggest error slice of the pie) is that xGF% vs GF% assumes an average NHL player is taking the shot in that location (against average goalie too), but that is never the case. Some teams are better at optimizing their shooters and also have much better talent taking those shots or in net.

I also may be lagging in my knowledge of the subject (5-10years) as most NHL analytic sites have been taken down or behind paywalls now. Please correct me where needed. Player tracking is much better now and should be a large part of the complex models.
 
View attachment 1003492
Only Winnipeg outscores their xGF, but by .51, so essentially they score the exact same. LA is in the same boat, scores at .74 less, so essentially the same. Many teams like Edmonton, Florida, and Carolina score WELL BELOW their xGF.

View attachment 1003499
Conversely, Chicago, Seattle, Buffalo, New York Rangers, and Toronto all score above their xGF. Buffalo has scored a staggering 24.79 goals above their expected, with many 4th liners and defensive dman scoring more than their xGF.

What exactly is the reason for how the top teams in xGF% perform far worse in actual goals scored, and vice versa?

Feel like if you look back to other seasons, would be similar.
 
Assume that xGF is affected more strongly by the team's skill than GF. As such, the top teams having higher xGF than GF, and bottom teams having lower xGF than GF, is only logical because xGF has less noise to it than GF.

Here, xGoals, of course, have calibration issues(total expected goals are different from total goals), so comparing them one-to-one isn't very fruitful. But this is exactly how the trends should be, if you look at how the stats are meant to work.
How are the stats meant to work? (Sincerely asking as I don’t know. All I know is it takes in a weighted avg of shot types, and situation (pp, 5v5) ).

The problem you just sited (can’t compare them one to one) is exactly what everyone does.

I one hundred percent agree with you on the calibration issues and the gap from skill. A shot in the slot from MacKinnon is given the same weight as a slot shot from vlad namestnikov.

I would say if we have built a stat that can’t be compared it’s not as useful a stat as it can be. What other stat is like that?
 

Ad

Ad