PlayerToBeNamedLater
Well-known member
Hell, I argue with Mirtle on twitter all the time about how it's being used.
Advanced Stats Thread
- Thread starter PlayerToBeNamedLater
- Start date
Montana
Champion
...and no one here has argued otherwise.
Hell, the only "argument" is msun incorrectly assuming people here don't know what they're referring to when they use the term "possession".....when they quite clearly do.
Last edited:
Montana
Champion
What's the correlation between goal ranking and final season standing?
What's the correlation between sv% ranking and final season standing?
None are great predictors on their own, (and people would be just as incorrect to try and use them as such) but all three together trump any of them individually. Adding Corsi/Fenwick improves the overall equation.
It's not a be all/end all variable....but it's a variable that when used right, is a worthwhile one.
PlayerToBeNamedLater
Well-known member
Of the top 16 save percentage teams last year, 12 made the playoffs.
Of the top 16 goal scoring teams last year, 10 made the playoffs.
Of the top 16 fenwick teams last year, 12 made the playoffs.
All these stats are good indicators of a successful team, but none of them on their own can consistently predict success from year to year. But combined, you might have something.
Of the top 16 goal scoring teams last year, 10 made the playoffs.
Of the top 16 fenwick teams last year, 12 made the playoffs.
All these stats are good indicators of a successful team, but none of them on their own can consistently predict success from year to year. But combined, you might have something.
zeke
Well-known member
actually, goal differential is by far and away the best predictor, with a correlation of over 80%.
obviously.
it's just not as repeatable as corsi.
Montana
Champion
not sure why you'd word it like that, when I made zero mention of goal differential........but to your point, I agree entirely, goal differential has a strong correlation, not unlike run differential in baseball.
zeke
Well-known member
which is probably worth talking about, actually.
because many of the teams that Corsi claims it is able to spot "overachieving" and who then regress, are also teams that seem to be obviously overachieving based on other metrics like goal differential - that you can predict for obvious regression without looking at the possession stats.
In fact, I could point out examples of where goal differential showed an obvious future regression that somehow Fenwick failed to see entirely.
I'm thinking of the 11-12 Florida Panthers.
They surprised everyone by finishing in first place in the southeast with 94pts. It wasn't a tremendous point total, being 6th best in the East and 14th overall, but it was a solid season.
Looking at their Fenwick Close, they ranked 13th overall in the league, and 7th in the East - Fenwick told us that Florida was full marks for their 94pts, and that it was totally sustainable.
Of course, if we took a gander at the much more strongly win correlated Goal Differential (like I remember doing at the time), we saw that they had one of the worst goal differentials in the league at -0.23, which ranked 10th in the east and 21st overall.
The goal differential pretty clearly told us that team was a fluke who couldn't possibly keep winning, even though Fenwick indicated they should be able to.
And the next year, of course, the Panthers went right back to sucking something fierce, as we all expected they would based on the pre-possession stats we had.
So there's one example, at least, of where the posession stats were lying to us, and the much more strongly correlated stat trumped it quite easily in predicting future wins.
At some point I think I'm going to have to line up some sort of comparison between fenwick and goaldif predictions, and see which one actually works better in real life, and how many examples Corsi/fenwick is taking credit for that goal differential would explain even better, and how many times fenwick completely misses a future regression that goal differential makes obvious.
because many of the teams that Corsi claims it is able to spot "overachieving" and who then regress, are also teams that seem to be obviously overachieving based on other metrics like goal differential - that you can predict for obvious regression without looking at the possession stats.
In fact, I could point out examples of where goal differential showed an obvious future regression that somehow Fenwick failed to see entirely.
I'm thinking of the 11-12 Florida Panthers.
They surprised everyone by finishing in first place in the southeast with 94pts. It wasn't a tremendous point total, being 6th best in the East and 14th overall, but it was a solid season.
Looking at their Fenwick Close, they ranked 13th overall in the league, and 7th in the East - Fenwick told us that Florida was full marks for their 94pts, and that it was totally sustainable.
Of course, if we took a gander at the much more strongly win correlated Goal Differential (like I remember doing at the time), we saw that they had one of the worst goal differentials in the league at -0.23, which ranked 10th in the east and 21st overall.
The goal differential pretty clearly told us that team was a fluke who couldn't possibly keep winning, even though Fenwick indicated they should be able to.
And the next year, of course, the Panthers went right back to sucking something fierce, as we all expected they would based on the pre-possession stats we had.
So there's one example, at least, of where the posession stats were lying to us, and the much more strongly correlated stat trumped it quite easily in predicting future wins.
At some point I think I'm going to have to line up some sort of comparison between fenwick and goaldif predictions, and see which one actually works better in real life, and how many examples Corsi/fenwick is taking credit for that goal differential would explain even better, and how many times fenwick completely misses a future regression that goal differential makes obvious.
zeke
Well-known member
you said "goal ranking".....I thought you were talking aboiut both goals for and against.
Montana
Champion
nope....just meant goal rankings.
...but I think you touch on an accurate point. A metric that combined Sv%, goal differential, and possession would likely be the single most accurate predictor of success than anything we're currently using/discussing.
edit: although given a second though, while it may work on it's own as a simple tool.....I think I'd prefer sv%/GF/Corsi. They might come out with somewhat similar results, with the latter being a little more accurate.
Last edited:
Why don't y'all try and come up with a better advanced stat? Seems like this board has the right ppl to do this sort of thing. The effort to just get the fantasy points to an exact measure shows level of detail ppl would put in.
Ps: if this happens, def should be partly named after me for throwing out the idea haha
Ps: if this happens, def should be partly named after me for throwing out the idea haha
Montana
Champion
Fenwick doesn't "tell" us that though.......idiots in the media might have tried to argue that, but everyone here is long in agreement that Fenwick and Fenwick alone doesn't really tell us anything that strongly all by itself.
I feel almost like people are constantly arguing that SV% is better than GAA, when there's no one on the board ever arguing that GAA is better than SV% anyway......in fact the two of us that argue most consistently to not throw the baby out with the bath water, when it comes to these possessions stats, also point to sv% and goals are just as necessary to paint a truly accurate picture.
That Panthers team was 27th in the NHL in goals for.....and middle of the pack in SV%.
Middle of the pack in two categories, and near the bottom of the barrel of another, is a pretty good indication they weren't as good as their record.
That said, I would certainly be intrigued to see the predictive abilities of (Sv% + GF + Corsi/Fenwick) VS Goal Differential all by itself, and see which was more accurate.
Montana
Champion
We've basically already come up with two.......(Sv%/GF/Corsi) and SV% that normalizes PP opportunities against.
We've just been too lazy to do all the number crunching required and the requisite naming of said brilliant metrics.
PlayerToBeNamedLater
Well-known member
Goal differential isn't a very good projection stat because it's a team stat that isolates one specific year. Teams change from year to year and it doesn't account for that. If Team A has a +20 GD but loses it's best goalie and top C the next year, GD is pretty meaningless now. You need to use player stats to account for player movement from year to year. That's what I don't like about what some of these stats guys have done this year, they're focusing on team models instead of player models, which don't account for team changes.
If you create some sort of formula using goals for, sv%, corsi per player, for every team, I think you can create a pretty nice tool. But I would take some time to do it properly.
If you create some sort of formula using goals for, sv%, corsi per player, for every team, I think you can create a pretty nice tool. But I would take some time to do it properly.
MindzEye
Wayward Ditch Pig
Yeah, that shit sounds a lot like work.
BeLeafer
Well-known member
If you ran a proper regression model and looked at colinearity, you'd in all likelihood find a high degree of colinearitty between goal differential and wins or winning percentage. Colinearity is a measure of sameness ... do the two variables measure much the same thing? The Panthers were a truly lucky team to post that record. It should not have happened, but the goals for/against shook out strangely over the course of the season.
I think the reason people have issue with all this possession/shooting percentage stuff is that Mirtle concluded that the Leafs are lucky. He put way too much weight on the indicators ... much more than he currently professes they should have.
leafman101
Well-known member
Here is something i tested out. PP% + PK% + SV% + Fenwick% + SH%. Its crude here are the results:
2012-13 ForumIce Rating (regular season finish/playoff finish)
1. Chicago - 2.59 (1st/Cup Winner)
2. LA - 2.57 (7th/ 3rd round)
3. Ottawa - 2.57 (12th/2nd round)
4. Boston - 2.56 (5th/Cup Finals)
5. San Jose - 2.55 (11th/2nd round)
6. St. Louis - 2.54 (6th/1st round)
7. Pittsburgh - 2.54 (2nd/3rd round)
8. Washington - 2.54 (9th/1st round)
9. Toronto - 2.53 (9th/1st round)
10. Philly - 2.53 (20th/DNQ)
11. Montreal - 2.53 (4th/1st round)
12. Detroit - 2.52 (12th/2nd round)
13. Vancouver - 2.51 (7th/1st round)
14. Anaheim - 2.51 (3rd/1st round)
15. New Jersey - 2.48 (21st/DNQ)
16. NYRangers - 2.48 (12th/2nd round)
17 NYIslanders - 2.48 (15th/1st round)
18. Edmonton - 2.48 (24th/DNQ)
19. Dallas - 2.47 (21st/DNQ)
20. Tampa - 2.46 (28th/DNQ)
21. Minnesota - 2.46 (15th/1st round)
22. Calgary - 2.46 (25th/DNQ)
23. Columbus - 2.45 (15th/DNQ)
24. Phoenix - 2.41/DNQ)
25. Winnipeg - 2.41 (18th/DNQ)
26. Colorado - 2.41 (29th/DNQ)
27. Carolina - 2.41 (25th/DNQ)
28. Florida - 2.38 (30th/DNQ)
29. Buffalo - 2.38 (21st/DNQ)
30. Nashville - 2.37 (27th/DNQ)
It gets 14 of the playoff teams (technically 15 as NY rangers and Islanders are tied for 15th with Jersey). It catches teams that were better in the playoffs than the regular season (LA/Ottawa/San Jose). It also catches teams that overachieved in the regular season (Montreal/Anaheim/Vancouver). Three of the final four teams ended up in the top 4.
Philly is the big outlier. Save% may need to be weighted more to correct that. Their other numbers were good.
Its rough, but at least its a complete picture.
2012-13 ForumIce Rating (regular season finish/playoff finish)
1. Chicago - 2.59 (1st/Cup Winner)
2. LA - 2.57 (7th/ 3rd round)
3. Ottawa - 2.57 (12th/2nd round)
4. Boston - 2.56 (5th/Cup Finals)
5. San Jose - 2.55 (11th/2nd round)
6. St. Louis - 2.54 (6th/1st round)
7. Pittsburgh - 2.54 (2nd/3rd round)
8. Washington - 2.54 (9th/1st round)
9. Toronto - 2.53 (9th/1st round)
10. Philly - 2.53 (20th/DNQ)
11. Montreal - 2.53 (4th/1st round)
12. Detroit - 2.52 (12th/2nd round)
13. Vancouver - 2.51 (7th/1st round)
14. Anaheim - 2.51 (3rd/1st round)
15. New Jersey - 2.48 (21st/DNQ)
16. NYRangers - 2.48 (12th/2nd round)
17 NYIslanders - 2.48 (15th/1st round)
18. Edmonton - 2.48 (24th/DNQ)
19. Dallas - 2.47 (21st/DNQ)
20. Tampa - 2.46 (28th/DNQ)
21. Minnesota - 2.46 (15th/1st round)
22. Calgary - 2.46 (25th/DNQ)
23. Columbus - 2.45 (15th/DNQ)
24. Phoenix - 2.41/DNQ)
25. Winnipeg - 2.41 (18th/DNQ)
26. Colorado - 2.41 (29th/DNQ)
27. Carolina - 2.41 (25th/DNQ)
28. Florida - 2.38 (30th/DNQ)
29. Buffalo - 2.38 (21st/DNQ)
30. Nashville - 2.37 (27th/DNQ)
It gets 14 of the playoff teams (technically 15 as NY rangers and Islanders are tied for 15th with Jersey). It catches teams that were better in the playoffs than the regular season (LA/Ottawa/San Jose). It also catches teams that overachieved in the regular season (Montreal/Anaheim/Vancouver). Three of the final four teams ended up in the top 4.
Philly is the big outlier. Save% may need to be weighted more to correct that. Their other numbers were good.
Its rough, but at least its a complete picture.
Last edited:
leafman101
Well-known member
One other note on the Flyers, they finished last season 12-8-2 (97 point pace). They might have made a playoff push in an 82 game season.
leafman101
Well-known member
Its straight up addition of the raw data. Eg Last year the Leafs were 18.7% on the PP, 87.9% on the PK, .917 save %, 44.1 Fenwick% and 10.56 shooting%. .187 + .879 + .917 + .441 + .1056 = 2.53.
UWHabs
Well-known member
But adding those together, you're roughly just duplicating a GF/GA stat.
PP% + Fenwick% + SH% ~= GF
SH% + SV% ~= GA
It is still interesting to see, and might do a better job overall than simply GF/GA.
Going back to points before, it'd be great if we could find some sort of individual WAR stat. The problem will come down to how you treat something like shooting pct. Ideally it should be something like a BABIP type stat, where individual players should have "expected" ranges to see deviations from, but I really don't know how to account for that. Really, until someone finds a way to score each shot (ie. speed and distance from goal). Maybe that could be a better way to play things - divide the shots into a few categories (like baseball has GB/FB/LD), and then we might be able to see more trends on certain types of shots.