BeLeafer
Well-known member
That should not be surprising. The more variables you throw into a metric, the greater its chance of explaining variance. It's really just a function of xG having a lot of variables mixed into it.
The ****ing Season Thread 2018-19
- Thread starter BeLeafer
- Start date
That should not be surprising. The more variables you throw into a metric, the greater its chance of explaining variance. It's really just a function of xG having a lot of variables mixed into it.
BeLeafer
Well-known member
No surprise here, shots against is a largely useless stat.
leafman101
Well-known member
The point is, there is no support that simply looking at a couple of "against" stats alone is a worthwhile measure of defensive performance. The analysis suggests otherwise. And from a logical perspective that makes sense. There are just so many variables in hockey.
If the support is out there I am open to it. But I have not seen it.
If the support is out there I am open to it. But I have not seen it.
BeLeafer
Well-known member
SCA/60 clearly has predictive value based on the stats you posted.
I don't look for stats to be the great explainers. This search for the big all encompassing stat is futile. They are indicators and can help to understand things. Because they are limited does not mean you throw them out or never use them singlularly. The raw stats are always better than the indices. SCA/60 appears to me to be a solid indicator of defensive performance. The regression shows it and the rankings show it.
I don't look for stats to be the great explainers. This search for the big all encompassing stat is futile. They are indicators and can help to understand things. Because they are limited does not mean you throw them out or never use them singlularly. The raw stats are always better than the indices. SCA/60 appears to me to be a solid indicator of defensive performance. The regression shows it and the rankings show it.
leafman101
Well-known member
But the rankings don't show it. And the regressions don't show it either. The regressions show there are better measures, and none are particularly great on their own. You are relying on the same analysis to support SCA/60 that you criticize zeke for using with QoC. Which is not wrong, because there is no one all encompassing measure, and we have to be a little subjective digesting the numbers.
But IMO your mistake is relying on one so heavily and ignoring everythign else.
Unless you are looking at QoC, QoT, passing data, zone entires/exits, shot quality, shot volume, among others you simply are not getting a solid indicator of defensive performance.
But IMO your mistake is relying on one so heavily and ignoring everythign else.
Unless you are looking at QoC, QoT, passing data, zone entires/exits, shot quality, shot volume, among others you simply are not getting a solid indicator of defensive performance.
zeke
Well-known member
eh, SCA/60 is not a good measure of defensive performance. even worse when you don't adjust for zone, score, or team effects - all of which we have thorough adjustments for. Then on TOP of that we need to factor in QOC, which we don't have thorough adjustments for yet, but which I'm starting to get a pretty good feel for.
And this has nothing to do with Rielly. In fact, I first started suspecting that QOC was crucial back when the analytics committee was gushing about sheltered 3rd pair Jake Gardiner being not just good but an elite #1 dman......along with other vaunted names like Shattenkirk, Yandle, Klefbom, etc. And then, to the analytics community's surprise, but not to mine, all of those guys saw their metrics fall apart when they were subsequently pushed into actual tougher usage roles.
And then the analytics community goes silent when a bottom pair analytics darling like Nate Schmidt in washington sees his metrics fall apart moving to an elite usage role in Vegas....even though Vegas is wildly successful with Schmidt putting up poor metrics in that role. They can't explain it, but only because they refuse to consider QOC significant enough to make that kind of difference. The fact is Schmidt was excellent as an elite usage dman in Vegas last year despite his "poor" metrics - because when you adjust for competition aka compare his metrics with guys with similar competition, those metrics weren't bad at all. Unsurprisingly, the analytics community started gushing over a guy like Colin Miller instead, saying that he was the real gem on the vegas blueline, because of his metrics......and, shockingly, a quick look at his QOC showed that Miller was in fact the guy getting the sheltered bottom pair usage. again.
Meanwhile, schmidt got suspended to start this year, and Vegas went 8-11-1 in his absence. He returns this year, goes right back into his elite usage role pushing the likes of miller back down into their sheltered roles, and suddenly Vegas is on a 6-2-0 role and moving right back up the standings. And, of course, Miller posted negative relative possession numbers in the increased role in Schmidt's absence.
zeke
Well-known member
Nope. We are just as resistant to using SCF/60 to big up the leafs' dmen.
over the past 3 seasons including this one, here's how the leafs' dmen rank in terms of SCF/60 5v5:
of 261 dmen in total
#2 Dermott
#9 Rielly
#13 Gardiner
#15 Zaitsev
#24 Borgman
#26 Carrick
#46 Hainsey
Are you arguing that Zaitsev is a top-15 offensive dman in hockey? Hainsey is a top-50 offensive dman in hockey? that the leafs have one of the most incredible offensive bluelines of all time, made up entirely of top-25/50 offensive dmen?
zeke
Well-known member
tbh I think the data we have is solid enough to be useful.
the bigger problem is not exploiting the power of the shot attempts stats (i.e. exponentially increased data points over goal based stats) to adjust for context - and this is where the pro teams likely have a big edge.
mbow30
Spudmeister Supreme
It’s not that it isn’t useful; just that it isn’t really clear exactly how we can use it. It also isn’t a good predictive measure because we don’t really know how teams are assessing value from an analytics perspective beyond knowing that they aren’t using the data were using.
zeke
Well-known member
I think the basic premise of the numbers holds water - i.e. controlling the run of play is very important. And don't take for granted the importance of new things that we may now take for granted - i.e. adjusting for score effects, looking at points per minute, much more easily separating out even strength from PP production, etc. Also don't take for granted that we can actually make significantly better predictions now than we use to.
I'm not sure that teams are using wildly different data (i.e. I'm pretty sure most advanced teams most definitely track shot/scoring chance data as the main thrust of their data), but even if they were, I don't think it renders the data we have useless.
But given the weaknesses of the stats we know about, i'm kind of liking my new averaged data just to give a nice ballpark picture of player quality. Instead of getting lost in the weeds, these numbers just take a step back and average out multiple (individually flawed) measures to get a general impression of usage, production, and possession.
I'm not sure that teams are using wildly different data (i.e. I'm pretty sure most advanced teams most definitely track shot/scoring chance data as the main thrust of their data), but even if they were, I don't think it renders the data we have useless.
But given the weaknesses of the stats we know about, i'm kind of liking my new averaged data just to give a nice ballpark picture of player quality. Instead of getting lost in the weeds, these numbers just take a step back and average out multiple (individually flawed) measures to get a general impression of usage, production, and possession.
leafman101
Well-known member
Its not even necessarily that teams are looking for different things, but they may also be better at measuring. For example, Steve Valiquette, has been examining and cataloging quality of shots to evaluate goalie performance over the last decade at a level that just isn't available publicly. He has the percentage each type of shot has of going in, where Scoring Chances on corsia simply measures shots close to the net. His company is a consultant to a number of teams, including the Leafs.
So not only do they likely have access to different data, but they have access to better data. Corsica or NaturalStatrick scoring chance data is likely just not as useful as what teams have access to.
Thats why logically, scoring chances should be very useful, but when you look at the numbers it is no more valuable than shots. Because what we have isn't really scoring chances, or at least isn't an effective measure of scoring chances.
So not only do they likely have access to different data, but they have access to better data. Corsica or NaturalStatrick scoring chance data is likely just not as useful as what teams have access to.
Thats why logically, scoring chances should be very useful, but when you look at the numbers it is no more valuable than shots. Because what we have isn't really scoring chances, or at least isn't an effective measure of scoring chances.
Last edited:
zeke
Well-known member
no doubt. I'm sure they do a better job of measuring shot quality, at the very least. but I'm not sure that that would make a massive difference, in the end.
uncus
Well-known member
Zeke, you never answered my question.
Do your stats take in to account the coaching, the surrounding talent, other outside variables which can effect the numbers you put forth?
If you want to use tavares numbers as a comparable for the last 2 years when measuring him against someone else, does it not make sense to take in to account the players he is playing with?
Tavares's number would be less on the positive side and more on the negative side because of the quality of teammates he was surrounded by in New York. The talent level of his present teammates would surely surpass those of the islanders of the past few years. So how would you account for that?
Just like coaches, some like to have their players play an offensive style, and some want a complete defensive scheme, kitty bar the door.
How does these metrics you are using take that into account?
Just curious.
Do your stats take in to account the coaching, the surrounding talent, other outside variables which can effect the numbers you put forth?
If you want to use tavares numbers as a comparable for the last 2 years when measuring him against someone else, does it not make sense to take in to account the players he is playing with?
Tavares's number would be less on the positive side and more on the negative side because of the quality of teammates he was surrounded by in New York. The talent level of his present teammates would surely surpass those of the islanders of the past few years. So how would you account for that?
Just like coaches, some like to have their players play an offensive style, and some want a complete defensive scheme, kitty bar the door.
How does these metrics you are using take that into account?
Just curious.
zeke
Well-known member
I think the numbers take much of that into account, yes.
Let's look at tavares:
This Yr: A+ qoc, 2.84avgp/60, 51.8avgpos%, +2.9avgrel / PP 4.58avgp/60
Last 2: A+ qoc, 1.81avgp/60, 49.8avgpos%, +2.9avgrel / PP 4.68avgp/60
So if we look at the possession, Tavares does have better possession this year by a good amount, but the exact same posession relative to his team - which makes sense, given that TOR is a solidly better possession team than NYI was (though unfortunately not as much better as they should be). He's done this against similar elite quality of competition (which i imagine will start going down now that Matthews is back).
His PP offense is the same, but we do see a big boost in scoring at ES. A good chunk of that is that scoring is up league-wide this year by a good amount. But it also makes sense that playing with Mitch has helped his scoring too. On the downside, much of that may be due to his on-ice shooting percentage going from an average 8.5% to a maybe unsustinable 12.5% so far this year.