• Moderators, please send me a PM if you are unable to access mod permissions. Thanks, Habsy.

Kapanen Traded to Pitt

BeLeafer

resurrected
he was garbage in the series. hell, Mitchy even SAID he sucked. not even engaged in the first game.

My problem is that when these stats don't match up with real world results, it's reality that is questioned because the stats say x and that's that.

It piles a whole bunch of credibility onto something that, frankly, completely lacks it from any objective view.

Statistical/analytical examination of hockey is a great idea and maybe one day we can have a set of reliable measures on the chaos of the game we all love. It's a pity that the argument is so often oversimplified into "stats vs. eyetest" and all the characterizations both sides have of each other.
I don't agree with this at all, to be honest.

The stats are a far more accurate 'objective' measurement of reality than the eye test. This is not to say these stats are without flaw or, more commonly, misused/valued. Corsi was all the rage initially but it was never a great stat.

To give a sense of this measurement error issue, I saw someone track manually all the shot locations recorded and it showed significant variance. This is what I was referring to above about the ME being pretty well established. But it's really important to note that we're talking about maybe 5 to 10% error.
 

Preston

Well-known member
I thought Mucho was great after game 1 personally. He created a fuckton. He's frustrating at times because he has the puck so fucking much.. So he's bound to make more mistakes with the puck. That can play tricks on the eyetest.
 

Volcanologist

Well-known member
I don't agree with this at all, to be honest.

The stats are a far more accurate 'objective' measurement of reality than the eye test. This is not to say these stats are without flaw or, more commonly, misused/valued. Corsi was all the rage initially but it was never a great stat.

To give a sense of this measurement error issue, I saw someone track manually all the shot locations recorded and it showed significant variance. This is what I was referring to above about the ME being pretty well established. But it's really important to note that we're talking about maybe 5 to 10% error.
I'm not an "eye test guy", I just want better data and better derivatives from it. I think the current quality is not sufficient that we can start judging performances by reducing players to one number.
 

BeLeafer

resurrected
I thought Mucho was great after game 1 personally. He created a fuckton. He's frustrating at times because he has the puck so fucking much.. So he's bound to make more mistakes with the puck. That can play tricks on the eyetest.
I think that's exactly where the eye test fails most commonly -- mistakes are often magnified way out of proportion.
 

Volcanologist

Well-known member
I thought Mucho was great after game 1 personally. He created a fuckton. He's frustrating at times because he has the puck so fucking much.. So he's bound to make more mistakes with the puck. That can play tricks on the eyetest.
If you're addressing that at me, I never said anything about mistakes with the puck.
 

BeLeafer

resurrected
I'm not an "eye test guy", I just want better data and better derivatives from it. I think the current quality is not sufficient that we can start judging performances by reducing players to one number.
Apologies for suggesting that. It should all be moot once they get the tracking going.
 

Preston

Well-known member
If you're addressing that at me, I never said anything about mistakes with the puck.
Nope, not you. I just think a lot of complaints in general directed towards Mucho are a result of how puck dominant he is (so obviously you will notice his mistakes with the puck more often). Which is actually a good thing.
 

zeke

Well-known member
I don't agree with this at all, to be honest.

The stats are a far more accurate 'objective' measurement of reality than the eye test. This is not to say these stats are without flaw or, more commonly, misused/valued. Corsi was all the rage initially but it was never a great stat.

To give a sense of this measurement error issue, I saw someone track manually all the shot locations recorded and it showed significant variance. This is what I was referring to above about the ME being pretty well established. But it's really important to note that we're talking about maybe 5 to 10% error.
tbh, it may be even smaller than that.

and moreover, it may be the type of "error" that doesn't really mean much anyways - i.e. some shots being mislabelled by a couple feet might not actually mean much at all in the end.


and people are far too wrapped up hating on the numbers.

the truth is we and talking heads have always cared about shot totals even when we were kids in the 80s. there was no denying their importance. there was also constant discussion of how late game shot barrages by trailing teams don't mean as much (i.e. score effects). there was also always discussion about quality scoring chances vs. weak shots. So now we use the same data as we have always used (actually much better data tracking by the nhl now than there was back then), but we're also able to improve it by adjusting for game situations, shot location, shot distance, and don't do things like credit the defense for missed shots by the offense.

the idea that we should ignore this stuff is just dishonest from the get go.
 

BeLeafer

resurrected
Yeah, the biggest weaknees of xg is that it doesn't measure puck movement. Measurement error is fairly minor and is significant only in small datasets (e.g,, single game).
 

zeke

Well-known member
the biggest issue with analytics imo comes on the player side, not the team side.

and this is entirely due to the analytics community pretending that matchups don't matter just because they are hard to quantify.
 

BeLeafer

resurrected
the biggest issue with analytics imo comes on the player side, not the team side.

and this is entirely due to the analytics community pretending that matchups don't matter just because they are hard to quantify.
They're coming around to usage. Still weighting (pardon the pun) for someone to develop a coefficient/adjustment. Silly we still don't have one.
 

Volcanologist

Well-known member
the biggest issue with analytics imo comes on the player side, not the team side.

and this is entirely due to the analytics community pretending that matchups don't matter just because they are hard to quantify.
From what I've read quality of teammate is significantly more important than quality of opposition. ie Micah Purdy who has actually done regression on the data.
 

Habsy

Wrong Thinker Extraordinaire
I don't know why zeke loves his ex-girlfriends so much. Constantly talking about them. xgf this and xgf that. I bet he has a spreadsheet with advanced analytics on those poor women.
 

zeke

Well-known member
They're coming around to usage. Still weighting (pardon the pun) for someone to develop a coefficient/adjustment. Silly we still don't have one.
1. it's very tough
2. it requires a lot of people admitting they made a major mistake
 

zeke

Well-known member
From what I've read quality of teammate is significantly more important than quality of opposition. ie Micah Purdy who has actually done regression on the data.
this is a red herring that is used just to dismiss qoc completely, though, and the fundamental mistake they keep making.

1. Of course quality of teammate matters, which is why have always used team-relative stats from the get go.
2. This question is "regression on WHAT data?". the whole mistake comes from early experiments using guesses at proxies for qoc and qot, not actual hard data. as well as a massive assumption that using the same number for both proxies makes sense. of course teammates will show a bigger correlation than opponents - all looking at qot says is that "this line did X together", whereas looking at qoc means something completely different - i.e. how did this player due against hundreds of opponents. what's worse is that qot teammate data is already influenced by qoc but we ignore it - playing on a matchup line vs playing on a sheltered line can mean a massive amount of qoc influence on your numbers, but that will always have the same impact on your teammate's numbers, so this will be interpretated as qot influence when it's really qoc influence.
3. even saying "qot is more important than qoc" is like saying "apples are more important than oranges". it's a nonsensical statement that makes major faulty assumptions from the get go, and they know that it's a nonsensical statement designed to just ignore qoc completely instead of actually trying to deal with it.
 
Top