Neil Greenberg

[Daniel]

Does anyone here read this guy? Even if you're all about esoteric statistical analysis, he seems like a waste of air.

It's behind the espn (yeah I know) pay wall, but his latest thing is calculating the team with the most "lucky" wins, this year, the Penguins, and noticing a trend that these teams generally don't go far in the playoffs. He then comes up with a percentage chance of an upset in each particular playoff series. It isn't even wrong.

Sent from my iPad using Tapatalk HD

Edited April 30, 2013 by Daniel

[Devils731]

Someone on another forum posted the whole article if you Google it but I won't post it here.

He's really just using expected wins using a formula derived from goal differential against actual wins.  That's kind of a "no duh" of who might be a little overrated, since I think most everyone agrees the best teams generate the greatest goal differentials.

The more interesting part, for this forum, is how well score adjusted Fenwick has done in projecting playoff results.  In the last 6 years, the team with the best score adjusted Fenwick have won the cup 3 times, lost in the finals once, lost in conference finals once, and lost in the first round(Pitt last season).  This also cheats a little by only counting LA last season after the Carter trade.  

 

A lot of people don't like Fenwick/Corsi but the best team in them seems to have done fairly well and the shot differential chart for the playoffs this season was fairly overwhelming too.

[Marshall]

 
A lot of people don't like Fenwick/Corsi but the best team in them seems to have done fairly well and the shot differential chart for the playoffs this season was fairly overwhelming too.

 

For those who haven't seen it.

 

[Derlique]

Daniel- When I brought this up a few weeks back, you said it was "Cargo Cult Science".

[Daniel]

Daniel- When I brought this up a few weeks back, you said it was "Cargo Cult Science".

This is different. He places a percentage chance of a certain outcome for a one time event, i.e a playoff series that will never happen again. He can't be wrong, the same way I can't be wrong by giving a 70 percent chance that Krys Barch is going to score 50 goals next year. When he doesn't score that many goals, I can just say that it turned out the thirty percent scenario was correct. Thus the saying, he isn't even wrong.

Last year he did something similar to give percentage chances to the outcome of all the first round playoff series. Later on he bragged that his methodology predicted the Kings would beat the Canucks. Of course he didn't mention that he called at least half of the series wrong. I pointed this out to him in the comments (to his credit he does participate) and got crickets in response.

I can live with the esoteric stats that people like Tri believe in. I don't, or at least I think that they're ok so long as you don't get too caught up in them. Greenberg likes to throw out a lot of formulas to make it sound like what he's doing is scientific. Turns out his sciency sounding jargon is as good as any half way knowledgeable fan that has a pair of eyes and a little bit of common sense.

Sent from my iPad using Tapatalk HD

[squishyx]

For those who haven't seen it.

I don't want to turn this into another microstats debate, because there is some correlation between a teams corsi and how they finish.

But it's substantially stronger if you looked at just team +/-

Rank     Team           GP      GF      GA      +/----------------------------------------------------1	 P - CHICAGO	48	155	102	532	 Z - PITTSBURGH	48	165	119	463	 Y - MONTRÉAL	48	149	126	234	 Y - ANAHEIM	48	140	118	225	 X - BOSTON	48	131	109	226	 Y - WASHINGTON	48	149	130	197	 X - NY RANGERS	48	130	112	188	 X - LA	        48	133	118	159	 X - ST. LOUIS	48	129	115	1410	 X - TORONTO	48	145	133	1211	 X - OTTAWA	48	116	104	1212	 X - DETROIT	48	124	115	913	 X - SAN JOSE	48	124	116	814	 Y - VANCOUVER	48	127	121	615	 COLUMBUS	48	120	119	116	 X - ISLANDERS	48	139	139	E17	 TAMPA BAY	48	148	150	-218	 X - MINNESOTA	48	122	127	-519	 PHOENIX	48	125	131	-620	 PHILADELPHIA	48	133	141	-821	 EDMONTON	48	125	134	-922	 DALLAS	        48	130	142	-1223	 WINNIPEG	48	128	144	-1624	 NEW JERSEY	48	112	129	-1725	 BUFFALO	48	125	143	-1826	 NASHVILLE	48	111	139	-2827	 CALGARY	48	128	160	-3228	 CAROLINA	48	128	160	-3229	 COLORADO	48	116	152	-3630	 FLORIDA	48	112	171	-59

Edited April 30, 2013 by squishyx

[Daniel]

For those who haven't seen it.I don't want to turn this into another microstats debate,

Neither do I. There's plenty of that to go around. It was really just seeing something that annoyed me at a visceral level. I was wondering if there was some common ground that this guy was basically making up a bunch of nonsense.

Sent from my iPad using Tapatalk HD

[Triumph]

I don't want to turn this into another microstats debate, because there is some correlation between a teams corsi and how they finish.

But it's substantially stronger if you looked at just team +/-

Rank     Team           GP      GF      GA      +/----------------------------------------------------1	 P - CHICAGO	48	155	102	532	 Z - PITTSBURGH	48	165	119	463	 Y - MONTRÉAL	48	149	126	234	 Y - ANAHEIM	48	140	118	225	 X - BOSTON	48	131	109	226	 Y - WASHINGTON	48	149	130	197	 X - NY RANGERS	48	130	112	188	 X - LA	        48	133	118	159	 X - ST. LOUIS	48	129	115	1410	 X - TORONTO	48	145	133	1211	 X - OTTAWA	48	116	104	1212	 X - DETROIT	48	124	115	913	 X - SAN JOSE	48	124	116	814	 Y - VANCOUVER	48	127	121	615	 COLUMBUS	48	120	119	116	 X - ISLANDERS	48	139	139	E17	 TAMPA BAY	48	148	150	-218	 X - MINNESOTA	48	122	127	-519	 PHOENIX	48	125	131	-620	 PHILADELPHIA	48	133	141	-821	 EDMONTON	48	125	134	-922	 DALLAS	        48	130	142	-1223	 WINNIPEG	48	128	144	-1624	 NEW JERSEY	48	112	129	-1725	 BUFFALO	48	125	143	-1826	 NASHVILLE	48	111	139	-2827	 CALGARY	48	128	160	-3228	 CAROLINA	48	128	160	-3229	 COLORADO	48	116	152	-3630	 FLORIDA	48	112	171	-59

 

No one would dispute that.  You may as well say there is a correlation between scoring goals and winning, or preventing goals and winning.  The trouble with goal differential is that it isn't predictive.  Shot differentials tend to be much more sustainable than goal differentials.

Edited April 30, 2013 by Triumph

[squishyx]

No one would dispute that.  You may as well say there is a correlation between scoring goals and winning, or preventing goals and winning.  The trouble with goal differential is that it isn't predictive.  Shot differentials tend to be much more sustainable than goal differentials.

Perhaps you know of one but I haven't found a website that provides snapshots of past seasons at a given date. Something like "2011 standings at date XX/XX/XXXX, or 50% of the way through the season. That would probably lead more accurate assessments.

Short of that, we do have season totals, and can then look at playoff performances to see if there is any connection there. I'm going to look at the top 5 finishers in corsi and goal differential (GDiff) for a given year, then compare it to the SC finalists. I'll also list where the finalists finished in [corsi, +/-] as a reference.

*I used http://stats.hockeyanalysis.com/teamstats.php for corsi %, and nhl.com for goal diff.

-- 2013 --

Actual: TBD

Corsi top 5: LA, NJD, Bos, Chi, Ott

GDiff top 5: Chi, Pit, Ana, Mon, Bos

-- 2012 --

Actual: LA[2,11] vs NJD [13, 9]

Corsi top 5: Det, LA, Pit, Bos, Stl

GDiff top 5: Bos, Pit, Van, Det, Stl

Result: Corsi, Corsi had one of the finalists, but missed on the Dev's, GDiff had as bad a year

-- 2011 --

Actual: Bos[14, 2] vs Van [5, 1]

Corsi top 5: SJ, Det Pit, Chi, Van

GDiff top 5: Van, Bos, Pit, Phi, SJ

Result: GDiff, Good year for GDiff, nailed it (albeit wrong winner), Corsi just grabs Vancouver, but pretty wide right miss on Bos

-- 2010 --

Actual: Chi[1,2] vs Phi[13, 12]

Corsi top 5: Chi, Det, Was, Tor, Bos

GDiff top 5: Was, Chi, Van, SJ, NJ

Result: Tie

-- 2009 --

Actual: Pit[19,9] vs Det[1, 3]

Corsi top 5: Det, Cal, Chi, Was, SJ

GDiff top 5: Bos, SJ, Det, Chi, NJ

Result: Tie, but I wouldn't argue too hard against anyone who said they would give a slight edge to Corsi

-- 2008 --

Actual: Det [1,1] vs Pit [29,4]

Corsi top 5: Det, Was, NYR, SJ, Cal

GDiff top 5: Det, Mon, Dal, Pit, SJ

Result: GDiff, Corsi did nab the wings as #1, but their eventual challenger was pegged at #29th

(stats.hockeyanalysis.com doesn't have data prior to this, thats why I stopped here)

What I learned from this: They both kind of suck at accurate playoff predictions. And while I certainly wouldn't declare GDiff a winner with a meager 2-1-2 record, I would hope we could agree that Corsi has no clear cut predictive power over it. Again, I recognize that I am comparing season stats to playoffs, a better value would be to look at snap shots at quarterly marks or something, but I didn't have that on hand.

This goes backto an intrinsic argument I've always held, stats do not predict future results with enough consistency to be used reliably. Stats, including microstats, are good for qualitative comparisons of past data sets, or setting reasonable expectations. And now I apologize for letting this become another microstats debate.

Edited April 30, 2013 by squishyx

[Triumph]

First off, goal differential isn't absent Fenwick Close or score-adjusted Fenwick or whatever microstat you choose to use - they're correlated with one another.  Teams that have a good FenClose tend to have a good goal differential.  

 

Second, your argument about goal differential was like saying 'Hey guys, I sorted by points and I found that most years, the teams with a lot of points make the playoffs and the ones without don't!'.  I mean, yeah, occasionally a team has a decent goal differential and misses out on the playoffs because of a poor shootout record or something like that (or vice versa), but yeah, if you score more goals than your opponents, you will win more games than them.  If you are +30 goal differential in a full season you are basically a lock to make the playoffs.  I mean, you could sort by points and raw Corsi and almost certainly find the same thing.

 

http://blogs.thescore.com/nhl/2013/02/25/breaking-news-puck-possession-is-important-and-nobody-told-the-cbc/

 

This one's probably better, regarding PDO, although it's old:  http://www.mc79hockey.com/?p=2996

 

Teams have SOME ability to sustain shooting percentages and save percentages, but it is much smaller than the ability to retain outshooting ability.

[squishyx]

First off, goal differential isn't absent Fenwick Close or score-adjusted Fenwick or whatever microstat you choose to use - they're correlated with one another.  Teams that have a good FenClose tend to have a good goal differential.  

 

Second, your argument about goal differential was like saying 'Hey guys, I sorted by points and I found that most years, the teams with a lot of points make the playoffs and the ones without don't!'.  I mean, yeah, occasionally a team has a decent goal differential and misses out on the playoffs because of a poor shootout record or something like that (or vice versa), but yeah, if you score more goals than your opponents, you will win more games than them.  If you are +30 goal differential in a full season you are basically a lock to make the playoffs.  I mean, you could sort by points and raw Corsi and almost certainly find the same thing.

 

http://blogs.thescore.com/nhl/2013/02/25/breaking-news-puck-possession-is-important-and-nobody-told-the-cbc/

 

This one's probably better, regarding PDO, although it's old:  http://www.mc79hockey.com/?p=2996

 

Teams have SOME ability to sustain shooting percentages and save percentages, but it is much smaller than the ability to retain outshooting ability.

But I am not arguing "hey, teams with the best +/- tend to be good teams with lots of goals and points!". My actual argument from my last post is "an intrinsic argument I've always held, stats do not predict future results with enough consistency to be used reliably".

That was reinforced when I was researching your implied argument that Corsi was better at predicting future results then GDiff. As it turns out is absolutely measurable. We can (and probably always will) debate on how much weight we should tether to a given stat, but we can conclusively prove (or disprove) if one stat predicts results better then another. 5 years of season-playoffs data is not that much in the grand scheme of things, but for me it's certainly enough to suggest they are both poor at predicting future success, and neither being measurably better then the other.

Thanks for the second link it is somewhat close to what I was looking for, I think I just need to write a script to compile the data I need.