Page 1 of 1

Nate Silver Article (BobboFitos, 2011)

Posted: Mon Apr 18, 2011 7:20 pm
by Crow
Author Message
BobboFitos



Joined: 21 Feb 2009
Posts: 200
Location: Cambridge, MA

PostPosted: Sat Jan 15, 2011 12:56 pm Post subject: Nate Silver Article Reply with quote
http://tinyurl.com/4swblez

Solid article. Had a question though - the main contention was Melo's effect of raising his teammates' TS% completely overpowers his own TS%. This is basically OAPM attributed to a marketable 4 factor (well, actually 2, both eFG and FT/FG, bc it uses TS) but I was wondering:

What the effect of raw usage is on average (for your teammates, not yourself)? It makes sense that ANY high volume shooter raises the TS of his teammates, since they're taking the marginal shots away from teammates that otherwise would take them; it's not that they're creating inherently easier looks. (If that was the case, it would be reflected in their apg tallys) Has anyone run a regression on usage vs teammates' TS? Is Carmelo especially special in this regard, or is he simply a 30%+ usage guy who is doing what most 30% usage guys do.

Basically, the old usage vs efficiency argument, but instead of looking at the individual's skill curve, rather the skill curve of the average teammate. (Seems like the sample is much larger other than individual players increasing or decreasing their shot load)

EDITED TO SHORTEN URL LINK
_________________
http://pointsperpossession.com/

@PPPBasketball

Last edited by BobboFitos on Sat Jan 15, 2011 10:06 pm; edited 1 time in total
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3604
Location: Hendersonville, NC

PostPosted: Sat Jan 15, 2011 1:23 pm Post subject: Reply with quote
How would you know what is affecting what?
Does a player's raised Usg% cause his teammates' TS% to rise?
Or does a team's low TS% cause the player to raise his Usg%?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
jmcfaul13



Joined: 12 Jan 2011
Posts: 5


PostPosted: Sat Jan 15, 2011 1:32 pm Post subject: Reply with quote
Didn't George Karl start coaching the team the same year Carmelo came into the league? Perhaps it is his system that raises a player's TS% and not Carmelo.
Back to top
View user's profile Send private message
Crow



Joined: 20 Jan 2009
Posts: 817


PostPosted: Sat Jan 15, 2011 2:03 pm Post subject: Reply with quote
Mike G's important question: "How would you know what is affecting what?"

Not sure of the specific split of what actions are affecting teammates but there is another estimate that can be a stepping stone in thinking about the "who" is having impact.

Joe Sill's multi-season (thru 08-09), factor level Adjusted +/= splits had Anthony at the 95th (100 being best) percentile for team level eFG% impact. It would seem the positive impact would "have to" come from helping teammate shooting as his own shooting is weak.

That is, if you find the method to be valid or reasonably useful despite estimated error. Note that this is measuring the impact of Anthony's field shooting, not scoring, as getting free throws is its own factor.

You still have to separate the direct impact of a player's own shooting from the other stuff he does that might affect his teammates (if that is your goal) but you could find the statistical +/- impact of his own shooting and deduct it from his total "eFG% impact" converted to point value.


In case it is of interest or helps the discussion at all, here was the top 5% on this (full) eFG% impact factor of Adjusted +/- metric from highest estimate on down:

Steve Nash
Rashard Lewis
Ray Allen
Kobe Bryant
LeBron James
Hedo Turkoglu
Dwight Howard
Dwyane Wade
Deron Williams
Tim Duncan
Pau Gasol
Dirk Nowitzki
Leandro Barbosa
Kevin Garnett
Shaquille O'Neal
Chris Paul
Vince Carter
Carmelo Anthony
Raja Bell

I haven't separated what I have in the past called the "local" impact of a player (in this case his own shooting) from his "global" impact on teammate shooting for these guys, but it could be done into at least those two piles.

Separating out what action exactly are having positive impact on teammates and adding positive to the second pile is beyond the intent or reach of this metric as implemented. Probably would have to count more uncounted things and for every game to go further with this approach. May instead have to rely on more limited counting and the eye and judgment.


The high player usage / teammate TS% or offensive efficiency check that Rob suggests would be worth looking at. With the Adjusted +/- factor level information you can estimate a player's overall impact found in his own shooting but also resulting from his spacing, moving. passing, usage and attention drawing and that holistic rating is valuable. I'd think it would be hard to isolate the impact of usage only separate from these other impacts but maybe seeing the data from both approaches would be helpful.


I agree with jmcfaul13 that coaching impact is indeed a variable that should be recognized and perhaps included in a more sophisticated model.

Last edited by Crow on Sat Jan 15, 2011 11:34 pm; edited 10 times in total
Back to top
View user's profile Send private message
greyberger



Joined: 27 Sep 2010
Posts: 51


PostPosted: Sat Jan 15, 2011 2:30 pm Post subject: Reply with quote
Quote:
I'd think it would be hard to isolate the impact of usage only separate from these other impacts but maybe seeing the data from both approaches would be helpful.


Indeed and the raw 'team eFG with player on/off-court', available by season at 82games, is yet another look at the same thing. I sorta expected Silver to go that route - comparing the Nuggets with Anthony to lineups without him.
Back to top
View user's profile Send private message Send e-mail
Crow



Joined: 20 Jan 2009
Posts: 817


PostPosted: Sat Jan 15, 2011 2:44 pm Post subject: Reply with quote
It probably would be helpful to compare the eFG% Adjusted Factor for a player with the raw team eFG%s with player on/off-court (for the covered years) to perhaps try to spot outliers where the error in the Adjusted Factor may be high. But there are real reasons why the 2 values will vary and the Adjusted Factor is an attempt to boil it down and sort it out.

Last edited by Crow on Sat Jan 15, 2011 11:35 pm; edited 2 times in total
Back to top
View user's profile Send private message
Jeff Fogle



Joined: 11 Jan 2011
Posts: 68


PostPosted: Sat Jan 15, 2011 3:15 pm Post subject: Reply with quote
I know article writers generally don't write the headlines for major publications like this. I do think it's odd to call Anthony "the ultimate team player" when:

"he’s accumulated 3.1 assists per 36 minutes played, considerably less than that of other high-volume scorers like Kobe Bryant (4.6 assists per 36 minutes) or LeBron James (6.2)."

and...

"One would also need to think about Anthony’s defense and rebounding, where he is no standout."

He prefers not to pass the ball to his teammates, and they have to cover for his weaknesses defensively and on the boards. One could argue that's the opposite of being the ultimate team player, or even a normal team player.

I get that his teammates end up shooting more efficiently. It's not like those shots are coming off assists from Carmelo though. He's not a big assist guy. Is he just missing shots they would have missed anyway?

"although he scores a lot of points, he does not do so especially efficiently."

Will be fun to see what everyone thinks about both sides of this Silver/Berri debate...
Back to top
View user's profile Send private message
bchaikin



Joined: 27 Jan 2005
Posts: 689
Location: cleveland, ohio

PostPosted: Sat Jan 15, 2011 3:24 pm Post subject: Reply with quote
is this really a "with or without carmelo anthony" analysis, or is it more of a "played on denver or did not play on denver " analysis?...

look at the large increases in shooting by nene hilario and j.r. smith, supposedly due to carmelo anthony. the "without anthony" minutes for nene came when he was 20 years old, the rest when he was 21-28. the "without anthony" minutes for j.r. smith came when he was the ages of 19/20, the rest the ages of 21-25. most players that come into the league that young almost always shoot better from the ages of about 22/23 to 27/28 than 22/23 and younger, regardless of who their teammates are...

also chris anderson was with denver in 03-04, and 08-09 to 10-11, both times when carmelo anthony was on denver. he shot poorly in 03-04 but excellent in 08-09 and 09-10...

lastly from 03-04 to 08-09 (6 years), lebron james played more minutes than any player in the league and averaged playing 3190 minutes a season for the cavaliers. he was on the floor for 19142 of 23811 possible minutes, i.e. he was on the floor 80% of the time the team played. during these 6 years, he threw for 3159 assists, more than all but 3 players (steve nash, jason kidd, and andre miller). he was a far better passer than carmelo anthony, and i'm guessing he also drew as many double teams as anthony did...

yet from 03-04 to 08-09, a full 6 years, and not including the shooting of lebron james, the rest of the cavaliers as a team shot an eFG% (2s and 3s) of just 47.5%, the 2nd lowest in the league those 6 years among all teams (only the clippers were worse)...

would that then mean he was less of a team player than carmelo anthony was during this same time?...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
BobboFitos



Joined: 21 Feb 2009
Posts: 200
Location: Cambridge, MA

PostPosted: Sat Jan 15, 2011 4:14 pm Post subject: Reply with quote
Mike G wrote:
How would you know what is affecting what?



I don't know! I imagine there is some sort of trade off both ways though.

Quote:


Does a player's raised Usg% cause his teammates' TS% to rise?
Or does a team's low TS% cause the player to raise his Usg%?


Again, probably some sort of balancing act. It can't be a hard fast number, but it likely is within a reasonable range w/ low-ish error.

On a personal level, when I play basketball w/ someone who gobbles up most of the shots, I tend to shoot far less often, and my baskets tends to be garbage type putbacks and maybe a wide open 3. That said, they also are taking away some shots I would have created at the rim for myself anyway, or successful post ups/getting fouled/etc... And even with someone shooting all the time, it's not like I still never take a bad shot. Again, all trade offs.

jmcfaul13 wrote:
Didn't George Karl start coaching the team the same year Carmelo came into the league? Perhaps it is his system that raises a player's TS% and not Carmelo.


This certainly could be the case, but couldn't we look at other datapoints (ie. the Sonics) since Karl has coached a lot, in order to see his overall impact at the team level? This seems to be opening a larger can of problems/questions though, in terms of raw coaching impact.

Crow wrote:
Mike G's important question: "How would you know what is affecting what?"

Not sure of the specific split of what actions are affecting teammates but there is another
estimate that can be a stepping stone in thinking about the "who" is having impact.

Joe Sill's multi-season (thru 08-09), factor level Adjusted +/= splits had Anthony at the 95th
(100 being best) percentile for team level eFG% impact. It would seem the positive impact
would "have to" come from helping teammate shooting as his own shooting is weak.

That is, if you find the method to be valid or reasonably useful despite estimated error. Note
that this is measuring the impact of Anthony's field shooting, not scoring, as getting free
throws is its own factor.



I remember JSill's work and found it to be the best usage of APM (RAPM, specifically) I've seen. I wish it was still public.

Quote:


You still have to separate the direct impact of a player's own shooting from the other stuff he
does that might affect his teammates (if that is your goal) but you could find the statistical +/-
impact of his own shooting and deduct it from his total "eFG% impact" converted to point
value.


In case it is of interest or helps the discussion at all, here was the top 5% on this (full) eFG%
impact factor of Adjusted +/- metric from highest estimate on down:
[LOADS OF PLAYERS, TRUNCATED DUE TO SIZE]

I haven't separated what I have in the past called the "local" impact of a player (in this case
his own shooting) from his "global" impact on teammate shooting for these guys, but it could
be done into at least those two piles.


I like how you put that. Clearly when you're 1 of 5 on the court, you're responsible for your own shots - the local side. The other 4 of the 5 are the global. The result is the net production of the 5 man unit. We have fairly reliable stats about how productive the individual/local man is of the 5... but the other 4, and how responsible the 1 is for the 5, is far more error prone. I would imagine most people trust the numbers that place those with great local stats and yet think little of the ones relying fully on global data. From the list of players, that would be Rashard Lewis, Hedo, Barbosa, and Raja Bell. (Raja Bell, wtf???)

Quote:


The high player usage / teammate TS% or offensive efficiency check that Rob suggests
would be worth looking at. With the Adjusted +/- factor level information you can estimate a
player's overall impact found in his own shooting but also resulting from his spacing, moving.
passing, usage and attention drawing and that holistic rating is valuable. I'd think it would be
hard to isolate the impact of usage only separate from these other impacts but maybe seeing the
data from both approaches would be helpful.


I agree with jmcfaul13 that coaching impact is indeed a variable that should be recognized and
perhaps included in a more sophisticated model.


I think coaching impact is something that hasn't really been analyzed too well so far... Certainly it should be recognized.

Jeff Fogle wrote:
I know article writers generally don't write the headlines for major publications like this. I do think it's odd to call Anthony "the ultimate team player" when:

"he’s accumulated 3.1 assists per 36 minutes played, considerably less than that of other high-volume scorers like Kobe Bryant (4.6 assists per 36 minutes) or LeBron James (6.2)."

and...

"One would also need to think about Anthony’s defense and rebounding, where he is no standout."

He prefers not to pass the ball to his teammates, and they have to cover for his weaknesses defensively and on the boards. One could argue that's the opposite of being the ultimate team player, or even a normal team player.


Yeah, it's a strange article headline because the normal traits of "team player" (defense and passing, I imagine) aren't exactly strengths of Carmelo.

Quote:


I get that his teammates end up shooting more efficiently. It's not like those shots are coming off assists from Carmelo though. He's not a big assist guy. Is he just missing shots they would have missed anyway?


One small explanation is he creates offensive rebound opportunities which leads to a greater expected team ORtg (off the rebound) that they otherwise wouldn't have had... But this only explains a small % of overall points + Denver hasn't been a great offense because of their ORB%. They've done it largely through low turnovers and high foul rates.

Quote:


"although he scores a lot of points, he does not do so especially efficiently."

Will be fun to see what everyone thinks about both sides of this Silver/Berri debate...


I can't really wrap my head around Carmelo being worth a mere "6" wins, as WP would have us believe. Again, the number is probably somewhere in the middle, between that number and the 6+~15 (which is what Nate says is possible)... the thing is that's a huge difference, so it sorta matters which is more accurate!

bchaikin wrote:
is this really a "with or without carmelo anthony" analysis, or is it more of a "played on denver or did not play on denver " analysis?...


I'm not sure, the article didn't really reveal how he arranged his stats. If it is truly on/off Denver, all your criticisms are very well deserved and somewhat nullifies the conclusion.

Quote:


look at the large increases in shooting by nene hilario and j.r. smith, supposedly due to carmelo anthony. the "without anthony" minutes for nene came when he was 20 years old, the rest when he was 21-28. the "without anthony" minutes for j.r. smith came when he was the ages of 19/20, the rest the ages of 21-25. most players that come into the league that young almost always shoot better from the ages of about 22/23 to 27/28 than 22/23 and younger, regardless of who their teammates are...

also chris anderson was with denver in 03-04, and 08-09 to 10-11, both times when carmelo anthony was on denver. he shot poorly in 03-04 but excellent in 08-09 and 09-10...


Right, we could probably do the same thing about Nene (who played with Melo for that same block) and show "wow, he increases his teammates TS by X% too!" [all the while maintaining a stellar individual ORtg!] and thus give HIM more credit for the Nuggets offensive success. Etc etc. This probably needs more work/investigation.

Quote:


lastly from 03-04 to 08-09 (6 years), lebron james played more minutes than any player in the league and averaged playing 3190 minutes a season for the cavaliers. he was on the floor for 19142 of 23811 possible minutes, i.e. he was on the floor 80% of the time the team played. during these 6 years, he threw for 3159 assists, more than all but 3 players (steve nash, jason kidd, and andre miller). he was a far better passer than carmelo anthony, and i'm guessing he also drew as many double teams as anthony did...

yet from 03-04 to 08-09, a full 6 years, and not including the shooting of lebron james, the rest of the cavaliers as a team shot an eFG% (2s and 3s) of just 47.5%, the 2nd lowest in the league those 6 years among all teams (only the clippers were worse)...

would that then mean he was less of a team player than carmelo anthony was during this same time?...


well of course, LBJ is all about the $tat$, didn't you know?

_________________
http://pointsperpossession.com/

@PPPBasketball
Back to top
View user's profile Send private message
back2newbelf



Joined: 21 Jun 2005
Posts: 274


PostPosted: Sat Jan 15, 2011 4:23 pm Post subject: Reply with quote
If what was said in the article was true Anthony should show up pretty good in (multiyear, regularized) offense APM. While his offensive multiyear RAPM is positive it's worlds behind Billups', with J.R Smiths' coming in second for Denver. Lots of high volume shooters, even players that shoot really bad %s can look very good in this metric but he doesn't (at least not by much).

Meanwhile, Landry Fields is 5th in one-year-offensive RAPM
Back to top
View user's profile Send private message
greyberger



Joined: 27 Sep 2010
Posts: 51


PostPosted: Sat Jan 15, 2011 4:50 pm Post subject: Reply with quote
Quote:
Meanwhile, Landry Fields is 5th in one-year-offensive RAPM


But that's only 1200 minutes or so... certainly less weight than anything you're doing with 4000+ minutes of recent Anthony performance.

Melo from '07-'09 looks very average in APM, from '08-'10 better but still third on his team. He doesn't have a consistent raw impact on team eFG, but you can find high-usage low-ast% players who do (Dirk, Durant).
Back to top
View user's profile Send private message Send e-mail
Crow



Joined: 20 Jan 2009
Posts: 817


PostPosted: Sat Jan 15, 2011 5:02 pm Post subject: Reply with quote
Generally I think a with /without article should be based on player data with / without Anthony actually on the court when they are. You can argue the other side... sort of... but I think the main analysis should a truly with / without.

From the article:

"In fact, this is true of almost every Nugget who has played a sufficient number of minutes with Anthony. I identified 16 players who have accumulated least 2,000 minutes with the
Nuggets in years when Anthony was on the team, and have also played at least 2,000 minutes in the N.B.A. without Anthony (either because they were playing for a different team or because they were on the Nuggets before Anthony’s rookie season). All but 2 of the players – Marcus Camby and Voshon Lenard – posted a higher TS% playing with Anthony than without him, and on average, he improved his teammates’ TS% by 3.8 points (to 55.0 percent from 51.2 percent).

This appears to be simply seasonal data, team based and not with / without Anthony actually on the court when they are.

That makes this data less compelling but the eFG% Adjusted factor data for Anthony probably helps shore it up. It does for me at least.

Last edited by Crow on Sat Jan 15, 2011 11:48 pm; edited 2 times in total
Back to top
View user's profile Send private message
Kevin Pelton
Site Admin


Joined: 30 Dec 2004
Posts: 979
Location: Seattle

PostPosted: Sat Jan 15, 2011 6:18 pm Post subject: Reply with quote
Crow wrote:
Joe Sill's multi-season (thru 08-09), factor level Adjusted +/= splits had Anthony at the 95th (100 being best) percentile for team level eFG% impact. It would seem the positive impact would "have to" come from helping teammate shooting as his own shooting is weak.

Did you save this from when it was posted? Curiosity about Sill's numbers was one of my first responses to Silver's numbers.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
EvanZ



Joined: 22 Nov 2010
Posts: 292


PostPosted: Sat Jan 15, 2011 6:50 pm Post subject: Reply with quote
This is really interesting stuff. It's motivated me to look further into a question that many Warriors fans have had lately, which is whether Curry and Ellis play better with each other, and whether the team does better with one or the other or both.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Crow



Joined: 20 Jan 2009
Posts: 817


PostPosted: Sat Jan 15, 2011 6:57 pm Post subject: Reply with quote
Kevin Pelton wrote:
Crow wrote:
Joe Sill's multi-season (thru 08-09), factor level Adjusted +/= splits had Anthony at the 95th (100 being best) percentile for team level eFG% impact. It would seem the positive impact would "have to" come from helping teammate shooting as his own shooting is weak.

Did you save this from when it was posted? Curiosity about Sill's numbers was one of my first responses to Silver's numbers.


Yes I saved many of the pages but not all of them. I still use them, will cite them. Not sure about sharing the files. I might. I'd listen to Joe if he wanted to speak about that. I don't see or recall seeing any copyright stamp on them.

Re: Nate Silver Article

Posted: Mon Apr 18, 2011 7:21 pm
by Crow
Author Message
DSMok1



Joined: 05 Aug 2009
Posts: 611
Location: Where the wind comes sweeping down the plains

PostPosted: Sat Jan 15, 2011 8:26 pm Post subject: Reply with quote
In reference to Nate's table: there are so many effects not accounted for, I would consider the data virtually useless. With or without you at the yearly level? My goodness. Aging, teammate impacts, random effects...the standard error on the estimate must be incredible.
_________________
GodismyJudgeOK.com/DStats
Twitter.com/DSMok1
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Crow



Joined: 20 Jan 2009
Posts: 816


PostPosted: Sat Jan 15, 2011 9:11 pm Post subject: Reply with quote
I don't have the specific error estimate for Sill's multi-season factor level Adjusted +/= split for Anthony but I recall he said the factor splits were about as good as the mutli-season data for overall Adjusted +/- and those estimate errors seemed pretty low to me.

It is possible of course that Anthony's eFG% factor estimate is too high but being estimated at the 95th percentile I'd guess the odds are pretty good that he is "truly" in the top third.

Among 8-10 versions of Adjusted +/- that I have with different timeframes and 3 different authors and with a break out of offense from defense, all but one have him as positive on overall offense. Add a few more that weren't split out to offense and defense and that comes to a dozen or more Adjusted +/- runs and all but 3 have him as a positive overall and 2 of the negative estimates are very small.

It is not a unanimous decision but it is a pretty strong trend that he is doing something right to help his team. The data suggests it is on offense. On eFG% impact more than anything else, but actually he is estimated to have a positive team impact across all 4 offensive factors, directly and / or indirectly.

I am not really a fan of Anthony as a team's superstar and the standards for this can be higher than a supporting star or role player, but that is what I see in the Adjusted +/- data with regard to his shooting impact and more broadly.
Back to top
View user's profile Send private message
Jeff Fogle



Joined: 11 Jan 2011
Posts: 68


PostPosted: Sat Jan 15, 2011 11:08 pm Post subject: Reply with quote
Other possibilities that might explain the phenomenon based on a little study and comments from others in this thread and the thread attached to the original article:

*Players generally improve their TS% as they get older, which pollutes the data because anyone who was with Denver before Carmelo got there was obviously older with Carmelo...and anyone who came over from another team then aged as a Nugget. I personally haven't done any research with this particular stat. Is that true? That players generally improve their TS% as they age because they're smarter in shot selection?

*George Karl generally improves his players TS% because of the way he coaches offense. Karl wasn't there in Carmelo's first year, coming over within the season during Carmelo's second year there. That's still enough time to have an impact on the players in the sample. They were either coming over from another team into Karl's system. Or, they were with Denver before he arrived, and improved under his system.

*Whenever you study something happening in Denver in any sport, you have to wonder whether or not altitude is mucking things up. Is there something about playing at alitude that wears down opposing defenses enough to push Nuggets TS% up? Home/road splits would help here. If players were near norms in Denver's road games, but improving noticeably in their home games...what we're seeing might be altitude related rather than Carmelo related.

Not suggesting these debunk Silver's theory. Just trying to frame the issues in a way that might be studied.

Was also thinking that Carmelo might get more credit as a team player if assists were granted in the NBA the way they are in the NHL. If he's kicking out of double teams to somebody who them whips it to an open man...Carmelo was being a team player but not getting an assist. Silver said 70% of Denver's possessions were ending with somebody else in charge of the ball. Might be a characteristic of the way Denver spreads the floor that the ball gets whipped around post-Carmelo in a way that hides Carmelo's contribution to that possession.

Anyway...food for thought in either direction...
Back to top
View user's profile Send private message
Crow



Joined: 20 Jan 2009
Posts: 816


PostPosted: Sun Jan 16, 2011 1:14 am Post subject: Reply with quote
Hockey assists might help some but the next to last pass in a basketball game is not necessarily that important and even if it is sometimes quite helpful it is probably not the same level of help as for the rare hockey goal.

There is also a lot more out there of possible value to count. I'd be for counting more things. But unless most or all of the uncounted helpful actions are counted in public for the full league on a sustained basis, I'll still drift back to Adjusted +/= and comparing the difference between Adjusted +/- and box-score player stats to estimate the sum of that pile of currently uncounted actions that separates the two ratings.


Also on Melo:
A recent study marks him as the 4th highest scorer in clutch-time and a check of his raw clutch-time +/- per 48 minutes shows he was a pretty strong positive the last 3 seasons. Down this partial season for whatever reason or set of reasons- distraction, extremely small sample size or other things.
Back to top
View user's profile Send private message
jmethven



Joined: 16 May 2005
Posts: 51


PostPosted: Sun Jan 16, 2011 10:54 am Post subject: Reply with quote
DSMok1 wrote:
In reference to Nate's table: there are so many effects not accounted for, I would consider the data virtually useless. With or without you at the yearly level? My goodness. Aging, teammate impacts, random effects...the standard error on the estimate must be incredible.


Is this necessarily the case, or can it be assumed that a lot of the effects would even out? Admittedly it is a back of the envelope type of calculation, but I personally think Silver's post does demonstrate some sort of effect that is likely related to the presence of Carmelo. Perhaps it would be worth performing a similar study on some other players to see what the results would be.
Back to top
View user's profile Send private message
EvanZ



Joined: 22 Nov 2010
Posts: 287


PostPosted: Mon Jan 17, 2011 9:33 am Post subject: Reply with quote
I find it ironic how everyone posting on Berri's blog is ripping this article to shreds. Berri is formulating a "response". Rolling Eyes
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
DSMok1



Joined: 05 Aug 2009
Posts: 611
Location: Where the wind comes sweeping down the plains

PostPosted: Mon Jan 17, 2011 9:50 am Post subject: Reply with quote
EvanZ wrote:
I find it ironic how everyone posting on Berri's blog is ripping this article to shreds. Berri is formulating a "response". Rolling Eyes


I'd love to see if he actually addresses the possibility of TS% being increased by another player's usage. Unfortunately, Nate just did a quick look at this... I know he's capable of much, much better work.

Hard to inject a 3rd position into such a discussion.
_________________
GodismyJudgeOK.com/DStats
Twitter.com/DSMok1
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Mon Jan 17, 2011 10:01 am Post subject: Reply with quote
DSMok1 wrote:
EvanZ wrote:
I find it ironic how everyone posting on Berri's blog is ripping this article to shreds. Berri is formulating a "response". Rolling Eyes

I'd love to see if he actually addresses the possibility of TS% being increased by another player's usage. Unfortunately, Nate just did a quick look at this... I know he's capable of much, much better work. Hard to inject a 3rd position into such a discussion.

Exactly right. Nate did a lazy job, and it will be easy for Berri to knock down his analysis. Which is unfortunate, because Nate's underlying point -- that WP undervalues high-usage players -- is of course correct.

Evan: why do you think the response from the Berrites is "ironic?"
Back to top
View user's profile Send private message
EvanZ



Joined: 22 Nov 2010
Posts: 287


PostPosted: Mon Jan 17, 2011 10:19 am Post subject: Reply with quote
Guy wrote:


Evan: why do you think the response from the Berrites is "ironic?"


From Wikipedia:
Quote:
In certain kinds of situational or historical irony, a factual truth is highlighted by some person's complete ignorance of it or his belief in the opposite of it.


Maybe I'm not technically using the term right? What I meant to say is more along the lines of the pot calling the kettle black.

They see all these flaws in Silver's article, but not one of them sees anything wrong with WP?
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3597
Location: Hendersonville, NC

PostPosted: Mon Jan 17, 2011 11:51 am Post subject: Reply with quote
Quote:
I identified 16 players who have accumulated least 2,000 minutes with the Nuggets in years when Anthony was on the team, and have also played at least 2,000 minutes in the N.B.A. without Anthony (either because they were playing for a different team or because they were on the Nuggets before Anthony’s rookie season). All but 2 of the players – Marcus Camby and Voshon Lenard – posted a higher TS% playing with Anthony than without him, and on average, he improved his teammates’ TS% by 3.8 points (to 55.0 percent from 51.2 percent).

Taking career subtotals, with and without Carmelo, as others have noted, invites some strange results. Mostly due to career trajectories rising or falling coincidentally. We could look only at seasons immediately before and after playing with Anthony.

For several (of these 16) players, there is more than one transitional sequence, and for Chris Andersen there are 3.
In total, they provide 24 datapoints. Only 6 shot a lower TS% with Anthony; but these 6 averaged 25% more true shot attempts (FGA+.88*FTA).

In total, player-seasons immediately before or after Anthony had 16.1 TSA/36 at .529 TS%.
First and last seasons with Anthony, these players averaged 14.8 TSA/36 and .547 TS%.

The difference is .018, contrasting with .038 cited in the article.

There might be even less than .018 improvement if we look at all players, rather than just those with 2000 minutes w/wo Anthony. That's rather a select group of players who got along well enough to play together for 2000 minutes.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
EvanZ



Joined: 22 Nov 2010
Posts: 287


PostPosted: Mon Jan 17, 2011 12:13 pm Post subject: Reply with quote
Let's say we had a player who in theory does improve his teammates. Let's call him "LeBron" or "Magic". Very Happy

This is what Adjusted +/- is (theoretically) trying to capture, right?
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3597
Location: Hendersonville, NC

PostPosted: Mon Jan 17, 2011 12:21 pm Post subject: Reply with quote
APM should be measuring how a player affects his team. Both by --
-- his own contributions, and
-- his effect on his teammates' play.

Most observers feel relatively more comfortable with sizing up a player's individual contributions, in points, rebounds, etc.
The arguments that run the gamut (e.g., regarding Carmelo) are about influence on teammates.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Mon Jan 17, 2011 1:47 pm Post subject: Reply with quote
My assumption was that Silver was probably wrong, or at least significantly overstated Melo's impact on teammates. But perhaps not. Berri has put up a response here http://dberri.wordpress.com/2011/01/17/ ... lo-effect/. He runs a regression to determine the Melo effect, reports that it is not statistically significant, and the coefficient was....well, he doesn't say. Which is interesting, and makes me wonder if the coefficient did indicate a Melo impact on TS%. Since Berri limited his analysis to players' first season alongside Melo, it would likely take a very large effect to be statistically significant. In fact, one question someone should ask Berri is whether Nate's reported effect, if also found in his model, would have been significant?

So I remain skeptical of Nate's claim, but Berri may convince me otherwise.
Back to top
View user's profile Send private message
schtevie



Joined: 18 Apr 2005
Posts: 411


PostPosted: Mon Jan 17, 2011 1:51 pm Post subject: Reply with quote
If you read Nate Silver's blog post, I don't think that there is any statement written that is specifically untrue. (To the contrary.) One might quibble about emphasis, a bit. One might wish for greater thoroughness (though it is just a post, after all, one off topic to the blog's primary focus). But the primary argument he makes is in complete accordance with informed opinion, that Carmelo Anthony is an above-average offensive player, in part because of the positive effect he has on teammates. And he specifically says that his remarks are aside from issues relating to rebounding and defense, areas where Anthony has always been (historically) suspect. Furthermore, Silver offers no comment on Anthony deserving star status.

But he does make a point of using David Berri as a foil to kick around. Which is really, pretty darn funny. Both as a stand alone fact and also because of what Berri's reaction has been.

Though Silver is nothing but an undergraduate who happened to finish his degree, the tenured professor who deserves respect and deference is, in reply, extraordinarily deferential and has curiously eschewed his usual attack dog tactics. One might ask, why? Why, indeed.

I really like Nate Silver and would not recommend him following up and getting himself in what would be a highly unpleasant mud fight. But what is interesting is that Dave Berri, despite his apparent love of attention, is clearly uninterested in a well-placed and highly thought of undergraduate publicly taking him to task on his model and its curious assumptions.

Very amusing.
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3597
Location: Hendersonville, NC

PostPosted: Mon Jan 17, 2011 2:22 pm Post subject: Reply with quote
Quote:
In total, player-seasons immediately before or after Anthony had 16.1 TSA/36 at .529 TS%.
First and last seasons with Anthony, these players averaged 14.8 TSA/36 and .547 TS%.

For 1.3 true shot attempts to drag a .547 TS% down to .529, they would have to be .320 effective.

If Anthony is converting 1.3 TSA/G from .320 to .542 (his career TS%), that's about +0.8 point per game.
_________________

Author Message
DSMok1



Joined: 05 Aug 2009
Posts: 611
Location: Where the wind comes sweeping down the plains

PostPosted: Tue Jan 18, 2011 10:56 am Post subject: Reply with quote
Finally, some one weighs in with a good, solid empirical look at the Carmelo question--Kevin Pelton:

http://www.basketballprospectus.com/art ... cleid=1412

This I agree with. Well done, KP!
_________________
GodismyJudgeOK.com/DStats
Twitter.com/DSMok1
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mike G



Joined: 14 Jan 2005
Posts: 3623
Location: Hendersonville, NC

PostPosted: Tue Jan 18, 2011 11:57 am Post subject: Reply with quote
There's also the matter of 7 Carmelo-less games for the Nuggets this season:
Code:
Dec opp ts% TO Pts OpPts
8 @Bos .500 18 89 105
10 @Tor .696 22 123 116
22 @SAS .567 13 103 109
25 @Okl .637 17 106 114
26 Phl .523 13 89 95
28 Por .550 15 95 77
29 @Min .573 12 119 113

wo Melo .578 15.7 103.4 104.1
season .573 14.4 107.2 104.1

These 7 opponents allow .002 above NBA avg TS%.
Without Anthony, the Nuggs have been .006 better than with him.
Probably more than offset by the increase in TO (1.5 per game)
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
schtevie



Joined: 18 Apr 2005
Posts: 414


PostPosted: Tue Jan 18, 2011 12:13 pm Post subject: Reply with quote
Kevin's summary (and Henry Abbott's too, for that matter) are nice. But a bunch of virtual ink could have been spared, and greater clarity provided, if the best, most relevant data were called forth initially to carry the story: Steve and Aaron's multi-year offensive and defensive 2007-08 APM at 82 games.

These have the weakness and strength of ending in the 2007-08 season, but they clearly tell the story of an above average (but by no means elite) offensive player (2.32, standard error 1.2, 43rd ranked NBA player - and 13th ranked 3 - of those playing over 2000 minutes, maximally 16th best in the same ranking if you add two ses) and sucky defender (-4.15, which is really bad in terms of ranking, so why bother counting). Overall that put him at -1.83 which made him about the 100th best player playing major minutes in 2007-08. And if you choose to believe that the Gods of Randomness have conspired against him, adding two ses gets him to the non-elite status of 61st best player in the NBA in 2007-08 (again, of those playing 2000 minutes or more).

The weakness, of course, is that these data aren't quite recent. The strength, however, is the end point picks up the narrative that, curiously, hasn't been recalled (I think) in the recent kerfuffle, the story about how he wised up for his Olympic experience. Indeed, the following year is the first where APM showed him to be net positive (yes, small sample blah, blah, blah) and simple on offs didn't show him to be a disaster on defense and also suggested offensive maturation. And these become an apparent trends, continuing in 2009-10 as well.

So, hats off, yet again, to the sadly too silent (and inexplicably ignored) Steve and Aaron (and, pssst, let's get Memphis' data back up on the site.) You are missed.

(Now will Carmelo Anthony please go away....or not...whatever.)