Page 2 of 2

Re: 2011 players and historic statistically similar seasons

Posted: Mon Oct 31, 2011 9:56 pm
by Mike G
Crow wrote:Our comparisons for Harden will vary in large part because I made the choice to look at wings while you stuck with SGs. ..
I did not, and do not, nor intend to restrict by any alleged position. Productions define a player much better than any arbitrary tag.
Harden is an above-average scorer who gets about 1/4 as many rebounds as points and 2/3 as many assists as rebounds. He's a 16-4-3 guy, as are the others, more or less. Is Grant Hill a SG, or is he a 'wing'?

I'm thinking the players a given player resembles might define his position pretty well. Not as 12345 or 123, but as 3.47 and such.

Probably a solid team is looking for a particular type of scorer -- a spot-up 3pt ace, a slasher who draws fouls, etc -- more than a floundering offense, which will appreciate points however they come. However, productivity at scoring (or anything) is hard to replace with any subset of the production. A great offensive rebounder who doesn't hold up his end on the defensive boards is still just a so-so rebounder.

Re: 2011 players and historic statistically similar seasons

Posted: Mon Oct 31, 2011 10:31 pm
by Crow
"I did not, and do not, nor intend to restrict by any alleged position."

Right. My mistaken assumption. I should have remembered that from the previous page of the thread; but, moving between threads and trying to be quick I forgot and just assumed based on seeing all or nearly all SGs on your list.

There has been a lot of different versions of your player comparisons of yearly stats and career stats but you have been consistent on not restricting by position.


2/3rds of your top 15 similars for Harden weren't in my sample. It really isn't possible to do a full and fair comparison between the two measures unless and until I load the RAPM data for the other players. But I wanted to at least get started. Might not got back to it for at least awhile.


"He's a 16-4-3 guy, as are the others, more or less."

Ok that is similarity using one approach but the top 9 guys on your similarity list for Harden vary from +1.6 to -3.6 on offensive RAPM and +1.4 to -1.9 on defensive RAPM and between +2.4 and -4.1 on overall RAPM for 2010-11. I think there is also a place for a similarity metric that gives fairly large weight to these larger roll-up metrics. The differences are worth seeing.

Harden was +1.7 on offensive RAPM which is outside the range of his top similars on your list but 7 of the 9 were more than a full and real game level point different than his rating. He was 0 on defensive RAPM which is sorta near the middle of the range of his top similars on your list. 5 were within minus or minus 1 point but they are still pretty spread out and not that similar to each other on this metric. 7 of the 9 were more than a full and real game level point different than his overall RAPM rating and 7 also being lower than Harden's +1.7, in fact 7 were +0.3 or lower on down to -4.1. Again not too similar of this overall measure.

On the per 36 minutes stats you used everyone is pretty much within 50% plus or minus of Harden's marks. Of RAPM they are often several multiples of his marks distant from his marks.


Even on boxscore stat based Offensive Rating the ratings vary from 101 to 109 and Defensive Rating from 105 to 114. That ranges from modestly above league average to well below average on each with each representing half of a player's total activity. Not too similar a group from this perspective especially when taken in combination.


If I understand it correctly based on what is listed, your metric uses 5 parts offense, 2 parts defense. Mine only has 5 defensive ingredients to 9 offensive but defense does get about half the total available weight compared to less than 30% in yours.


"the players a given player resembles might define his position pretty well"

It might define his role pretty well but role and position are not always the same and it seems worth noting to me when player roles diverge from traditional position-based roles and when roles overlap and when roles are understaffed.


The more perspectives, the more information for consideration. I find your similarity system and others (present and past) helpful but I wanted something else too. If I had more time I'd also try to do an updated version of Ed Kupfer's old factor analysis based similarity system or like David Sparks'. Both were position agnostic, similar in that regard to your approach and both likewise relied entirely on simple boxscore based rate stats. I'd use more complex stats including the bigger roll-ups. In this old thread http://sonicscentral.com/apbrmetrics/vi ... view=print
Eli Witus showed interest in some of kinds of the stats that I've added. Dean Oliver talked about achieving more balance by including more defense. I've upped the degree of defensive representation using Defensive Rating and Defensive RAPM. Conceivably counterpart defense (from 82 games on Evan Zamir site or queencityhoops) could get a share too.