APBRmetrics

I haven't thought it all the way thru yet, but I bet there would be value in correlating RAPM to S4PM and looking particularly at the comparison ratio outliers.

Monta Ellis has gone a couple years now without any positives:

http://stats-for-the-nba.appspot.com/pm/417.html

Probably not a good thing for a guy who is supposedly the best player on the team.

_________________
The City: A Golden State Warriors-Centric NBA Blog
NBA WOWY: Find statistics of your favorite team with any arbitrary combination of players on or off the court.

Yeah, Evan. that has been my feeling on him even before this data.

Favors only better than Jefferson. Kanter way worse team results than with Jefferson. Maybe they need to move one or two of these guys.

Probably not right away, no reason to rush, but unless something changes, someone should change. Keep / get who works best with Milsap. Or even consider changing that if you don't think that building block is high enough quality to eventually get deep contention happening again.

Bosh -7 compared to Haslem this season and only half & half on w/l for comparisons last season.

Brand, all moderate to large negative comparisons this season? Still at about +1 on non-prior RAPM. I guess he does better in the time that isn't the same 4 other guys as the time that is? More detailed analysis needed to crosswalk between the metrics.

S4PM suggests playing more Odom instead of Marion and less Odom and more of the substitutes. If I wanted to check team analysis and strategy more I'd bookmark this thought and see what they do from here and what the results are.

Hibbert only better than Hansbrough this season after having a majority of winning comparisons last season? Most of these comparison are fairly small minutes. I wonder how stable the comparisons are year to year for those over say 1000 possessions.

Nelson double digits worse than Duhon.

Gay 6 points worse than Quincy Poindexter this season after being 6 points than Battier last season? hmm...

Folks like Trevor Booker and Jason Smith have some good numbers. related to their play and / or their sub.

Boozer -19 compared to Gibson. How many of these issues are known to teams and being worked on? Or just call it / assume it to be mostly noise?

Splitter -11 to Duncan but even to Blair. Splitter-Bonner seems to be working very well. And somewhat better than Duncan-Bonner.

Tony Allen +15 over Mayo.

Mahinmi moderately strongly over his alternatives.

Blatche with double digit leads on 2 of his alternatives and about even with the other two. Keep him despite the other information? Lots of inter-related decisions. Knowledge of technical programming languages probably aren't going to clearly guide / resolve such basketball analysis questions.

Bargnani has only looked better than Ed Davis on this metric since 2008, the guy with the worst RAPM in the league. When do you stop trying to make it work?

Kyrie Irving 14 points worse with same 4 than Ramon Sessions. Small sample but would you dismiss this or look into it more to see if you could learn anything useful?

Carmelo was +14 compared to Amare last season...

Blair about 4-1 negative comparisons to positive over his career.

Dragic -22 compared to Lowry. Only better in Phoenix than his replacement Brooks, formerly the pick n prominent play of Houston. For some reason(s) he had a favorable set of comparisons in 2010 but not before or since. What was different?

Delonte West terrible comparisons in Boston. Ok in Dallas. How much is player, state of mind and how much is system and coach?

Jordan Hill one favorable comparison in his career, 8 negative.

Thabeet all bad to terrible comparisons last season but 50 / 50 for career but Hill plays 20 times more than Thabeet this season? Almost the exact same terrible prior informed RAPM so it is not a good choice to have or chosen to have, at least temporarily til they expire on the cap.

Flynn and T Williams terrible RAPM. Does Houston use APM or not? Much? I thought when Battier was going good they hinted they did. Less references to it after Battier stopped putting up strong RAPM for them and these trades were made and the trade for Martin. For these 4, their prior informed RAPM has either always been terrible and steady or terrible and declining. With Martin is has been 8 straight years below -1. What, if anything, did they see in their databases built on advanced programming languages & tools different from and more powerful than this?

When glancing over the numbers for a player it's always important to realize what team the player was playing on, and who his backups were. It's definitely OK to have slightly red comparisons on this years' Philly team because they're such a good team. The same holds true for any top tier team. On the other side of the coin, someone who has only positive comparisons but was playing on a bad team might still not be an above average NBA player

_________________
http://stats-for-the-nba.appspot.com/

Well, I just had a discussion about that here: http://forums.realgm.com/boards/viewtop ... &start=270

Goes on until page 21/22. After that it starts Deng vs. Josh Smith, while people comparing boxscore numbers. It is quite funny.

Well, in another discussion with a Warriors fan, I learnt that the Warriors FO thought that Brown and Jordan are better in terms of interior defense and rebounding than Dalembert. Well, Brown can't really rebound and Jordan gets pushed around in the paint, no idea, but if that was really the result of an evaluation process of the Warriors staff, you really should be worried.

_________________
http://bbmetrics.wordpress.com/

I think that will confuse a lot of people, because most are unable to put stats into context anyway. Maybe there is a way to compensate for this by presenting a base level for those different situations. Meaning, if a player A + 4 teammates is -3 on the 76ers in comparison to player B and the same 4 teammates, the lineup is still at +4 above average.

_________________
http://bbmetrics.wordpress.com/

I've been tweetin' up a storm on these nuggets.

One thing that might be beneficial Jerry, is the "expected NET" your system comes up with (i.e. rAPM of the player in question minus the rAPM of their replacement).

i.e.

Collison - Perk = 7.
expected(Collison - Perk) = 6.4

!

Good idea, but I want to keep things as simple as possible. I'm not gonna trust these numbers over RAPM anyway, but they're nice to illustrate difference in team impact to someone who never heard of regression

_________________
http://stats-for-the-nba.appspot.com/

On Kevin Love -

True, but his impact was much greater than what rAPM predicts (+9.55 better, weighted per Poss)

Small sample size, Nathan. Exactly what RAPM is adjusting for with its built-in regression to the mean.

_________________
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

5,979 possessions isn't exactly "small sample size." Correct me if I'm wrong there...but Jerry told me that the "# poss" is the minimum # of shared possessions with the same-four-man lineups.

This + Kevin Love's SPM makes him ACTUALLY look like the MIP last year, whereas rAPM makes him look quite weak.


EDIT: From a "regression" perspective: Perhaps if SPM was used to inform the priors, the lambdas could be lower?

That is the sample for all players together. You can't just add them up in such a fashion. You might need a lot of more possessions for each of those players with the respective lineups in order to get a really useful value out of this.



The lowest error in such a test, like you did it, will likely have APM.

_________________
http://bbmetrics.wordpress.com/

Also, remember that it's not listing any combination with <380 possessions. Who knows what happened there

_________________
http://stats-for-the-nba.appspot.com/

Quick question: 380 possession means roughly 190 on offense + 190 on defense?

_________________
http://bbmetrics.wordpress.com/

Usually, yes. It might sometimes be something like 180/200, but it's almost always close to 1/1

_________________
http://stats-for-the-nba.appspot.com/