APBRmetrics

Regularization prevents overfitting and that is not an overfitting issue here anyway. It has also NOTHING to do with the prior informations, because Carter is leading in non-prior informed version and Love has only 0.3 in the same.

And the issue is indeed defense with Love. He does a bit better than last season, but he is still not going for the close out, because he is more concerned with getting the rebound. He is still not going quickly back on defense, there are numerous situations in which Love is going for the offensive rebound while not having the chance to get the ball. Love is still way too often going for the ball first and not for the man in defensive rebounding situations or is skipping a rotation in order to be in a better position for the rebound. That helps his rebounding numbers, but not the defense. Love is overrated by boxscore numbers, his overall impact on the game is much lower than any boxscore-based metric will show. He got compared to Moses Malone last season, and in fact Malone wasn't so much better in the end (better defensively, but not by much). High boxscore numbers, but his teams didn't play that much different when he was out, as we would expected when we see such numbers. Even during his peak seasons with the 76ers he hardly made more than 2 to 3 points difference, clearly less than we saw by other players in the history.

Love is rather similar, and he will get even more overrated when the team around him gets better and he doesn't improve a lot defensively.

_________________
http://bbmetrics.wordpress.com/

I agree that your premise is plausible, but just because Carter is having a good non-informed season does not mean that rAPM is not overfitting...why can't it just mean Carter is playing well?

Maybe overfitting is the wrong word - how about "incorrectly biased" or having a "flawed bias."

The biggest disagreements seem to come from very young and very seasoned players.

EDIT: Maybe it doesn't "love" older players, but there are definitely some huge gaps in what they think of certain players, specifically those deviating from the average amount of seasons played.

I guess this is a question I have for Jerry. In general, ridge regression reduces the variance of the ratings compared to APM (or just O.L.S.), but all else being equal, shouldn't we expect the ranking to be roughly the same (for the non-informed case)?

_________________
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.

Yes, overfitting is the wrong word, and yes, using the wrong parameters can cause "a flawed bias". As far as I see it J.E. used crossvalidation in order to come up with the lambdas. You would need to show that a different set of lambdas would give BETTER results to show that "flawed bias".



Maybe the reason is that older players are smarter, more experienced and are making less mistakes? While younger players are eager to show their value and not necessarily doing the right stuff. As visitors we get caught up in highlight plays, most times done by younger players. Our own bias (caused by the excitement) may let us think those players are better than they really are. We might focus too much on the good things they are doing, while ignoring the errors, especially when the errors are not that easily seen. Who is paying attention to the correct rotations? Who is paying attention whether a bigger player is closing out on a 3pt shooter? Who is looking at the work of the players off the ball, when the ball is handled on the perimeter? Who looks at the boxout of players when the ball is in the air? All of those things aren't shown in the boxscore stats, they are not things we are paying more attention to than other stuff when we watching games. It is easy to miss those things.

Carter plays pretty smart defensively, he goes back on defense and is not celebrating like he did in earlier years, he is passing the ball well and he still draws the attention of the opponents defense, even when he has not the ball in his hand. The Mavericks are forcing more turnovers with Carter on the court, the opponents have to take tougher shots (less fastbreak opportunities), the fouling rate goes down and the opponents are getting less FTA. And on the opposite site the Mavericks are turning the ball over less, drawing more fouls, are getting more easy opportunities (including more fastbreak opportunities). And all of this is not register in his personal boxscore numbers. For Love on the other side all he does is included in the boxscore and the way he is getting to his numbers is not the smartest way in terms of winning games. It is smart for Love to do it, because that will earn him a big paycheck and he might as well do much better within a team concept once he has his new contract, it wouldn't surprise me. And that would have likely a decline in boxscore numbers as a result.

_________________
http://bbmetrics.wordpress.com/

Not really. We have really big overfitting issues in APM, especially with the small sample. When we make the sample bigger, the ranking of APM and RAPM looks more similar. Take the 6yr APM study by Iliardi for example and the 2009er RAPM:

Iliardi from 2003-2009
https://docs.google.com/spreadsheet/ccc ... l=en#gid=0

J.E. 2009 starting with 2002 as priors:

http://stats-for-the-nba.appspot.com/ranking09



The overfitting issue can be seen for example on the 76ers. Young and Williams are playing a lot of minutes together. APM thinks Young is +19.5, and Williams -14.9, that is the result of overfitting. RAPM (non prior informed) has Young with +2.6, and Williams with +0.3.
Now we go to the prior informed RAPM and see Young with +4.2 and Williams with -0.1. We compare that with the 2yr APM in which we have Young with +6.6 and Williams with -3.6. We are getting closer. I would think that a APM study for the last 3 or 4 years would show a similar difference between Young and Williams as the prior informed RAPM values.

_________________
http://bbmetrics.wordpress.com/

Roughly, yes. At least once you use a healthy amount of data.

But, sometimes, the regularization has the following effect:
say you have a team performing at +10 in a healthy amount of games. if you sum up APM*minutes for every player of that team it will have to add up to +10.
If you sum up RAPM*minutes for every player it will not add up to +10, but close (the more data, the closer). With a good amount of games it's probably something between 8 and 9. Let's say +9.

Now, APM has no constraints on player value, it is not afraid to ship all the credit to 1 single player and might say that he's a +15. The entire team will have to add up to +10, so the remaining players will have to add up to -5

RAPM does have constraints on player value, and subsequently will give the top player a lower rating, say +6. The entire team will have to add to +9, thus the remaining players will have to add up to +3

Mystic has cited the Young/Williams situation for Philly. With no constraints on player value APM gives Young a +19. I think we can agree that this is an unrealistic assessment. Because Young's rating is that high, and Philly's APM*minutes will have to add up an overall rating of +12, APM will then just give a (heavy) negative rating to some of it's other players, and Lou Williams ends up with a -15.
Things like that happen less often with more data, but they still happen.

Another example. Say you play two vs two with a friend, there are no substitutions. You two win games against average opponents by an average of 2 points. APM will give both of you a +1, RAPM does almost the same (maybe it says +0.9/+0.9). Now, your friend plays a single game without you, but instead with an average teammate against two average teammates, and wins the game by 2. APM will now give him a +2, you get a zero. RAPM will generally still rate you positive. It will give your teammate a higher rating than yours (the closer the more games you two played together), but it will probably say something like +0.92/+0.88

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

Thanks! Great explanation. I should probably bookmark it.

_________________
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.

I kinda chuckle at the "ripped rebounds from teammates", because I always felt Perkins did the same while here in Boston...

Anyway, I will pay more attention to these observations before I wholly comment. One note, though, if you look at the Wolves' defense, it appears they are defending the rim far better than in previous years. This is - in spite? - of the high number of minutes Love is playing..

_________________
http://pointsperpossession.com/
@PPPBasketball

Interestingly, Love's defensive rebounding %'s are way down this season (34.2% vs. 26.6% DRB%). Minnesota is ranked a few places lower than last season, but it hasn't affected them that much. Maybe someone is stealing those rebounds back from Love now.

_________________
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.

We can even say more about that. When Love is on the court the Timberwolves last season got in average 31.9 defensive rebounds per 36 minutes. Love got 10.8 of those. In this season the Timberwolves are getting 32.2 defensive rebounds, while Love gets 8.7 of those per 36 minutes. The Timberwolves lost the equivalent of 0.4 defensive rebounds per 36 minutes from last, if we adjust via DRB%. Love has 2.1 defensive rebounds per 36 minutes less.

I think a bit of that is related to his defensive plays. Love in this season even went for the close-out a couple of times already, he even stayed with his man instead of going into a better position for the rebound. Overall Love is better this season than last season. Just still not as good as his boxscore numbers indicate. His prior informed RAPM improved from last season to this season, his SPM also improved from last season to this season. Somehow his WP48 decreased ...

_________________
http://bbmetrics.wordpress.com/

I went and looked for things he didn't too well on defense and made some mini videos out of it
http://www.youtube.com/watch?v=rPiAea6NGQg
http://www.youtube.com/watch?v=WZ_sJH8TgUc
http://www.youtube.com/watch?v=qwbHCCgWpjo
http://www.youtube.com/watch?v=8S_W1gA-4rI
http://www.youtube.com/watch?v=Cztli4j3ifo
http://www.youtube.com/watch?v=mGA2JPQ3KE0
http://www.youtube.com/watch?v=lVRTRqCZ_dg
http://www.youtube.com/watch?v=YM4w4cOkLmI
http://www.youtube.com/watch?v=JrDWSaOoKIc
http://www.youtube.com/watch?v=STvR1TguwnM
http://www.youtube.com/watch?v=lY4mE1P2KrM

Forgive the glitches at the start of (probably) every video.
There's one or two videos where he can't be seen. That's why he's not running back (the MIN announcers weren't impressed by it, either). Most other videos are him being in perfect position to contest shots but not actually doing it

I only looked at the first quarter of the CHI and HOU games to find stuff. Obviously, these videos only show what he did wrong, not what he did right. He's definitely not the only one doing these kind of mistakes. Nevertheless, the videos highlight areas he can improve in

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

A lot of players chose not to run back when they'd rather have friendly dialogue with an official...

I dunno, there's too much of a disconnect between his box score results and his +/-. I prefer asking the hypothetical "just how bad is his defense" and approximating a reasonable #. And going from there.

_________________
http://pointsperpossession.com/
@PPPBasketball

RAPM says Love is better on offense and just as good on defense (which is average) compared to last year. Win Shares says he's worse on offense and better on defense. His PER is up and his WP is down. APM has him as a positive, and more of a positive than last year, with his raw numbers showing more of an improvement on defense. Quite a mixed bag. But of players with a decent number of minutes they all think he's a top-15 player except for RAPM.

Requesting the option to vote for Chris Paul

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

Done.
The only player who had dropped out of the top 10 in my chart was Carmelo, so he's gone.
I tried to put Paul in alphabetical order; but then Kevin Love got all of LeBron's (9) votes, as the 7th name in the list.
Crazy.

With his games missed -- 5 of 23 = about 14 of 66 -- Paul is currently 25th in projected eWins this year.
Among that top 25, he's 16th in eW/min.; ahead of J Smith and behind the other candidates plus Duncan, Pierce, Anthony, Westbrook, Jefferson, Bynum, and Millsap.