Luol Deng vs Derrick Rose

[Fast Dont Fib]

Really nice post, micro.

Indeed.

And thanks for the links. Very informative stuff.

[Steve Brule]

Just a random aside: Luol Dengs midrange jumper is butter. I don't think he gets enough credit for it. I'd say top five in the league but I know some of you stat buffs will disprove that. Regardless, that shot is hot fire.

Actually, this year, he's been hot crap. He's shooting 14.3% from 10-15 feet and 32% from 16-23 feet. He's shooting around 31.25% on shots from 10-23 feet and 33.3% from 3-9 feet.

He's been TERRIBLE everywhere between the rim and under the arc, and he hasn't been amazing at the rim either, where he's shooting 2% below positional league average for SFs playing 25+ mpg at the rim.

Deng's been pretty much only good at hitting 3s and competent at hitting FTs this year.

Historically speaking, he's been bad or mediocre in most of those zones as well, but he's been empirically terrible this season, no questions. He doesn't take a lot of shots from 10-15 feet (0.5 per game), so it's not really a meaningful zone for him, but he's been quite bad from 16-23 feet this year and he's taking almost 4.5 of those per game.

From an NBA.com article the other day:

Shooting just 31.2 percent (39-for-125) from mid-range, which ranks 109th among 118 players who have attempted at least 75 shots from between the paint and the arc. The only 2012 All-Star shooting a lower percentage from mid-range is Andrew Bynum, who's 5-for-25.

I'm an ignoramus.

[micromonkey]

Look at this detail:
http://stats-for-the-nba.appspot.com/pm/755.html

Interesting--I think the 2010 season is still "dragging" Rose down.
Basically a cluster-F of VDN lineups (and different positions) he's being compared to (ie. not having Rose and Deng on as much as possible).

Once synergies are found and coaches are smart to play upon them the guys look better.

Funny that Thibs is best coach per this (probably maximizing lineup based on some RAPM-like stat--got to think the pros have more variables and adidtioanl data to mine...)--again looks like he has some version of this and probably more advanced stats too make his lineups....

Thanks tsherkin for posting this thread--really made last night and this afternoon extremly unproductive investigating this stat!

[SideshowBob]

Nice explanations micro. I've been following the thread on the PC board, and the last few posts are trending towards the same conjecture you've laid down. That Rose hasn't played long enough to be properly evaluated in the RAPM system and his early years have been dragging him down.

[tclg]

Very interesting development I really learned alot about RAPM , This really has been a great thread

[Rerisen]

This too me is more interesting:
http://stats-for-the-nba.appspot.com/5mu_11
Knowing Thibs you know why he rode that line so hard in the playoffs and start of this year (Taj/Omer/Brew/Rose/Deng).

This lineup caught discussion last year before the playoffs. Thibs used it most often to close down games in which we already had leads. If was effective because the massive defense they brought could shrink the clock down, since teams took so long to score points on them. Meanwhile Rose just handled the offense by himself, creating plays.

But this lineup would be (and was) far less effective in closer games or where we were behind, because then it became easier to put the pressure back on the Bulls with few scorers out there. I think Thibs went to it vs Miami more out of desperation than anything else, when Boozer and Noah were kind of choking.

Which shows a good cautionary lesson about some of these +/- numbers. Which is that many are based on specific contexts, or situations, and can't necessarily be extrapolated into all other scenarios. Which is what is probably happening to some degree, if writ large, with the Deng / Rose split as well.

Run a whole year without Deng or Rose, and the complicated interactions between players would probably begin to show out more meaningfully against what RAPM guesses they would be, with it having very small per game samples of Deng without Rose instances to go on; mostly just Deng and backups vs other backups a few minutes a game. And then also a short 10 games stretch this year, when we played mostly scrub teams, and also CJ Watson was hot and playing over his head for most of. Play a whole year without Rose, and CJ is not going to be putting up 18-20 PER like he was for a large portion of that run. He was just hot, and has already begun to revert back to his norms. But such a streak by one player, esp the one replacing Rose, can skew these small samples greatly.

[alucr]

Further digging I see r square of .67. So if I read that right then it means with 99% confidence we can say play x is good or bad (overall a + or -) but with only 67% confidence on the magnitude. Meaning we are 67% sure Deng is “more valuable” than Rose.

One small note on this, it is not saying there is a 67% percent chance Deng is more valuable than Rose. R squared is a measure of how well a curve of best fit represents a set of points. Also, it comments on how much error comes from included variables versus how much error comes from other variables or natural variance. 0.67 is saying that 67% of the error that leads to the line of best fit not fitting perfectly is from the variable the line was fit to while 33% is from other stuff. R squared makes no comment on the accuracy of the data the line of best fit represents unless the data has zero error. The information that 0.67 is the R squared actually says there is only a 67% certainty that it is even representing the data correctly. This data that it is representing also has its own error. By saying there is only a 67% chance of the flawed data being represented correctly means that the errors are stacking on top of it.

For example, I can say Rose has a plus minus of + a trillion while Deng has + 100 billion based on numbers I pull out of my ass. I can also make these numbers up so that my line of best fit spits out an R squared of 1. It doesn't mean that Rose has a + 1,000,000,000,000 with 100% accuracy because the data used to reach that conclusion has 0% accuracy. This is an extreme analogy, but plus minus is a beast of uncertainty.

[micromonkey]

^^Yes that makes sense re: r sq.

It is an interesting stat but Rerisen's points of who is playing against whom is very relevant and I was starting to notice as well.

This year JL3 has an insane defense rating --most of it skewed because of the blowout Wiz/Cats wins.
68/314 (21.6%) of his min against Wiz and Bobcat 64 pt against games. A lot of the rest are spot scrub minutes with the super D line against 2/3rd rate opponents.

Or you implicitly believe everything about a number and his D is better than all other PG's in the league and PPP less than Dwight or KG.....

It actually pulls Roses D rating down this year--where other stats show he's been great.

Again as 1 number for everything there are issues--also looking at Roses earlier years the crazy lines have him playing "against" Centers/SFs, etc--so defensively he's worse than Brad Miller--yeah I think most PG's would be (product of some VDN nelly ball lineups I think). And I'm sure Rose isn't the only one these kind of weird "matchups" are happening too. Over time it could wash out but for right now it doesn’t tell you a lot.

in a group of other numbers it's interesting but you would really have to know how teams play their lineups--I think good coaching is a huge part--like the super Def linup--playing that in the right situation and against the right opponent.


PPP is another good stat and this year it shows Rose's OFF/DEF growth. .
PPP Off/Def (opp PPP) (synergy sports)

.93/.86 2010
.94/.78 2011 MVP
.97/.78 2012

The work in the gym and watching film is paying off.

[alucr]

What any variation of plus minus is decent at saying is what actually happened in the past as a part of the lineups they were in. That is literally the extent of what it should be used for. Trying to seperate out one player's impact to compare it to another's based on plus minus is going into the realm of the stat that loses all meaning. Plus minus is a TERRIBLE predictor. Trying to say that going forward Deng would be more valuable than Rose is using the stat to predict. There is massive error involved in this to the point where the stat is saying for example Deng is +5.0.....but it has an error of +/- 5. So really he is either 0.0 or +10.0...somewhere in there (even that is only 68% accurate).

[StopItDRose]

The only way to truly compare would be to keep all variables constant, so you would have the exact same Bulls lineup except one is with Rose, one is with Lu, against the same team with the same lineup, and put Rose up against the same player Lu was guarding, and Lu against the same player that Rose was guarding. >.>

[SKN]

I agree, with some of the other posters that Luol Deng is a marginal all-start at his best. Luol hasn't shown the offensive ability to even take over a game for a few minute stretch. I have no qualms at all about his defense, he gives a ton of effort, and plays his heart out hurt every night. If he had Rose's decisiveness on offense, this wouldn't have been Deng's first all-star game.