APBRmetrics

Evan, I noted the possibility of using your developed database for this task (though maybe I should have left that to you) but of course it is up to you to volunteer and act.

Crow wrote:Evan, I noted the possibility of using your developed database for this task (though maybe I should have left that to you) but of course it is up to you to volunteer and act.

No worries. I was just giving you a hard time.

To be honest, I'm flattered whenever anyone suggests anything I do can be useful for something.

It is not player level with / without data, but at team level in the regular season the Bulls players are better on offense with Rose than without in every case that I have data on (within the constraints of what nba.com provides on +/-) but are also better on defense without Rose in 75% of the cases.

Only Boozer is better on net team +/- (considering both offense and defense) without Rose than with and by 1.8 points. Looking at the individual stats for this with and without might be interesting case- usage, eFG%, TS% and Offensive Rating or Evan's O100.

Korver and Noah were only 3 and 2 net team points better with Rose than without. 4 others were at least 6 net team points better with Rose.

Overall the Bulls were only 2 points better with Rose than without. The other guys I don't have pair data on like Gibson and Asik and deeper bench players are probably mostly better without Rose than with to pull the average gap down to just 2.



On the Magic, the 8 guys I could compute with and without Howard team +/-s for were all better on offense with Howard and all but one better on defense with Howard (Reddick was the exception). The overall on / off for Howard was about +10 per 100 possessions.


One year RAPM has Rose at +3.4, Howard +4.4. One year traditional APM has them both easily over +10. It looks like one method got the rating pretty consistent with raw +/- in each case and 1 varied widely and vice versa for the other guy. But which Adjusted +/- is "right" or more might? If either is. Maybe one or both "right values" lies in-between? And why the divergence in the match-up with raw +/-? Surface mystery or deep mystery, more work needed to reconcile or try to understand anything better, results or metric.

The greater consistency on team +/- for Magic players with Howard than without and on offense and defense might give me somewhat more confidence in his rating and in a higher rating for him than Rose.

When there's a big difference between APM rating and rAPM rating or a surprising APM result I like to look at the pair and trios pages at NBA.com to see how much overlap there is in teammates' minutes.

For example in the 2011 regular season Kobe had a B-Value.com APM rating of -8.5 (SE 6.4) while Gasol finished +11.9 (SE 5.8). rAPM has Kobe at +1.3, Gasol at 3.9, including the playoffs.

The Lakers had consistent rotations through the regular season - 86.7% of the time Bryant played Gasol was playing, and 64.5% was spent in the quartet of Fish Kobe Artest Gasol. I think that pure APM isn't very reliable in such cases of extreme collinearity, which is partly reflected in a large SE for the players affected relative to their total minutes played.

I think rAPM's numbers for the quartet are a lot more plausible than pure APM:

Pure APM: Gasol 11.9, Fish 1.8, Artest -0.2, Kobe -8.5
rAPM : Gasol 3.9, Fish 2.3, Artest 2.2, Kobe 1.3

Having another element at work beyond pure APM solving with the given lineups (some of which suffer from colinearity) provides an interesting perspective on those pairs, trios and quartets. rAPM rates Luol Deng at 3.6, Derrick Rose at 3.4. APM has Rose at 11.6 (SE 8.8 (!)) Deng 1.0 (SE 5.8). 85.7% of Rose's minutes are spent in the company of Deng, next closest companion is Boozer at 56%.

Having rAPM samples that have just one regular season (and perhaps just one post-season) would allow for apples-to-apples comparisons with B-Value.com APM.

Along those lines...

I've been working tonight on modifying my code to look at WOWY at the play-by-play level. Here are some initial results for Chicago:

W/O Rose W/O Deng W/O Boozer W/O Noah

Using the total EZPM100 data and the newly provided without Rose data for players to find the with Rose data, here is the summary:
4 better with Rose than without on EZPM 100, 1 tie, 3 worse with.

not taking sides in any debate about win shares, +/-, adjusted +/-, or any kind of +/-. but...

Al Jefferson having almost identical win shares compared to Steve Nash this year is a joke...

al jefferson was actually quite good in 2010-11. he scored 1528 points with just 105 turnovers, and if you check the statistical history of the league you'll find that since 1977-78 (when the league first kept track of turnovers) of the 942 times a player scored 1450 or more points in a single season, jefferson's 105 TOs were the fewest. even though his overall shooting or ScFG% (2s, 3s, and FTs) of 52.2% was less than that of the league average C (53.9%), and ranked just 45th among the 65 Cs to play at least 500 minutes in 2010-11, his turnovers were so low such that his offensive efficiency (pts/0ptposs or points scored per zero point team possession personally responsible for) actually ranked 7th highest among those same 65 Cs...

he rebounded a bit better than the league average C, blocked shots at a rate a bit better than the league average C, and committed fouls at a rate almost 1/3 less than did the league average C. if you look at defensive ON/OFF data for utah with/without him in 2010-11:

http://www.82games.com/1011/10UTA15.HTM#onoff

ON OFF
2940 1035 minutes
113.3 106.0 pts/100poss allowed (+7.3 pts/100poss worse)

you'd get the impression he was poor on defense because the team was poor on defense when he played. but notice that he played 3 times as many minutes ON versus OFF, thus that OFF data is much more likely to be skewed because it's a much smaller sample size of data. this is similar to data for chicago with/without derrick rose in 2010-11:

http://www.82games.com/1011/10CHI3.HTM#onoff

ON OFF
3027 937 minutes
103.0 95.3 pts/100poss allowed (+7.7 pts/100poss worse)

this is an even worse differential on defense than with jefferson, so one might get the impression that rose was not good on defense based on these team pts/100poss values. but again he was ON the floor for over 3 times as much time as he was OFF, so that OFF data has a much higher chance of being skewed. also Synergy data shows defensive numbers for rose than are extremely good, and defensive numbers for jefferson that are good...

Utah was -3.1 net per 100 poss with Jefferson on the court this year compared to +4.7 for Nash, despite them both being starters on 39-40 win teams...

this is not an individual player rating of any kind, but is team dependent. among the 6 players to play the most minutes on utah in 2010-11 (al jefferson, paul millsap, raja bell, deron williams, and andrei kirilenko), millsap had the best ON/OFF at +5.8:

http://www.82games.com/1011/1011UTA.HTM

and jefferson had the worst at -3.6. but according to StatsCube:

http://www.nba.com/statscube/player-vs- ... 4;season=r

millsap played 84% of his minutes with jefferson. so which is it? is jefferson's overall +/- so "bad" because he had to play so much with millsap, or is millsap's overall +/- so "good" because he played so much with jefferson? the fact is millsap played just 16% of his minutes in 2010-11 without jefferson...

or is it that in reality the overall ON/OFF team data is not really telling us anything concrete about any individual player?...

...to me that explains why we see absolutely crazy results like equating Jefferson and Nash, despite that not passing the eye test...

what exactly was it that you saw with your eyes that convinced you nash was a better player than jefferson in 2010-11? i ask because as good of an offensive player that steve nash was in 2010-11 (and previous seasons):

- he played the 2nd most total minutes on a 2010-11 suns team that was the league's 6th worst team defensively (and the player to play the 3rd most total minutes on the suns - grant hill - finished 7th in DPOY voting)...

- he played the most total minutes (7641) of any player on phoenix over the past 3 seasons (1/8 of the team's total minutes played) and phoenix was the league's 7th worst team defensively (7th most pts/poss allowed) during this time...

bchaikin wrote: you'd get the impression he was poor on defense because the team was poor on defense when he played. but notice that he played 3 times as many minutes ON versus OFF, thus that OFF data is much more likely to be skewed because it's a much smaller sample size of data. this is similar to data for chicago with/without derrick rose in 2010-11:

I think this is a really important point that I come up against again and again on various forums and have tried to explain. Most people only consider the sample size of the player they are looking at his on/off. If he played 80% of the team minutes, they say it is a large sample size, but the problem is that the minutes he was off is a small sample size. I think this is an issue with Monta Ellis who many Warriors fans bring up how bad his +/- is. The fact is that he was only off the floor around 700 minutes this season (19% of the total minutes).

Of course, coaches don't give out minutes to satisfy our desires for proper sample size, but it seems to me the ideal would be for every player to play an equal number of minutes (10 guys playing 50% of the minutes in random permutations of units would be ideal). Of course, that doesn't happen. But we can look at players who play somewhere in the range of 40-60% of a team's total minutes (compared to say <30% or >70%), and probably make a stronger statistical case for +/- (even unadjusted).

To analyze lineups, I like reformatting the bv lineup pages. You can do more filters and more complex filters that way compared to NBA.com info. Here's the Bulls' 2011 data. Change the filters and the ratings numbers will change.

A couple things with Rose's defensive numbers. First, CJ Watson is a stud defender, I think better than Rose just from watching them both all season. But more importantly, the Bulls' bench is full of stud defenders. Watson, Brewer, Taj, Asik would all be candidates for all-defense votes given more recognition, and also Luol often plays with those guys since the Bulls don't have a true backup SF. Comparatively, Rose plays a lot with Boozer. Rose+Boozer lineups had a 104 DRating this year, Rose minus Boozer lineups had a 99 DRating. I also think Noah is a worse defender than Asik, though that's probably not a popular opinion. I'm pretty sure based on watching Rose and looking at Synergy numbers that he is an above average PG defender, though I don't think he's elite.

As for "Overall the Bulls were only 2 points better with Rose than without" stat, couple things as well. First, Thibs ran a strict rotation all year, with the bench mob playing at the beginning of the 2nd and 4th quarters of nearly every game, and they usually played against several opponent bench players. This shows up in bv lineup data, as the two most-used bench mob lineups had Adj +/- ratings about 7pts per 100 worse than their raw +/- ratings. Second, Keith Bogans stinks, and he played like 95% of his minutes with Rose on the floor this year. Rose+Bogans lineups were +7 net, Rose minus Bogans lineups were +10 net.

Crow wrote:One year RAPM has Rose at +3.4, Howard +4.4. One year traditional APM has them both easily over +10. It looks like one method got the rating pretty consistent with raw +/- in each case and 1 varied widely and vice versa for the other guy. But which Adjusted +/- is "right" or more might?

IMO neither. It's a 1-year APM, I think it's fairly meaningless. Rose had a +11.7 APM with an 8.8 standard error. That's why I like that the MVP votes usually goes to the best story. I don't know of a way to say for sure who was most/least valuable between Rose/LeBron/Wade/Howard/Dirk/Manu/Garnett/Paul this year, but there's not much fault to find in Rose's 62-20 record, #1 overall seed, or his dozen clutch performances.

bchaikin wrote:al jefferson was actually quite good in 2010-11...

Just from looking at a bunch of rAPM data and comparing it to box score numbers and Synergy numbers to find a correlation, my thoughts...

Jefferson's style helps him load the box score, but he doesn't have much that translates to team efficiency. Team efficiency pluses: he's a decent post defender (so AK/Millsap don't have to shade over) and post scorer (which helps spread the floor and attract some doubles/shades). Unfortunately, he doesn't pass well out of the post to exploit those doubles/shades, and he's not elite enough in the post to draw constant doubles like Howard does, so what's left is just his actual production on post scoring attempts (and threat of post scoring), and he averaged just .89 PPP on post attempts this season, which is barely above NBA average efficiency. He had just a .528 TS% overall. He was also barely used in the PnR, he's a poor help defender, and he had an off year rebounding despite playing with undersized teammates.

In that perspective, how much of what Jefferson was doing was improving his team's efficiency? If a player scores 19ppg on a .528 TS% on a team with other decent scoring options, and he doesn't pass well, how much is he actually helping? Compare that to Nash who shot a 60 TS%, helped keep the floor spread by shooting 43% from 3, led the league with 11.4apg, made probably 60+ additional passes per game that weren't recorded as assists, ran an elite PnR, and offensively every possession he's helping his team play more efficiently.

Going to chime in here about Big Al...
If you look at his splits with Sloan and without, he played muuuuuch better with Corbin. Sloan's offense didn't fit at all, and certainly impacted his boxscore stats. (Negatively) He was the only* Jazzman who "got better" when Deron left.
*I'm not actually positive, I looked into this a little while ago but I believe everyone else shot a bit worse/turnovers increased, or at least that was the story, without Deron. Al Jeff was the exception.

He has one of the more bizarre offensive profiles, as well. Despite being thought of as a post player, (he takes too many jumpers) he rarely draws fouls. As bchai pointed out, he has a historically low turnover rate, as well. Looking at his four factors, he resemblesan old school F who crashes the glass hard. (Lots of midrange shots, ball in hands, generate a lot of ~avg to below avg shots)

Al's raw rebounding numbers look OK, (a 16.3 TRB% I believe is slightly above average for a big - off the top of my head a 16.0 is average?) incorporating counterpart data as well as on/off, he actually shrinks a bit to slightly below average.

The main criticism is that as far back as I can research, Al's defensive numbers look bad. (Combing through counterpart via PBP, on/off, etc.) The Jazz fell from 105.0 last year to 110.1 this year. The most obvious change was Boozer, out (not really known for his defense!) and Al Jeff, in. It's ridiculous to pin that on Al Jeff, but it also doesn't bode well.

As far as Bchai pointing out Nash's defense->the Suns, it sorta brings about another discussion, which is the relative impact of PGs (and guards) on defense... How much do they matter? And, is Nash responsible for the poor defense, or is it that through the Suns' run, they've employed very few able post defenders? (In fact, after they acquired Gortat, and actually played 2 bigs rather than Nash + 4 wings, their defense stabilized closer to league average)

I don't know, probably not the thread for it. That said, throw me in the lot that thinks Nash's value is >> Al Jeff.

If nothing else, Jefferson is an interesting case statistically in that his 4-yr DRAPM (-2.5) suggests he is one of the worst defenders in the league, but his Synergy stats are actually pretty good (0.82 PPP overall, ranked #72; 0.74 PPP on iso, ranked #46). Not a good help defender, I guess.

Crow wrote:Using the total EZPM100 data and the newly provided without Rose data for players to find the with Rose data, here is the summary:

Code: Select all

NAME total EZPM100 w/o Rose with Rose
C.J. Watson	0.6	0.3	1.3
Carlos Boozer	0.2	2.4	-0.8
Joakim Noah	4.3	3.7	3.7
Kyle Korver	-1.4	0.1	-3.2
Luol Deng	2.1	0.7	1.4
Omer Asik	3.7	2.5	6.1
Ronnie Brewer	5.3	4.8	7.4
Taj Gibson	1.9	2.1	1.7

4 better with Rose than without on EZPM 100, 1 tie, 3 worse with.

Interestingly, here is what I get when I look at the most used CHI lineup this season:

The EZPM breakout of that most used lineup is interesting, not exactly in-line with the most common overall narrative or with all other average metric ratings of players.

Lineups are almost always considered small sample but how much of their performance is unique / inherent and how much are they influenced by noise? Or in-between those things, how much by unique match-ups or types of match-ups? I've looked at some lineups year to year, probably should do more.

Overall Rose-Boozer on average is a little better on team +/- than this lineup's sum of performance on EZPM but maybe / probably the difference is because of better opponent quality faced by this lineup than average.

Rose- Brewer- Deng- Gibson- Thomas has very strong +/- in both the regular season and playoffs, far far stronger than the most used lineup. In the regular season the most used lineup got 4 times as many minutes as this alternative. In the playoffs, its minute advantage is 15-1. Maybe the numbers don't tell the whole story but I'd want to hear why Rose- Brewer- Deng- Gibson- Thomas is barely being used.

I'd want to compare these lineups when facing 4+ starters vs 3 or less to decide where to use and not use them. It is still small sample but maybe better to study this data closely but carefully rather than not giving it attention or much attention?


On EZPM it appears Rose and Boozer are both better without the other than together on average. That might suggest lower offensive usage when together and probably less first or best usage looks. A good portion of the good team +/- when they are together is coming from others. That might still be an ok lineup to go with but maybe they are not a necessarily a pairing to maximize minutes together now or keep as a core long-term pairing. I guess it would be good to check back on this next season or the one after that.

Kurt got big minutes when Noah was hurt, so a lot his minutes in general come from those situations. With Noah healthy, all else equal, Omer gets slightly more court time than Kurt (neither play much), and because of the Hawks' athleticism and big man versatility Thibs has gone with Omer all series.

Even better than the Rose-Brewer-Deng-Taj-Kurt lineup was the Rose-Brewer-Deng-Taj-Asik lineup though. Thibs finally featured that lineup in game 5 of the Hawks series, and it won the game with a dominant 4th quarter performance IMO that's the lineup of the Bulls' most efficient player at each position, especially now with Boozer banged up, and they've just rocked all season.

Yeah, that's a pretty good lineup: Although to give you an idea of the sample size (and opponent quality) issue, here is the list of players with above average number of possessions against this unit sorted by ezpm100: