2015 APM/RAPM/etc.

[Doctor MJ]

Yeah, I'll wait for JE's PI set, this is full of noise. I just don't believe Harden, Ginóbili, Curry, McLemore or Lillard are close to -3 on defense.

But isn't NPI RAPM what people really want out of +/- stats: an unbiased look at what is happening on the court? I don't see how PI RAPM actually reduces noise; it just gives biased results.

RAPM is always biased. Ridge regression or any other regularization method introduces bias (toward 0) in order to reduce variance.

Right, if you want the purest +/- one-number stat it's APM, not RAPM. I frankly wish we had regular access to that just like we do for RAPM or RPM, though not because I'd only use APM. I went them all. The differences you see when you compare tell a lot.

[Moonbeam]

But isn't NPI RAPM what people really want out of +/- stats: an unbiased look at what is happening on the court? I don't see how PI RAPM actually reduces noise; it just gives biased results.

RAPM is always biased. Ridge regression or any other regularization method introduces bias (toward 0) in order to reduce variance.

Right, if you want the purest +/- one-number stat it's APM, not RAPM. I frankly wish we had regular access to that just like we do for RAPM or RPM, though not because I'd only use APM. I went them all. The differences you see when you compare tell a lot.

I actually think something like elastic net or its variations might be an improvement on RAPM in terms of a regularized approach. You'd probably get more bang for your buck in terms of reduced variance at a cost of the incurred bias through shrinkage. I posted in the KG thread about it in case you're interested.

[Doctor MJ]

RAPM is always biased. Ridge regression or any other regularization method introduces bias (toward 0) in order to reduce variance.

Right, if you want the purest +/- one-number stat it's APM, not RAPM. I frankly wish we had regular access to that just like we do for RAPM or RPM, though not because I'd only use APM. I went them all. The differences you see when you compare tell a lot.

I actually think something like elastic net or its variations might be an improvement on RAPM in terms of a regularized approach. You'd probably get more bang for your buck in terms of reduced variance at a cost of the incurred bias through shrinkage. I posted in the KG thread about it in case you're interested.

Oh certainly. To be honest, as you may have noticed, I've had misgivings with every advancements after APM. It's not that I think any of them are inherently bad, it's just that so many people don't seem to understand that RAPM isn't an upgrade to APM, it's a different stat with different strengths and weaknesses. By no means would I think it a given that some other sophisticated technique couldn't give us other strengths.

[Doctor MJ]

Yeah, the more of the NPI I'm seeing, the more I'm starting to not believe in its real usefulness. It passes some of the general "sniff test" (Harden/Curry), then I'm seeing Z Bo with very good defense, same for Lillard (who I don't think is as bad as his rep, but not that good), Beal rating much more highly than I'd put him, etc. I'm becoming less a fan of this as I go on, as strange as it may seem. I get there's a lot of noise, but it seems the noise is starting to outweigh the usefulness to my mind.

Your feelings are understandable.

Here's what I'll say:

1 - Remember that there are other factors that could make the data look odd other than noise. I gave the example before about Westbrook's defensive rebounding actually being something that helps his offense at the expense of defense, that's tricky, but still relatively "fair". It can get weirder: A guy getting steals whose actually helping the offense more with his steals than the defense. A guy who "helps" defense because when he's on the floor they let the offense flow through him and other guys can concentrate on defense. A guy who is getting played selectively and thus isn't on the floor when the worst mismatches are out there.

2 - I'm very cautious about using these stats as something beyond tier placement and general estimation for most players, in part because your average player isn't important enough to me that I'll look at every little detail about him. As such, I tend to look at this data to see things more along the lines of whether a guy is "working" for a team as they currently use him as opposed to that guy actually being X good.

3 - With the big name players though I do try to understand what's going on with them, particularly when we see something weird, and while I can often figure oddities out, I can't always. Up until the point where I feel I understand where a guy's +/- numbers are coming from, I'm cautious about using them to push my assessment of the player.

[acrossthecourt]

RAPM is always biased. Ridge regression or any other regularization method introduces bias (toward 0) in order to reduce variance.

Right, if you want the purest +/- one-number stat it's APM, not RAPM. I frankly wish we had regular access to that just like we do for RAPM or RPM, though not because I'd only use APM. I went them all. The differences you see when you compare tell a lot.

I actually think something like elastic net or its variations might be an improvement on RAPM in terms of a regularized approach. You'd probably get more bang for your buck in terms of reduced variance at a cost of the incurred bias through shrinkage. I posted in the KG thread about it in case you're interested.

Elasticnet drops variables, so I don't see why that's useful when you want a ranking of every player in the league. Heavy penalization keeps the low minute guys from being +10 or -10, and in elasticnet they wouldn't even be used as variables.

What you want is a better way to control the variables through Bayesian methods, which is what was done. I know people don't want to see box score stats mixed in, but that improves the results. You can try different things through. Use a prior that's a multi-year RAPM? Adjust the weights based on how many teams a guy has played on? i.e. the more teams you've played on, the less you get penalized because you're in new environments.

[Moonbeam]

Right, if you want the purest +/- one-number stat it's APM, not RAPM. I frankly wish we had regular access to that just like we do for RAPM or RPM, though not because I'd only use APM. I went them all. The differences you see when you compare tell a lot.

I actually think something like elastic net or its variations might be an improvement on RAPM in terms of a regularized approach. You'd probably get more bang for your buck in terms of reduced variance at a cost of the incurred bias through shrinkage. I posted in the KG thread about it in case you're interested.

Elasticnet drops variables, so I don't see why that's useful when you want a ranking of every player in the league. Heavy penalization keeps the low minute guys from being +10 or -10, and in elasticnet they wouldn't even be used as variables.

What you want is a better way to control the variables through Bayesian methods, which is what was done. I know people don't want to see box score stats mixed in, but that improves the results. You can try different things through. Use a prior that's a multi-year RAPM? Adjust the weights based on how many teams a guy has played on? i.e. the more teams you've played on, the less you get penalized because you're in new environments.

I don't think dropping variables (i.e. assigning certain players an estimate of 0) is necessarily a bad thing, especially since a lot of RAPM scores are probably not significantly different from zero anyway. As it stands, I've seen RAPM used often to draw conclusions that may not be appropriate, noting differences in the estimates when there may be significant overlap in credible intervals.

[JLei]

Anyone got insider that can compile the multi-year RPMs in this story?

http://espn.go.com/nba/insider/story/_/id/12727323/ranking-all-80-playoff-starters-nba

[Atmanne]

Anyone got insider that can compile the multi-year RPMs in this story?

http://espn.go.com/nba/insider/story/_/id/12727323/ranking-all-80-playoff-starters-nba

1. Stephen Curry, Golden State Warriors
Base winning pct.: .758 | Multiseason RPM: 8.51 | Rating: 8.83

2. LeBron James, Cleveland Cavaliers
Base winning pct.: .744 | Multiseason RPM: 8.45 | Rating: 8.49

3. Chris Paul, Los Angeles Clippers
Base winning pct.: .730 | Multiseason RPM: 7.89 | Rating: 7.92

4. Anthony Davis, New Orleans Pelicans
Base winning pct.: .748 | Multiseason RPM: 6.69 | Rating: 7.71

5. James Harden, Houston Rockets
Base winning pct.: .712 | Multiseason RPM: 7.28 | Rating: 7.25

6. Kawhi Leonard, San Antonio Spurs
Base winning pct.: .644 | Multiseason RPM: 8.05 | Rating: 6.36

7. Tim Duncan, San Antonio Spurs
Base winning pct.: .664 | Multiseason RPM: 3.86 | Rating: 4.61

8. Kevin Love, Cleveland Cavaliers
Base winning pct.: .661 | Multiseason RPM: 3.63 | Rating: 4.45

9. DeAndre Jordan, Los Angeles Clippers
Base winning pct.: .640 | Multiseason RPM: 4.17 | Rating: 4.35

10. John Wall, Washington Wizards
Base winning pct.: .608 | Multiseason RPM: 4.78 | Rating: 4.10

11. Kyle Lowry, Toronto Raptors
Base winning pct.: .618 | Multiseason RPM: 4.42 | Rating: 4.10

12. Draymond Green, Golden State Warriors
Base winning pct.: .559 | Multiseason RPM: 5.94 | Rating: 3.90

13. Danny Green, San Antonio Spurs
Base winning pct.: .600 | Multiseason RPM: 4.39 | Rating: 3.79

14. Paul Millsap, Atlanta Hawks
Base winning pct.: .614 | Multiseason RPM: 3.88 | Rating: 3.76

15. Damian Lillard, Portland Trail Blazers
Base winning pct.: .595 | Multiseason RPM: 4.32 | Rating: 3.66

16. Marc Gasol, Memphis Grizzlies
Base winning pct.: .607 | Multiseason RPM: 3.49 | Rating: 3.45

17. Tyson Chandler, Dallas Mavericks
Base winning pct.: .594 | Multiseason RPM: 3.73 | Rating: 3.35

18. Blake Griffin, Los Angeles Clippers
Base winning pct.: .612 | Multiseason RPM: 2.84 | Rating: 3.20

19. Kyle Korver, Atlanta Hawks
Base winning pct.: .537 | Multiseason RPM: 5.16 | Rating: 3.15

20. LaMarcus Aldridge, Portland Trail Blazers
Base winning pct.: .567 | Multiseason RPM: 4.17 | Rating: 3.13

21. Jimmy Butler, Chicago Bulls
Base winning pct.: .580 | Multiseason RPM: 3.58 | Rating: 3.04

22. Kyrie Irving, Cleveland Cavaliers
Base winning pct.: .602 | Multiseason RPM: 2.53 | Rating: 2.89

23. Klay Thompson, Golden State Warriors
Base winning pct.: .552 | Multiseason RPM: 4.00 | Rating: 2.80

24. Mike Conley, Memphis Grizzlies
Base winning pct.: .593 | Multiseason RPM: 2.61 | Rating: 2.7

25. Zach Randolph, Memphis Grizzlies
Base winning pct.: .550 | Multiseason RPM: 3.77 | Rating: 2.66

26. Tiago Splitter, San Antonio Spurs
Base winning pct.: .553 | Multiseason RPM: 3.59 | Rating: 2.61

27. Andrew Bogut, Golden State Warriors
Base winning pct.: .536 | Multiseason RPM: 4.03 | Rating: 2.58

28. Al Horford, Atlanta Hawks
Base winning pct.: .592 | Multiseason RPM: 2.06 | Rating: 2.48

29. Khris Middleton, Milwaukee Bucks
Base winning pct.: .480 | Multiseason RPM: 5.41 | Rating: 2.40

30. Joakim Noah, Chicago Bulls
Base winning pct.: .573 | Multiseason RPM: 2.36 | Rating: 2.32

31. Dwight Howard, Houston Rockets
Base winning pct.: .580 | Multiseason RPM: 2.02 | Rating: 2.27

32. Pau Gasol, Chicago Bulls
Base winning pct.: .603 | Multiseason RPM: 1.23 | Rating: 2.25

33. Deron Williams, Brooklyn Nets
Base winning pct.: .540 | Multiseason RPM: 2.85 | Rating: 2.05

34. Dirk Nowitzki, Dallas Mavericks
Base winning pct.: .571 | Multiseason RPM: 1.66 | Rating: 1.94

35. Zaza Pachulia, Milwaukee Bucks
Base winning pct.: .497 | Multiseason RPM: 3.85 | Rating: 1.87

36. Brook Lopez, Brooklyn Nets
Base winning pct.: .609 | Multiseason RPM: -0.10 | Rating: 1.68

37. Marcin Gortat, Washington Wizards
Base winning pct.: .536 | Multiseason RPM: 1.77 | Rating: 1.45

38. Tyreke Evans, New Orleans Pelicans
Base winning pct.: .527 | Multiseason RPM: 2.07 | Rating: 1.45

39. Jeff Teague, Atlanta Hawks
Base winning pct.: .565 | Multiseason RPM: 0.61 | Rating: 1.33

40. Chandler Parsons, Dallas Mavericks
Base winning pct.: .519 | Multiseason RPM: 1.98 | Rating: 1.29

Players 41-80Rank Player Team Base win % MS RPM Rtg.
41. DeMarre Carroll Atlanta Hawks .530 1.56 1.24
42. Paul Pierce Washington Wizards .556 0.71 1.22
43. Robin Lopez Portland Trail Blazers .511 1.77 1.05
44. Timofey Mozgov Cleveland Cavaliers .493 1.96 0.88
45. Trevor Ariza Houston Rockets .524 0.91 0.83
46. Monta Ellis Dallas Mavericks .476 2.30 0.77
47. Matt Barnes Los Angeles Clippers .470 2.39 0.74
48. Mike Dunleavy Jr. Chicago Bulls .470 2.30 0.69
49. Nicolas Batum Portland Trail Blazers .531 0.35 0.65
50. Bradley Beal Washington Wizards .467 2.31 0.64
51. Terrence Jones Houston Rockets .539 -0.09 0.56
52. J.J. Redick Los Angeles Clippers .499 0.88 0.42
53. Joe Johnson Brooklyn Nets .469 1.67 0.36
54. Thaddeus Young Brooklyn Nets .491 0.94 0.33
55. J.R. Smith Cleveland Cavaliers .499 0.62 0.29
56. Omer Asik New Orleans Pelicans .495 0.28 0.06
57. Marcus Smart Boston Celtics .443 1.85 0.04
58. Giannis Antetokounmpo Milwaukee Bucks .453 1.44 0.00
59. Tyler Zeller Boston Celtics .491 0.16 -0.07
60. Jonas Valanciunas Toronto Raptors .542 -1.52 -0.10
61. Tyler Hansbrough Toronto Raptors .454 1.04 -0.19
61. Ersan Ilyasova Milwaukee Bucks .537 -1.52 -0.19
63. Nene Hilario Washington Wizards .438 1.20 -0.36
64. Jason Terry Houston Rockets .470 -0.15 -0.53
65. DeMar DeRozan Toronto Raptors .453 0.13 -0.66
66. Derrick Rose Chicago Bulls .471 -0.53 -0.70
67. Michael Carter-Williams Milwaukee Bucks .491 -1.49 -0.88
68. Harrison Barnes Golden State Warriors .431 -0.61 -1.37
69. Courtney Lee Memphis Grizzlies .436 -0.86 -1.42
70. Avery Bradley Boston Celtics .380 0.54 -1.58
71. Eric Gordon New Orleans Pelicans .454 -1.75 -1.59
72. Tony Parker San Antonio Spurs .477 -2.56 -1.64
73. Brandon Bass Boston Celtics .445 -1.73 -1.71
74. Terrence Ross Toronto Raptors .433 -1.73 -1.89
75. Rajon Rondo Dallas Mavericks .477 -3.37 -2.03
76. Quincy Pondexter New Orleans Pelicans .408 -1.79 -2.31
77. Evan Turner Boston Celtics .419 -2.32 -2.40
77. Jeff Green Memphis Grizzlies .462 -3.64 -2.40
79. Arron Afflalo Portland Trail Blazers .400 -2.30 -2.69
80. Markel Brown Brooklyn Nets .406 -3.30 -3.10

[SideshowBob]

Tossed that into a GoogleSheet and sorted by RPM.

[colts18]

[SideshowBob]

So this is multi-year 2015 RAPM through the RS. Link

Here's the 2014 set for comparison: https://stats-for-the-nba.appspot.com/ratings/2014.html

And here is the same 2015 set through mid-January: https://stats-for-the-nba.appspot.com/ratings/RAPM_2015.html

NOTE: This isn't PI RAPM, like Engelmann was posting from 01-12 (which is what's included in colts' GoogleDoc), or like GotBuckets lists from 08-14. Those are single-year with a prior, this is straight up multi-year, with more weight on the current year.

[ceiling raiser]

Just wanted to link to this (J.E. taking RAPM requests):

http://www.apbr.org/metrics/viewtopic.php?f=2&t=8964

If anybody has any requests, might be worth it to stop by and ask.

[colts18]

15 year RAPM

https://docs.google.com/spreadsheets/d/1CA4KxmzjZrTlYqxNU85jkUnCcqvJjsP5LT818LSYjkk/edit#gid=0

Top 10

Name Offense Defense Total
LeBron James 6.86 2.39 9.25
Kevin Garnett 1.91 7.2 9.11
Chris Paul 6.54 1.01 7.55
Tim Duncan 1.56 5.71 7.27
Dirk Nowitzki 4.97 1.88 6.85
Manu Ginobili 4.92 1.9 6.82
Steve Nash 6.94 -0.58 6.36
Draymond Green 2.1 3.85 5.95
LaMarcus Aldridge 2.56 3.18 5.75
Rasheed Wallace 1.26 4.25 5.51

[bondom34]

^^
Wow that spreadsheet had some shockers for me.

[colts18]

^^
Wow that spreadsheet had some shockers for me.

I'm guessing it's because of Durant. It's most likely due to his first 2 seasons where he was among the worst players in the league. His raw plus/minus numbers those years are atrocious. Since 2010, he has done fine in that area. Once he was moved from SG, he started to thrive.

[bondom34]

^^
Wow that spreadsheet had some shockers for me.

I'm guessing it's because of Durant. It's most likely due to his first 2 seasons where he was among the worst players in the league. His raw plus/minus numbers those years are atrocious. Since 2010, he has done fine in that area. Once he was moved from SG, he started to thrive.

Yeah him for sure, Westbrook too, Aldridge as well. Plus I never really realized how good Baron Davis was I think as well as Jeff Foster.

[cpower]

^^
Wow that spreadsheet had some shockers for me.

I'm guessing it's because of Durant. It's most likely due to his first 2 seasons where he was among the worst players in the league. His raw plus/minus numbers those years are atrocious. Since 2010, he has done fine in that area. Once he was moved from SG, he started to thrive.

Yeah him for sure, Westbrook too, Aldridge as well. Plus I never really realized how good Baron Davis was I think as well as Jeff Foster.

Amir Johnson is second coming of Wilt apparently lol

[SideshowBob]

2015 PI RAPM posted as well.

EDIT: May just be multi-year RAPM again (smaller weight the farther we go back).

[eminence]

Can someone try to explain to me in layman's terms why Exum does so well (relatively) in these stats? I really like him as a player and would like to be able to defend him adequately when he gets criticized. Thanks

[RebelWithACause]

^^
Wow that spreadsheet had some shockers for me.

I'm guessing it's because of Durant. It's most likely due to his first 2 seasons where he was among the worst players in the league. His raw plus/minus numbers those years are atrocious. Since 2010, he has done fine in that area. Once he was moved from SG, he started to thrive.

Yeah him for sure, Westbrook too, Aldridge as well. Plus I never really realized how good Baron Davis was I think as well as Jeff Foster.

If you never realized how good Baron was you may not trust your eye-test anymore and maybe should really rely solely on numbers