ESPN RPM

[aIvin adams]

i don't like espn's real plus-minus bc i don't think they explain anywhere what 'real' means (ie, how it's calculated)

but they're the flagship so here we go:

http://espn.go.com/nba/statistics/rpm/_/sort/RPM


suns in the top 50:

kieff is the highest rated Sun, at 18.

bledsoe is next, at 32.

IT at 49.

passes eye test - until recently kieff has been our best-performing player night in night out. surprising to me that his 'def RPM' is significantly higher than his 'off RPM,' but again-- i dunno how this is calculated :/

bledsoe is about even on his def/off split.

IT has a big off RPM and a negative def RPM. eye test BING

others: PJ at 129, plums at 164, mook at 175, gerald green at 201, goran dragic at 205, len at 232, ennis at 260, zoran 311, shavlik 313, tj warren 318, archie 383, tolliver 387

so obviously goran dragic sucks and zoran dragic should be getting all of tj warren's minutes

[RunDogGun]

But when we looked at the percentages on the defensive end, IT kept is opponent shooting lower percentages better than all our other guards.

[kennydorglas]

I think height has a great consideration in this formula.
All little guys with the exception of Collison (WTF?) and Bledsoe are underpreciated in DRPM.
In this list we have some notables great defenders barely being a positive one (Beverley, Conley)
even have some guys in negative (Lowry, CP3)

No way Collison is a better defensive prospect than Thomas.
Both of them have blazing speed, awesome lateral quickness (stay in front of his guy everytime), quick hands for steal. I dont see how Collison is a +1 and Thomas is a -1, barring his size advantage.

[Wannabe MEP]

i don't like espn's real plus-minus bc i don't think they explain anywhere what 'real' means (ie, how it's calculated)

Some info here:
http://espn.go.com/nba/story/_/id/10740 ... plus-minus

My breakdown:

PM (plus-minus) — Raw point differential while a player is on the floor. Heavily influenced by who he’s on the court with. (If, let’s say, Kerrsed only ever played with LeBron James, Kevin Durant, Chris Paul, and Anthony Davis, and only ever played against the 76ers, he’d have really good PM numbers.)

APM (adjusted plus-minus) — Take a big sample size of raw PM and churn in through mathematical formulas in an attempt to assign “credit” to players. Every player is a variable, so there are 10 variables on the court at any time. Take all the data from all NBA possessions and see what happens when those different variables interact. (Kerrsed no longer gets credit for being on the court with great players against bad players.)

APM gets a heck of a lot right, but there are some quirks. For example, players X and Y always share the court together, so APM struggles with differentiating credit between them. Huge example of this with the 2004 Pistons: all five starters played together all the time. In the tiny sample size from when they didn’t play together, some players had better PM numbers than the others, so APM had a huge range for those starters, from best player in the league to major negative.

So the question became: How do we take APM and make it better by normalizing it—removing some of the quirks?

Multi-year APM — Grow the sample size enough and you can filter out a lot of the problem areas. Problem is that players and roles change. Kerrsed dominated before he tore his ACL, and he hasn’t been the same since. We really want to know about his impact now.

SPM (statistical plus-minus) — An attempt to normalize APM by mixing other stats in with plus-minus (think PER mashed up with APM). If APM is trying to say that player Kerrsed sucks at offense, it’s helpful to see that he’s shooting 20%. Those things combine to confirm that yes, Kerrsed sucks at offense.

RAPM (regularized adjusted plus-minus) — Add regression techniques to those mathematical formulas to smooth things out. If APM is trying to say there’s a huge gap like +17 for Rasheed and -6 for Billups, regression says, “Hey wait, that sample size is too small to make that big of a claim. I think it’s +7 for Rasheed and +4 for Billups. Rasheed may be more valuable, but he’s probably not THAT much more valuable.

Jeremias Engelmann became sort of the RAPM guru in the advanced stats community, although he definitely wasn’t the only one spinning those numbers. He kept churning out different versions with different ways of weighting things (4-year RAPM all years weighted equally, 3-year RAPM with greater weight to current season, negative priors for rookies, RAPM with SPM priors, etc.).

ESPN finally decided to publish one of Engelmann’s versions of RAPM as "Real Plus-Minus". It’s actually a RAPM-SPM mashup, although I think I remember the SPM part is just used for priors. It’s pretty much xRAPM from this page by Engelmann: http://stats-for-the-nba.appspot.com/.

[Wannabe MEP]

I’m a huge fan of RPM/RAPM/APM, particularly for recognizing the under-the-radar guys who subtly/indirectly impact the game more than people realize. One caveat I want to stress, though, is that RAPM is really about assessing the value of a player with his current team, in his current role. While it wants to extrapolate further, that’s really hard to get right. Context affects impact.

Suppose…

- Player A is the only decent point guard on his team.
- Player B is one of three quality point guards on his team.

Player A will probably have a higher RAPM rating than player B, because he’s more valuable to his current team. That doesn’t mean he’s actually a better player. When he comes in, the team gets a lot better; when he sits, the team really struggles.

In that light, I want to take a look at a couple of our guys’ quirky RPM numbers: Dragic and Markieff.

Markieff RPM
-0.02 — 13-14
+4.03 — 14-15

Dragic RPM
+3.43 — 13-14
-0.97 — 14-15

Did Markieff really get that much better in a year? Did Dragic really get that much worse in a year??

I don’t think so. I think there are a few things at play here for each of them:

Dragic

1) I’ve said it before, and I’ll say it again: Dragic + Frye was something special. Dragic thrives in large amounts of space, which Frye provides as well as perhaps any big. Dragic has played a ton more minutes this year in lineups with significantly less spacing.

2) Dragic played much of the season a year ago as the only quality point guard when Bledsoe was injured. IT + a healthy Bledsoe means that even if Dragic was just as effective this year, he wouldn’t be as valuable to this team. We’re still effective at his position now when he sits.

3) He has flat-out played worse. It seems like his confidence has been shaken this year, probably largely because of both #1 and 2, and because of less touches with all these PGs. That’s partly evident in that he’s gotten some more quality touches recently with IT out and Tucker in the starting lineup and he's played quite a bit better.

Markieff

1) Frye’s absence leaves a serious vacancy in the frontcourt. Markieff is the only quality player taking minutes at the 4 for us right now. The main alternate options here thus far are Marcus and Tolliver, neither of which have been successful (both net negative at the 4). We had a lot more effective non-Markieff lineups last year, such as with Frye-Plumlee frontcourts. Right now there’s nothing really working (consistently) that doesn’t include Markieff, so his RPM numbers make him look like an all-star — RPM has him third in the league among PFs. (I’m still hopeful that Tolliver finds a rhythm with this team and has a more positive impact, and in fact that’s already started happening: he was -11.9 through Nov 15, but he’s +8.3 since.)

2) Markieff is thriving as a small-ball 5. RAPM/RPM doesn’t care whether he’s playing the 4 or the 5. He’s clearly the most skilled of our guys that have played the 5, and he’s the only one who can pass as a “stretch-5.” With our slasher guards, we often kill it when we go five out. Right now Kieff is +5.1 as the 4 (good) and +26.1 as the 5 (WOW!!). He was similarly effective last year when he played next to Frye.

3) He probably did actually improve a fair amount, as young/inexperienced players are wont to do (given playing time).

[lilfishi22]

I’m a huge fan of RPM/RAPM/APM, particularly for recognizing the under-the-radar guys who subtly/indirectly impact the game more than people realize. One caveat I want to stress, though, is that RAPM is really about assessing the value of a player with his current team, in his current role. While it wants to extrapolate further, that’s really hard to get right. Context affects impact.

[sunskerr]

Frye this, Frye that. Mate, we get it- he was beneficial. But he's gone.

[Wannabe MEP]

Fishi, I'm going more for this:



(Kerrsed's idea initially)