Who is the most valuable NBA player?

[Mike G]

Minnesota is playing at +2 with Love, +3 without. I know he'll get votes and it makes me cringe. Obviously his On/Off is bad because he's playing on a elite team and has elite PFs/Cs that back him up..... . He'll have to thank Rubio and Anthony Tolliver. If anyone's really considering to vote for Love, then please go watch some actual games of Minnesota instead of looking at BoxScores. I'm sure if you've done that, there's no way you still think he's one of the best players in the league

According to this site: http://basketballvalue.com/teamplayers. ... 2&team=MIN
Kevin Love has a 1-yr APM of +10.17 (and a 2-yr of +8.89)
Finding that the Wolves play 93.2 possessions per 48 minutes, here are their b-v APM and corresponding points above NBA avg contributed:

Code: Select all

.T-Wolves         Min  poss     apm   pts
Rubio,Ricky       477   926    13.8   128
Love,Kevin        594  1153    10.2   117
Tolliver,Anthony  339   658    16.9   111
Beasley,Michael   221   429     3.3    14

Milicic,Darko     279   542    -3.2   -18
Ellington,Wayne   268   520    -4.2   -22
Randolph,Anthony  152   295   -14.6   -43
Johnson,Wesley    334   649    -7.4   -48
Williams,Derrick  277   538   -16.0   -86
Ridnour,Luke      478   928   -16.9  -157

Not sure how Ridnour can be that bad and Rubio that good.
Just realized these APM do not include last night's game. Rubio (1-11 FG) was +2, Ridnour -1, Love -5 (-8 before final play?)

[J.E.]

Mike, APM with less than one month of data is absolutely worthless. Even APM with a whole season of data is pretty much wortless (you're better off saying everyone contributes 0 than trust one year APM)

[Mike G]

So, why is raw on/off +/- a fine way to find a "clear cut mvp candidate", and the same data adjusted for a player's opponents and teammates is "absolutely worthless"? What is missing from this story?

I did mention above that Love's 2-year APM was +8.89, and the SE is half that.

[J.E.]

What is missing from this story?

The fact that APM is overfitting the data like crazy, leading to funky results that make no sense and don't predict well

[Mike G]

So, unadjusted PM would have us think the best players for the Bulls are Gibson and Korver, while the worst are Boozer, Noah, Deng, Watson, and Rose.
APM indicates Deng and Rose, at least, among their top 5. I don't know which is funkier.

[J.E.]

That's cherry picking at it's best. You say you'd rather believe APM, which, for the team you used as an example, has Hamilton 75 points per game better than Noah?

The shortcomings of non-multiyear APM are well documented. Raw +/, at the extremes(!), can at least give you valuable information on who is having an exceptional good/bad season. I will not discuss this any further

[Mike G]

Sorting Atlanta players by minutes, there's not much correlation with players' minutes and either adjusted or unadjusted plus-minus.
http://basketballvalue.com/teamplayers. ... 2012&team=
Bos - hi correlation in unadjusted, not as much in adjusted
Cha - some corr. in unadjusted, a bit less in adjusted
Chi - strong corr in adjusted, entirely negative in unadjusted
Cle - no obvious correlation in either
Dal - bad correlation in adjusted, worse in unadjusted
Det - good correlation in adjusted, totally negative in unadjusted
http://basketballvalue.com/teamplayers. ... 2&team=DET
etc.

Perhaps some teams just tend to start slowly, and thanks to greater depth in their rosters eventually wear down most opponents.
On such teams, unadjusted seems to reward the 2nd-stringers who have mostly outscored other 2nd-strings.
Or in the Pistons case, just doing well in garbage time may explain it.
Adjusted plus-minus may be less apt to make that mistake, at least. Even after 12 games or so, there aren't teams with starters looking like the worst players on the team.

And I don't especially advocate APM, since there are so many anomalies. Neither do I detect that unadjusted is more consistent or meaningful.

Then there is the issue of a player being called more valuable because, when he misses games, the team flounders. Yet when Shaq would annually miss games, and his team would flounder, people said, "He can't be MVP; his team depends on him, and he misses too many games".
It's a two-edged sword.

[Mike G]

I've changed the name of the thread, and the poll question, to clarify that the poll does not solicit guesses as to who will win the MVP Award, but who is the most valuable player in the NBA.

[mystic]

And I don't especially advocate APM, since there are so many anomalies. Neither do I detect that unadjusted is more consistent or meaningful.

I don't understand what you are even arguing about. J.E. is correct, overfitting is a huge issue with APM. We have the example of Williams/Young on the 76ers, where normal APM gives all the good things to Young and all the bad to Williams, while RAPM has a positive value for both.
Unadjusted +/- at least tells us how a team performed with a certain player on the court in comparison to without him. We can draw conclusions like saying the team is winning during the time a player is on the court or when he is not on the court. And when a team is winning during the time the player is not on the court while being outscored during the time the player is on the court, it pretty much tells us exactly this. Unadjusted +/- is NOT a player ranking tool, that much should be clear.

Anyway, I think James is the MVP right now.


@J.E.

Nowitzki was not considered MVP last season, because he missed games and he was the one getting the blame for the "failure" of the Mavericks in the playoffs in years before. The MVP award depends on a couple of somehow subjective factors and is influenced by previous seasons. I tried to model the voting results of the past and I came up with a formula to predict past winners. In all years from 1975-76 to 2009-10 it has all the MVP winners scoring the most "MVP"-points. It includes factors like season and previous season record for the team, previous season MVP voting, a simple performance metric for that season and the comparison to the previous season, injuries to supposed to be top level players on that team. In essence: when a player's team exceeds record expectations and he himself is seen as a high performance player, he pretty much is always seen as a MVP candidate. When OKC improves and Durant keeps his level, he is pretty much a lock for MVP unless the Magic will end up with a much better record as expected and Howard improves his performance level. Chris Paul would have been another candidate, but he will probably miss too many games.

[Mike G]

And I don't especially advocate APM, since there are so many anomalies. Neither do I detect that unadjusted is more consistent or meaningful.

I don't understand what you are even arguing about. J.E. is correct, overfitting is a huge issue with APM. We have the example of Williams/Young on the 76ers, where normal APM gives all the good things to Young and all the bad to Williams, while RAPM has a positive value for both....

It didn't seem to me that I was making any argument, but inquiring about the information J.E. was using to come up with his MVP suggestions. And I gave some examples of conflicting evidence.

I suspect others are not familiar with the nuances of these various plus-minus measures. So it would be generally enlightening to know about these issues, I think.

And where do you find these RAPM numbers?

[J.E.]

And where do you find these RAPM numbers?

http://stats-for-the-nba.appspot.com/ranking12_no_prior lists RAPM player ratings using data from this season only. If you're not that familiar with it: The values are rather low because it is a very "cautious" metric. Players need to consistently perform above or below average before RAPM gives them a rating far from zero.

Here's a definition for MVP which I liked (although I had to cut off the weird part about having to be a good story for the meda):

Most valuable guy on a regular season team that won a lot of games (preferably most in the conference), who has a perceived significant gap between himself and the second best player on his team

Looking at Win% the teams with the best chances to finish 1st in their conference are Chicago, Indiana, Orlando for the east. Oklahoma, Denver, Utah for the West.

I don't think anyone would argue that there's a significant gap between the best and 2nd best player for either Indiana or Denver, that leaves us with 4 teams.
Chicago: Rose is perceived as the best player (with a sign. gap to the 2nd best) by PER. WS has him almost tied with Boozer and Deng. RAPM has Deng or Gibson at #1, depending on which version you're looking at.
Oklahoma: BoxScore metrics have Durant as the best player, PER says there's gap to Westbrook and Harden. WS says Harden has the same value. RAPM has Collison #1, Durant #2
Utah: Clearly led by Millsap according to PER/WS/RAPM
Orlando: PER and WS say Anderson and Howard are performing equally well, one year RAPM agrees.

Paul Millsap for MVP! (they had twice as many home as away games, I expect them to bounce back to earth soon)

[Bobbofitos]

Minnesota is playing at +2 with Love, +3 without. I know he'll get votes and it makes me cringe. Obviously his On/Off is bad because he's playing on a elite team and has elite PFs/Cs that back him up..... . He'll have to thank Rubio and Anthony Tolliver. If anyone's really considering to vote for Love, then please go watch some actual games of Minnesota instead of looking at BoxScores. I'm sure if you've done that, there's no way you still think he's one of the best players in the league

late to the party here, and maybe this was a while ago, but.. Eesh. I have been watching Minnesota - and I do think KLove is one of the best in the league - so what am I missing?

I voted LBJ, for the record.

[J.E.]

late to the party here, and maybe this was a while ago, but.. Eesh. I have been watching Minnesota - and I do think KLove is one of the best in the league - so what am I missing?

Do you think his offense is so good that it's enough to offset his bad defense, or do you think his defense is above average? I haven't watched all MIN games this year, but in those I've seen I saw him:
-literally ripping rebounds out of his teammates hands (not saying he's not a great rebounder but most people will read his 15RPG and might be impressed a little more than they should be)
-doing absolutely nothing when defending pick and rolls. Nothing. He just stands behind the man setting the screen. That sure doesn't put too much pressure on the offense. Also, opponent guards blew by him rather easily in P&R situations
-not contending many shots. There were multiple situations where he didn't even bother to raise his arm

Going by adjusted TO/REB/PPS the biggest issue appears to be influence on own and opponent turnovers, with the latter being the bigger problem. This is just me guessing and there is a good chance that he really is one of the best players, but you can try to keep an eye out for these things, anyway.

[@48_Minutes]

late to the party here, and maybe this was a while ago, but.. Eesh. I have been watching Minnesota - and I do think KLove is one of the best in the league - so what am I missing?

I voted LBJ, for the record.

I look at the stats and Love looks like an MVP candidate but when I watch Minnesota play, I just don't see Love impacting plays in the quantity or quality like Lebron or Dwight and I am not sure he's even on the second tier with someone like KD. It might only be a matter of time before Rubio moves ahead of Love as the most value T'Wolf.


I voted Dwight

[bbstats]

Eeeee. I love it. Box score versus plus-minus!

Love is definitely overrated, but as I noted before (here), APM and rAPM treat him quite differently.

As Daniel has suggested, perhaps the priors are inflating older players? Overfitting is still possible in the rAPM world, right?

EDIT: I mean, it makes sense that rAPM could be underrating younger players as it hasn't built 'confidence' about them and is therefore 'regressing' them. Maybe?