Most Improved Player. That’s not so easy to define, although there’s much less at stake than, say, with the MVP award. Among the names being mentioned this year are **Kevin Love**, **LaMarcus Aldridge**, **Eric Gordon**, **Dorell Wright**, and even **Derrick Rose** (can you win this and MVP?). Of course, mine is a statistics-focused blog, so I need a way to measure it. MIP shouldn’t just go to the guy who plays more minutes. In that case, **Dorell Wright** would easily win the award. Of course, **Blake Griffin** played 0 minutes last season, so by that criterion…Yeah, it’s really not about minutes. It should be about who played better in those minutes. The only criterion I have about minutes is that the player should have a major role on his team.

In the other awards that I’ve done so far (MVP, DPOY, and 6th Man), I’ve blended ezPM and 1-yr RAPM, but I’m going to do it a little differently for this one. I just got my hands on 2009-2010 1-yr RAPM data (courtesy of Jeremias Engelmann), and am planning to test the year-to-year correlation, but I expect that there is probably quite a bit of variance. I wouldn’t want large random errors to skew the blend, so I did this in two stages. First, I found the players with the biggest positive deltas according to ezPM100. Next, I calculated the change of 1-yr RAPM, and eliminated any players that were negative. I figure if both metrics can’t agree that the player improved, he’s probably not the best candidate, anyway. By the way, players eliminated after this step who would have been in contention according to ezPM alone included **Kwame Brown** (would’ve been #1 otherwise), **Chuck Hayes**, **Kevin Love**, **Tyler Hansbrough**, **Rudy Gay**, **Jordan Farmar**, and **Ray Allen**.

Here are the top 15 candidates, sorted by the change in their WARP per 36 minutes, along with their RAPM data and number of possessions in each year:

### The City Top 15 Most improved

RANK |
NAME |
POS |
DELTA |
WARP11 |
WARP10 |
RAPM11 |
RAPM10 |
DELTA |
POSS10 |
POSS11 |
%-PT |

1 | Thaddeus Young |
3.5 | 0.087 | 0.090 | 0.003 | 2.6 | 0.05 | 2.55 | 3925 | 4003 | 2% |

2 | Jodie Meeks |
1.98 | 0.080 | 0.097 | 0.017 | 1.7 | -0.35 | 2.05 | 1281 | 3914 | 206% |

3 | Emeka Okafor |
5 | 0.079 | 0.134 | 0.055 | 2.9 | -0.30 | 3.20 | 4433 | 4181 | -6% |

4 | Ronnie Brewer |
2.5 | 0.078 | 0.184 | 0.106 | 1.5 | -0.90 | 2.40 | 3329 | 3265 | -2% |

5 | Darrell Arthur |
3.99 | 0.077 | 0.050 | -0.027 | 0 | -2.00 | 2.00 | 848 | 3060 | 261% |

6 | Eric Gordon |
2 | 0.076 | 0.123 | 0.047 | 1.4 | -0.45 | 1.85 | 4251 | 4062 | -4% |

7 | Corey Brewer |
2.5 | 0.072 | 0.060 | -0.012 | -0.4 | -0.85 | 0.45 | 4688 | 3053 | -35% |

8 | JaVale McGee |
4.99 | 0.070 | 0.066 | -0.004 | 1 | -1.70 | 2.70 | 1683 | 4021 | 139% |

9 | Jose Barea |
1.02 | 0.068 | 0.139 | 0.071 | 0.5 | -1.45 | 1.95 | 2755 | 3130 | 14% |

10 | Kris Humphries |
4.49 | 0.067 | 0.157 | 0.090 | -0.5 | -1.30 | 0.80 | 2124 | 3753 | 77% |

11 | Derrick Rose |
1 | 0.061 | 0.132 | 0.071 | 2.8 | 0.10 | 2.70 | 5378 | 5612 | 4% |

12 | Jrue Holiday |
1 | 0.061 | 0.052 | -0.009 | 2.1 | -0.40 | 2.50 | 3175 | 5486 | 73% |

13 | Monta Ellis |
1.51 | 0.060 | 0.111 | 0.051 | -1.9 | -3.55 | 1.65 | 5418 | 6343 | 17% |

14 | Tony Parker |
1 | 0.060 | 0.124 | 0.064 | 0.8 | 0.75 | 0.05 | 3199 | 4878 | 52% |

15 | Reggie Williams |
2.49 | 0.053 | 0.076 | 0.023 | -0.1 | -0.95 | 0.85 | 1540 | 3171 | 106% |

As you can see, the 2011 The City MIP is **Thaddeus Young** of the 76ers. And hey, for all the Warriors fans reading this, Reggie and Monta made the top 15. That kind of surprised me actually.

I hate the Most Improved Award, because it’s so impossible to define. Your method produced a fine choice in Thaddeus Young, but it’s still possible to nitpick. Here’s Young’s Wins Shares and WS/48 over his career:

Season WS WS/48

2007-08 4.5 0.140

2008-09 5.5 0.103

2009-10 2.1 0.047

2010-11 6.2 0.139

My hunch is that if you found his WARP for 08-09, you’d find something similar to this, that basically it’s not so much that he “improved” but that he came back from a really crappy 09-10 season. This isn’t to say that your methodology is bad, it’s just to say that MIP is a really stupid award.

(ps: In your first graph, you refer to Derek Love. I’m guessing that’s a typo. And an ironic one for obvious reasons.)

Phil, thanks for point out that typo!

Sure enough, when I look at Young’s WARP number from ’08, it’s 0.076. What happened to him last season?

Could be how he’s used. This season, it looks like he was almost exclusive used with at the 4 with just one other big. <a href="http://basketballvalue.com/player.php?year=2009-2010&id=681"Last season, he played a pretty substantial number of minutes at the three, with both Brand and Dalembert (or some other bigs) on with him. Those lineups would probably play slower and take him further away from the basket. Although, according to hoopdata, he didn’t take a lot more long 2’s, so maybe not. Just an off year I guess.

What happened to Conley in eZpm?

What about most regressed? I know it’s not an official award but would still be cool to know

Do have have the correlation coefficient for RAPM for players that played more than, I don’t know, 500 or 1000 possessions in each season?

Conley is interesting. There’s a huge discrepancy between ezpm and RAPM with him. According to 1-yr RAPM, Conley was +3.5 this season, but his ezPM100 was -1.80. What’s your take on him?

In terms of most regressed, I’ll take a look when I get a chance. I’m not sure about your last question. Correlation between RAPM and what?

Conley: why does ezPM hate him so much? He’s above average in WinShares/48

I was talking about RAPM y-t-y

In general I believe RAPM as long as no other metric proves that it does better out of sample predictions

My thought about Conley is that it’s defense, which doesn’t really get captured by Win Shares.

I haven’t seen prediction results for 1-yr RAPM. According to Jeremias Engelmann, the best RAPM predictor is 4 years.

And for y-t-y correlation for RAPM, I’m going to work on it this weekend.

What kind of cutoff (minimum minutes, possessions) is usually used for y-t-y correlations?

Throwing out all players with less than 5000 possessions in either season gives me an r of 0.55 (122 players) for single year RAPM (2010 vs 2011).

When making the cutoff 2000 instead of 5000 r is 0.46 with 204 players.

2 year to 2 year correlation for RAPM is ~0.45 if you compare 08+09 with 10+11.

What everyone should keep in mind though, is that y-t-y correlation is not a great measuring stick for player rating systems. But if you have two systems that are equally good in out of sample prediction you should probably go for the one that has a higher y-t-y corr

Hey, “Jerry” (I assume you’re J.E., right?).

When I do y-t-y correlation, I usually choose a cutoff of 1000 possessions, and then I weight by possessions (using “lm” function in R).

So, you’re getting a correlation with R^2 ~ 0.2. That seems pretty good. The R^2 that I get with 1-yr EZPM is about 0.45. I completely agree that prediction is more important than y-t-y correlation alone.

Just multiplying rating with possessions with a 1000 possession cutoff I get an r of 0.52 for y-t-y, 314 players

What if you take 3-yr RAPM (using 3 prior seasons to this one) and correlate with current season 1-yr RAPM? That would be interesting to know.

Also, now that the season is over, which RAPM metric best retrodicts wins this season, using actual min/poss played 10-11?