Top values on new metric called "Individual Player Values"

[sideshowbob]

Just saw that the site's been updated with Offensive/Defensive splits. Just eyeballing the overall figures it tells me the spread looks a bit tighter as well.

[talkingpractice]

Updated as of today. Some comments ->

1, Deepak (sorry for delay in responding to your question from a few weeks ago), offense/defense split provided.
2, We've noted that boxscore stats don't provide much for defense, but the reality is that they do indeed show *something*, so it was silly to not use the information. As such, we're now using the information (a little bit).
3, We fixed any remaining bugs with some players (there were maybe a dozen), the values should be really clean now.
4, Moved them to 1 decimal as using 2 decimals was always silly anyway.
5, Don't want to answer v-zero's question (mean v median v mode) for obvious reasons though I will say that we originally preferred one way but now think a different way is preferable.
6, Made an amendment to how we do priors for rookies, now using a moderately better prior for what we call X-ROOKs ("rookies" with relevant international professional experience).
7, They are indeed a bit tighter now. tbh, it's because we turned the mean reversion back onto full blast (we had toned it down some)

[deepak]

Thanks for the update.

[Crow]

Comparing early January IPV to the current ratings, the order of the top 7 is virtually changed (2 changes of 1 and 2 spots respectively). Of the current top 30, the players with the biggest rating improvement since January’s measure (all with at least a 20 spot gain) are Tony Allen (up 42 spots), Vince Carter, Paul George and Nick Collison. Aldridge was next best at improvement at +12 then Griffin at +9. Nobody else more than +5 (Amir Johnson at +5). The biggest estimated rating slips among those still in the top 30 (and at least –5) are T Chandler (-12), Bryant, Ginobili and Harden. Duncan and Iggy are –4. Melo actually lost 18 spots (and out of the top 30) down to 37th, the biggest fall of anyone in the top 30 in January.

early
Jan. mid-
rank April Net Change (+ = improvement)

Player
1 1 0 LeBron
2 2 0 Chris Paul
3 3 0 Kevin Durant
5 4 1 Tony Parker
5 5 0 Dwyane Wade
6 6 0 Russell Westbrook
9 7 2 Kevin Garnett
4 8 -4 Tim Duncan
18 9 9 Blake Griffin
40 10 30 Paul George
8 11 -3 Paul Millsap
7 12 -5 James Harden
17 13 4 Dwight Howard
26 14 12 LaMarcus Aldridge
16 15 1 Marc Gasol
15 16 -1 Dirk Nowitzki
14 17 -3 Stephen Curry
23 18 5 Amir Johnson
13 19 -6 Kobe Bryant
12 20 -8 Manu Ginobili
10 21 -11 Tyson Chandler
25 22 3 Mike Conley
22 23 -1 Nene Hilario
20 24 -4 Andre Iguodala
67 25 42 Tony Allen
57 26 31 George Hill
52 27 25 Vince Carter
29 28 1 Josh Smith
49 29 20 Nick Collison
28 30 -2 Danilo Gallinari

The significance of these changes during the season (if any to you) is up for consideration / cross-check with other resources.

[talkingpractice]

Updated with final numbers (end of regular season).

[v-zero]

Would you be willing to reveal how close you thought the Nuggets vs Warriors series would be according to IPV?

[talkingpractice]

Would you be willing to reveal how close you thought the Nuggets vs Warriors series would be according to IPV?

I'd talk to you about it via pm after the series is over, with certain assurances.

One caveat is that we dont calculate things like that using IPV per se (ie "according to IPV" doesn't mean anything in that context really). The IPV's of the players are relevant, but other things are as relevant as (or more than) the raw player evaluation stuff (lineups, matchups, home/road, etc).

[v-zero]

I'd be interested to hear about that. I only asked about IPV as I know your efforts extend to matchups etc, which I had assumed you would rather not share either prior to or after the series, but if you are willing then I'd be very interested to hear. What assurances would be required?

[colts18]

Is there a reason why leBron is the only player on the list who doesn't have his last name listed too in the ranking?

[talkingpractice]

Is there a reason why leBron is the only player on the list who doesn't have his last name listed too in the ranking?

We thought that calling him King James woulda been going too far.

What assurances would be required?

I'll shoot you a note about it now.

[rshea]

Wayne Ellington with an impressive set of 0's across the board. He won't help your team, but then again he won't hurt your team either... which actually kind of helps.

[Crow]

A few observations about the IPV database:

There are 50% more +3 or better offensive IPV marks than defensive IPV. Almost the same levels of +2 or betters. At +1 there is a modest edge to offense.

For -2 or worse there are 5.5 times as many on offensive IPV as defense. At -3 or worse it is a 10 count on offense, none on defense. At -1 or worse it is 50% more common on offense.

8 of the top 9 on IPV overall are on the 4 best teams (MIA, OKC, SAS, LAC). New York's best is 20th and is Chandler.

For players better than +1 on offense or defense, they average a +2 overall rating. Likewise with those worse than -1 on one side of the ball.

[Crow]

JE’s 2012-3 RAPM and IPV have all the same guys in their top 12s and 28 of RAPM’s top 30 are also in IPV's top 30. After that more variation creeps in so that IPV has 40 of RAPM’s top 50.

The correlation of offensive RAPM to offensive IPV for the 60 players on either top 50 was .888. The correlation on defense was weaker at .586. The overall player value correlation was .639. I am not so surprised with the existence of some bigger variation on defense given past study, but I wonder why it is there and why it is of moderate size and what to make of it.

The average difference on offense for these players was very very very close to zero. Well less than .001. On defense it was almost a full point different, with RAPM judging these players better on defense by that amount on average.

I guess this RAPM used priors. I don’t recall if IPV does but I think it does. Correct me if I am wrong.

The players that the 2 metric's estimates are most different on are Chris Andersen and Brooks Lopez at over 5 pts and almost 4 pts respectively. Only 5 players in this dataset vary by more than 3 pts. They are all big men, coincidence or not.

About 75% of overall player values are within 2 pts per 100 game possessions. Almost 90% are within 2.5 pts. These results could be called pretty good or not good enough to inspire confidence and use or much of them. The average of the two or an average with even more metrics might reduce average error or alleviate some concern.

Of course this data is for only part of the overall league, the very top. I didn’t have the time to line up / match up everything and add data right now, so these results are suggestive and not the definitive word on similarity between the metric findings. If someone else wanted to match-up all the names and re-run some of this analysis, great.

[NBAMetrics]

I believe IPV does use priors and an aging curve with their numbers, but only +/- for the defensive side of things.

Briefly looking over the two different sets of data, it seems RAPM values interior defenders more than IPV, while IPV rates big men more favorably on offense.

Does either metric seem particularly more accurate than the other to anyone?

[Crow]

Thanks for the comments.

I'll have to look at RAPM / IPM further before I decide which looks more right to me, if either. When in doubt, I tend to blend.