14 Year Average RAPM Dataset

[Crow]

I may not have looked at this closely before. It may be the least in the pocket of conventional wisdom rapm run I've seen.

Amir Johnson at 9 probably unsettles others a lot more than me but Deng 18, Hiliaro 27, Rubio 28, battier 31, e Jones 36, Danny green 46, Tony Allen 50, josh Howard 55, Christie 57, mbah a moute 71, fisher 78, udoh 87, o jeffers92, Tolliver 94, Scott Padgett 109, fesenko 116, Iverson 140, ayon 139, Durant 159, etc.?

What values would a simple minutes weighted summation and avg. of the one yr results show? I'd guess they'd be more in the pocket for a lot of these guys.

[colts18]

What values would a simple minutes weighted summation and avg. of the one yr results show? I'd guess they'd be more in the pocket for a lot of these guys.

I already did something like that before.

Possession weighted RAPM of 1997-2014 RAPM.

https://sites.google.com/site/rapmstats/97-14-rapm-2

[Crow]

Thanks, I 'll review.

A number of the questions I get about old apm runs could be answered by your site. I've bookmarked, lost track of and now bookmarked again.

[Crow]

This approach produces estimate rankings very different from Jerry's single run in some cases and not as much as I thought it would for others.

Iverson is still 180. Durant 87.

[gzchen]

I'd asked Jerry about this on Twitter a month ago, and hopefully I didn't misunderstand him:

The 14y set is RAPM while the single year numbers are xRAPM. The numbers that stood out to me between the two datasets were that the 14y numbers love Rasheed more while the (possession-weighted) single-year numbers love Big Ben more, which make sense given how ridiculous some of Ben's box score stats were.

[Crow]

My perception / concern that a number of these values are out of the pocket of conventional wisdom may have been a bit too strong. A lot of these cases near the top are affected by strong defensive estimates that i don't immediately quarrel with. I am still not totally on board with this being the best resource, except if you are judging a full career. One or two year RAPM or XRAPM may have more noise but it is also more sensitive to real player changes not in step with a general aging curve.

(Is the aging curve for all positions and is that pretty consistent with how the positions behave separately?)

[permaximum]

Thanks again for J.E's work.

BTW, NBA PBP data goes back to 1996-97. I would be a lot more interested in 1996-2014 RAPM (playoffs included) with age adjustment AND without age adjustment. If it's too much work, no problem. However I'm sure, certain people here have 1996-2001 PBP data.

[fpliii]

Thanks again for J.E's work.

BTW, NBA PBP data goes back to 1996-97. I would be a lot more interested in 1996-2014 RAPM (playoffs included) with age adjustment AND without age adjustment. If it's too much work, no problem. However I'm sure, certain people here have 1996-2001 PBP data.

Yes, thanks again. Best dataset we have IMO.

Personally, I'd like to see 2001 and 2002 single season RAPM (with priors) from the stats.nba.com data first.

[Crow]

I still have some unease with it. How much does variation in the game over time or especially in the defense post the new hand-check rules affect the ratings for guys who mainly played early, middle or late in the 14 yr timeperiod? Is 14 years just too long and causing weird fitting?

Durant only +1.3 for his career? That is top 12% but still, +1.3 for a guy who turned into the MVP?

Only 13% of 1383 players were +1 or better? Less than 20% better than neutral? 8% worse than -5 and all the way down to -9, multiple times worse than replacement level?

The timeperiod has a lot of very good to great players at end of career.That threw me for a few moments.

How different would this look if one discarded first 2 seasons and last 2 seasons? What other tweeks were considered or should be considered?

75% of the top 40 are playing now. Seems a bit high.

Is a three point shot worth the same thing in 1991,2004 and now? I don't think so. In part because of the research nylon calculus shared recently on winning thresholds.

[Crow]

64 plus 3 or better, about 300 -3 or worse. 106 plus2 or better, 580 -2 or worse. 270 at or above neutral, over a 1000 below. 580ish below -2. 300 below -3. This distribution is endorsed? I continue to have some unease. Will have to get around to that comparison with weighted sum of 1 yr estimates.

[Statman]

64 plus 3 or better, about 300 -3 or worse. 106 plus2 or better, 580 -2 or worse. 270 at or above neutral, over a 1000 below. 580ish below -2. 300 below -3. This distribution is endorsed? I continue to have some unease. Will have to get around to that comparison with weighted sum of 1 yr estimates.

I don't normally chime in on RAPM because of my unease with the results, and I don't want to be THAT guy. But, Crow recently has pointing out much of what has been bothering me for a while with RAPM.

[Mike G]

If you are looking at career values, of course there are several below-avg players for every above-avg one. This isn't representative of a given season, because a better player goes more seasons.

Look at any random group: 100 tallest guards since 2001: http://bkref.com/tiny/rXMMW
Just 14 of 100 are/were above-avg in PER.
18 are above .100 WS/48, but only 10 of those for >100 career minutes.
16 were >0 BPM

Median career minutes in that random assortment are ~2800
If you have a higher cutoff for total minutes, you'll have a higher caliber of player.

[Crow]

I am more comfortable with top talent being scarce than I am with the bottom being so heavy but you are right to recommend comparing the distributions on other metrics. I've known that median performance is below mean but seeing 1000 of 1280 below neutral was more than I expected. Not a bell curve.

[Mike G]

In BPM for last season, the bell shaped curve is in seen in the minutes.
The top tier is just LeBron, Durant, and 10 minutes from 2 others.

Code: Select all

BPM       minutes   #
> +8        6034     4
+6 to 8    12046     6
+4 to 6    22862    12
+2 to 4    81139    40
0 to 2    178576   113

-2 to 0   166001   133
-4 to -2   94848   112
-6 to -4   24204    67
-8 to -6    7682    35
< -8        1801    35

With 15-man rosters, a large number of sub-replacement types will get a total of 1000+ minutes per team.
The size of the roster, however, doesn't improve the quality of players 7 thru 15. There are just more of them making appearances.

[Crow]

Thanks. That is helpful.