14 Year Average RAPM Dataset

[J.E.]

Is this adjusted for margin?

What do you mean?

[nbo2]

What do you mean?

Is there a scoring margin variable in the model?

i.e. if Team A is up 10 and scores at a certain rate per possession, the player ratings would be different than if Team A is down 10 and scores at the same rate per possession

[J.E.]

Is there a scoring margin variable in the model?

i.e. if Team A is up 10 and scores at a certain rate per possession, the player ratings would be different than if Team A is down 10 and scores at the same rate per possession

Yes, that's in there

[DSMok1]

J.E., could you add your "error of estimation" to this dataset when you have a chance? I'm interested in just how stable this is long-sample, and for use in estimating the error terms for ASPM.

[Crow]

I would think that a 14 yr playoff only rapm would have several good uses.

[colts18]

Is there an update on this with the whole 2014 season completed? I'm curious if Lebron is #1 now with the extra few months of data.

What's interesting to me about this list is the variance in rankings. From J.E.'s past RAPM data, the highest single season value was LeBron's +9.8 in 2010 but in this ranking KG is at +9.7 (leBron +9.5) for this whole 14 year span.

[DSMok1]

Is there an update on this with the whole 2014 season completed? I'm curious if Lebron is #1 now with the extra few months of data.

What's interesting to me about this list is the variance in rankings. From J.E.'s past RAPM data, the highest single season value was LeBron's +9.8 in 2010 but in this ranking KG is at +9.7 (leBron +9.5) for this whole 14 year span.

Larger sample size allows for larger outliers, I guess.

[colts18]

Is there an update on this with the whole 2014 season completed? I'm curious if Lebron is #1 now with the extra few months of data.

What's interesting to me about this list is the variance in rankings. From J.E.'s past RAPM data, the highest single season value was LeBron's +9.8 in 2010 but in this ranking KG is at +9.7 (leBron +9.5) for this whole 14 year span.

Larger sample size allows for larger outliers, I guess.

It should be the opposite. The larger the sample, the less variance and outliers there are. That's the whole point of RAPM. Small samples have wide spreads. Large samples don't. Based on that we should expect a small spread of talent from 01-14 RAPM.

[DSMok1]

Is there an update on this with the whole 2014 season completed? I'm curious if Lebron is #1 now with the extra few months of data.

What's interesting to me about this list is the variance in rankings. From J.E.'s past RAPM data, the highest single season value was LeBron's +9.8 in 2010 but in this ranking KG is at +9.7 (leBron +9.5) for this whole 14 year span.

Larger sample size allows for larger outliers, I guess.

It should be the opposite. The larger the sample, the less variance and outliers there are. That's the whole point of RAPM. Small samples have wide spreads. Large samples don't. Based on that we should expect a small spread of talent from 01-14 RAPM.

I don't think so. RAPM is regressing quite strongly to the mean; if you have 0 possessions you will be regressed all the way to the mean. As you add possessions, you can gradually push further from the mean. If you've got 14 years of data, then you are regressed very little toward the mean.

[huevonkiller]

Is there an update on this with the whole 2014 season completed? I'm curious if Lebron is #1 now with the extra few months of data.

What's interesting to me about this list is the variance in rankings. From J.E.'s past RAPM data, the highest single season value was LeBron's +9.8 in 2010 but in this ranking KG is at +9.7 (leBron +9.5) for this whole 14 year span.

Larger sample size allows for larger outliers, I guess.

It should be the opposite. The larger the sample, the less variance and outliers there are. That's the whole point of RAPM. Small samples have wide spreads. Large samples don't. Based on that we should expect a small spread of talent from 01-14 RAPM.

Well a few things: First, in 2011 Miami robbed LeBron of one of his prime years, he went from >+11 for 2009 and 2010 (according to J.E.), to +8. It is safe to say that in Cleveland he was surrounded with an inferior roster, but also did not have to play with a ball-dominant guard who clogged up the paint. LeBron likes playing with a Mike Miller type more than a Dwyane Wade type (see 2013 Finals).

With just that alteration their ratings (LeBron and KG) would be nearly identical. If we exclude LeBron's rookie season obviously he would look even more dominant. If you rank LeBron's top 4 or 5 seasons they're still the most dominant as well.

Additionally, KD also gets penalized for the relatively poor start to his career, but he's still not liked by J.E.'s ratings in general. His prime seasons are not particularly noteworthy on JE's website, especially compared to someone like Howard or Wade. Certainly an interesting issue.

[fpliii]

Great stuff, thanks for sharing J.E., this has to be the most extensive such dataset out there.

Just wondering, would you consider uploading single-season RAPM (without box score elements) for each season from 01-07, updated with your current research? acrossthecourt (97-00) and talkingpractice (08-14) have published the other seasons, but there's a seven-year gap between them.

[kmedved]

Great stuff, thanks for sharing J.E., this has to be the most extensive such dataset out there.

Just wondering, would you consider uploading single-season RAPM (without box score elements) for each season from 01-07, updated with your current research? acrossthecourt (97-00) and talkingpractice (08-14) have published the other seasons, but there's a seven-year gap between them.

James Brocato has published 2002-2014 single-season RAPM here: http://shutupandjam.net/nba-ncaa-stats/pi-rapm/ It has like a .99 R^2 with Talkingpractice's numbers (which makes me wonder where the 0.01 has gone, but no matter)...

[fpliii]

Great stuff, thanks for sharing J.E., this has to be the most extensive such dataset out there.

Just wondering, would you consider uploading single-season RAPM (without box score elements) for each season from 01-07, updated with your current research? acrossthecourt (97-00) and talkingpractice (08-14) have published the other seasons, but there's a seven-year gap between them.

James Brocato has published 2002-2014 single-season RAPM here: http://shutupandjam.net/nba-ncaa-stats/pi-rapm/ It has like a .99 R^2 with Talkingpractice's numbers (which makes me wonder where the 0.01 has gone, but no matter)...

Thanks.

I believe these are the sheets that were once hosted on J.E.'s site (at least the numbers looked the same)? Nice to have everything in one place, of course.

From my understanding though (not sure how correct/incorrect it is), there was originally something like 15% or so of missing games in the first few seasons (again, apologies if I'm wrong here). If the new data set J.E. linked in the OP has reparsed the PbP, so the first few seasons have less missing data and are more complete, it might improve the single season calculations (particularly for 01 and 02, since I believe J.E. had said in an earlier thread that flattening/stabilization occurs at around 5000 minutes, which most significant contributors wouldn't hit until their third seasons, so filling in missing data would seem to improve accuracy before they reach that threshold).

Again, could be off base, just my understanding.

[permaximum]

So, Kevin Garnett is better than Lebron thus making him the greatest ever

[Crow]

Forgot about this run. Will need to look at it again.