Appr. 5.x year reg. adj. +/- (J.E., 2010)

[J.E.]

Multiyear uninformed, non coach RAPM from '03/'04 to '11. Probably not too important for most people unless you want to do SPM calculations.

http://stats-for-the-nba.appspot.com/dsmok_edition

Let me know if you need anything in a different format or whatever.

The ratings are shifted so that
sum over (possessions*rating) ~ 0

[schtevie]

However - and it is a big however - it has a glaring weakness: it accomplishes this while clearly - and I mean clearly - biasing the contributions of star players downward (most conspicuously with a zero prior imposed). The notion that in 2002, for example, at the height of their powers, that Tim Duncan was only worth 3.0 points per 100 possessions and Shaq was at 3.6, well, I don't think you would find anyone on this board or elsewhere adhere to that point of view.

I don't think 2002 has all of its data yet? So the player are still being penalized to 0 a lot

bbstats is right. The ratings are low because there are only 3 months of data. With one full season of data the ratings can go as high as those http://stats-for-the-nba.appspot.com/ranking11.

I understood that in having selected the 2002 "dimmed" stars that I was making a point that was clearer than the truth. But the point is (or certainly appears to be) that it takes quite a while for RAPM to fully ramp up to reasonable results when starting with a zero prior (then using each year's results as the subsequent prior). Said another way, it isn't because of the partial season that you get nonsense results on the upper tail. Duncan and Shaq only get to 4.4 and 4.5 the following year. (What underscores the necessity of starting with a really good prior.)

More generally, if you look at the average RAPM of the top 10 players, year by year, you see a monotonic increase (save one year). Starting in 2002 and going up to 2010 , you get: 2.69 (in one third of a season, or whatever), 3.98, 4.77, 5.14, 5.59, 4.94, 5.76, 6.06, and 6.34. Perhaps this is some secular increase in the abilities of the elite, but I am guessing that that is not it.

And then there is the data presented for 2011. Hopefully I am not misrepresenting the results, but on Jeremias' site there are two 2011 estimates. One, a single year version with a zero prior (http://stats-for-the-nba.appspot.com/ranking11). Here the top ten average is 4.93. By comparison, the six year, "chained" prior (http://stats-for-the-nba.appspot.com/ranking_rec) shows a top ten average in 2011 of 7.18 - a difference of 2.25. That is a big difference - the equivalent of two championship level teams added to the talent pool.

Again, this suggests establishing a good prior is really important (and I am especially curious as to the relative predictive power of the two, distinct sets of 2011 estimates.)

Finally, let me return to 2002 and look at another player of recurring interest, one who clocked in at an RAPM of 1.9 at age 25, in expectation at the peak of his career, someone who must be an average player. This is Kevin Garnett. I think we might all agree that it is likely that his actual contributions of that year were greater than this estimate. But so what? A point estimate is a point estimate, one among 400 plus in a given year (or some such). I bring up KG because he is one and the same who leads the field by a country mile in Ilardi and Barzilai's multi-year APM for 2008. There his estimated APM of 14.47 (s.e. 1.41) is way above - in fact, incompatible with - the "chained prior" RAPM estimate of 7.3.

Now, my suspicion is that a more credible prior for KG entering the 2003 season, given the apparent long-lasting effect of such initial specification, is that these two, 2008 estimates would be if not statistically equivalent, at least, closer in line. And that matters. More generally, my hunch is that this would also substantively improve the predictive power of RAPM.

[EvanZ]

Theoretically, couldn't one use 2011 as a prior and go backwards? Could you keep going back and forth like this (start at 2011 -> 2002 -> 2011 -> etc.) and expect some kind of stable solution? Just wondering...

[bbstats]

Theoretically, couldn't one use 2011 as a prior and go backwards?

You just blew my mind.

[J.E.]

Theoretically, couldn't one use 2011 as a prior and go backwards? Could you keep going back and forth like this (start at 2011 -> 2002 -> 2011 -> etc.) and expect some kind of stable solution? Just wondering...

You could do this, but my goal is to get estimates that know nothing about the "future"

[DSMok1]

That would be interesting, though. It would show how much of an effect aging is having using your current methodology.

[DSMok1]

Multiyear uninformed, non coach RAPM from '03/'04 to '11. Probably not too important for most people unless you want to do SPM calculations.

http://stats-for-the-nba.appspot.com/dsmok_edition

Let me know if you need anything in a different format or whatever.

The ratings are shifted so that
sum over (possessions*rating) ~ 0

Wonderful. Thanks, J.E.!

So basically, if a player played a lot of minutes, he was on average a +1.3 or whatever, so to get the league to sum to 0, you had to shift downward? Interesting. What prior would you have to use to get the league to sum to 0 without shifting it afterwards? -3? That would be very interesting to know, and perhaps that would be a better set of numbers to use. How hard would it be to create a similar set of numbers like that, and also (perhaps) a set of plain APM without using a prior at all, for comparison?

[J.E.]

So basically, if a player played a lot of minutes, he was on average a +1.3 or whatever, so to get the league to sum to 0, you had to shift downward? Interesting. What prior would you have to use to get the league to sum to 0 without shifting it afterwards? -3?

Not sure but I can check. Might just be -1.3 (divided in half for offense and defense)

How hard would it be to create a similar set of numbers like that, and also (perhaps) a set of plain APM without using a prior at all, for comparison?

Are there two questions in that sentence or just one? I can create plain APM for that timeframe, too. Do you want me to? That would, again, take some time. And you'd have to tell me what I should use as a possession cutoff

You can now click on player names on this site http://stats-for-the-nba.appspot.com/ranking_rec and see a chart of their RAPM value over all years they have been in the league.
You can't click on players with just one year in the league, and you cannot click on player names on any other subpage as of yet. Should be coming in the not so distant future though.

My favorite chart is Amir Johnson's

[DSMok1]

So basically, if a player played a lot of minutes, he was on average a +1.3 or whatever, so to get the league to sum to 0, you had to shift downward? Interesting. What prior would you have to use to get the league to sum to 0 without shifting it afterwards? -3?

Not sure but I can check. Might just be -1.3 (divided in half for offense and defense)

How hard would it be to create a similar set of numbers like that, and also (perhaps) a set of plain APM without using a prior at all, for comparison?

Are there two questions in that sentence or just one? I can create plain APM for that timeframe, too. Do you want me to? That would, again, take some time. And you'd have to tell me what I should use as a possession cutoff

You can now click on player names on this site http://stats-for-the-nba.appspot.com/ranking_rec and see a chart of their RAPM value over all years they have been in the league.
You can't click on players with just one year in the league, and you cannot click on player names on any other subpage as of yet. Should be coming in the not so distant future though.

My favorite chart is Amir Johnson's

Brilliant Charts! (How'd you do it?)

When I do my ASPM, I think I'm going to use a cutoff somewhere in the 3000-5000 poss range (according to the poss counts you've got in the above listing).

Those were actually 2 separate requests--one to figure out what to regress towards to get it all to equal 0, and one to create a raw APM. Neither are critical though; I'm going to be really busy for the foreseeable future.

[EvanZ]

Steve Nash's offensive RAPM has increased tremendously over the past 8 years. It would be interesting to look at aging curves for each position. Do PGs tend to peak later?

[DSMok1]

I've got my spreadsheets set up to work on ASPM with this new 8 year data set.

Initial results indicate I was weighting rebounding too heavily, and defensive stats and efficiency not enough. I also had too big a spread, overall. R^2 (not properly calculated because of weighting) is roughly 0.6.

Players helped most with new weights (same construction, currently): Nash, Korver, Tony Allen, Brent Barry. Players hurt the most with the new weights: Kevin Love, Marcus Camby, Okafor, Jason Kidd.

2011 top 10:
Lebron
CP3
Russell Westbrook
Rose
Wade
Kobe
Howard
Dirk
KG
Manu

[Crow]

Among the 1 year RAPMs presented separately for 2008 to 2011, only 12 guys were estimated +2 or better in all 4 seasons. Miami has 3, Boston and San Antonio 2, no other team more than 1. Deng and Artest made this group and might not have been expected by all to do so.

Not too many more than a dozen other guys who were +2 in 3 of 4 seasons. Among the possible surprises- Fisher, Foster, Amir Johnson, JR Smith, Cardinal, B Davis, Josh Howard and Turkoglu.

There are players often treated as stars (and voted to All-Star games) who don't have RAPM estimates above +2 in at least 3 of 4 seasons. Some of them might have just missed making the standard on at least one occasion. Some are young stars just starting their RAPM prime, some have left that prime (for good or for some specific immediate reason that might be reversible). But some other "stars" just don't perform that strong on RAPM, don't consistently pull an estimate of much positive impact. That seems worth considering.

There are somewhere around 35-50 players whose RAPM performance is somewhat short of the top 30 or so found to be consistently strong by this approach. Multi-season estimates (and other considerations) might satisfy enough to decide to go ahead and treat some of these players as tier 1 or 2 stars more than others or for some reviewers more than others. But, if you give much weight to RAPM, most will probably end up staying considered as tier 3 or 4 players.

With 10 years of data it would be possible to check such things as the frequency of 3, 4 or more seasons above +2 for the whole period and calculate the average duration of such a performance level, the likelihood of repeat if attained or repeat if attained then lost and by age, position, height, player type or a specific stat or another metric or some other marker.

[huevonkiller]

Multiyear uninformed, non coach RAPM from '03/'04 to '11. Probably not too important for most people unless you want to do SPM calculations.

http://stats-for-the-nba.appspot.com/dsmok_edition

Let me know if you need anything in a different format or whatever.

The ratings are shifted so that
sum over (possessions*rating) ~ 0

I like that list (at the top at least). Also time for me to give credit for Kobe's 88628, second best on that list.

Glad to see he's not always doing so bad on this site.

[bbstats]

Kobe's 88628, second best on that list.

The last column represents possessions played...

[huevonkiller]

Kobe's 88628, second best on that list.

The last column represents possessions played...

Why isn't that an accomplishment? APBRmetrics gets too wrapped up in "per game", "per possession", "per minute".

The ability to stay on the court should also be recognized. This is a tougher, athletic, defensive era. Even more so than past eras.

Sorry but I don't recognize Dwyane Wade, Chris Paul, Manu Ginobili as superior RAPM players, even if they are slightly more efficient per possession. Kobe's durability should also be taken into account.

Last note, Steve Nash seems too high. Stunning to think he is "better" than Kobe, Chris Paul, and Wade, but I doubt it.