Prior-Informed RAPM (Updated 2-3 Times/Wk)

[talkingpractice]

A lot of people have been asking us on Basketball Twitter and APBR for a full list of RAPM values for the NBA, so we decided to oblige with the version that most people requested. The below numbers are created as follows:

- This is RAPM-informed RAPM.

- Seed year is 2006. All non-rookies start at {0,0} in the seed year.

- Mean reversion of 15% is used between seasons.

- An aging curve is applied to each player between seasons. It is currently specified as (27-AgeYrs)*.06 added as a linear term between seasons.

- ROOKS take a prior of {-1.2, -1.2}.

- XROOKS (rookies with substantial prior international experience) take a prior of {-0.6,-0.6}.

- Priors for Rose, Bynum, and Frye were set at {0,0}.

- All data parsed internally via NBA.com, crosschecked with 2 other sources, final verifications done via video review as needed.

- We do not (and you should not) use these numbers for wagering purposes. We wouldn't be giving them out to you if we did.

- Changing any/all of the above parameters is reasonably trivial, so if the community desires any changes to the above, we can make some changes before the next run. We will likely update these about 2-3 times/week.

It's fastest to get us on Basketball Twitter now, probably. So if anyone has any ideas for slightly different parameters, you can also let us know there, @talkingpractice.

[boooeee]

Do the numbers have a particular unit definition? For example, points per 48 minutes or 100 possessions?

[deepak]

Is it just me, or is Amir Johnson consistently near the top of the player's ranked by APM stats?

[talkingpractice]

Do the numbers have a particular unit definition? For example, points per 48 minutes or 100 possessions?

They're per 100 possessions.

[Crow]

Thanks very much for sharing these.

I look forward to looking at the list.

Initial observations:

Nick Collison near the very top. Last season he was only at about +1.5 on Jerry's RAPM. He was near the very top on some other form of APM. Maybe the pure APM? Never above +2.5 on what Jerry's site currently shows.

Granger's near top 10 ranking is encouraging and perhaps forgives his slow re-start shooting.

Westbrook at +4 but how much was that influenced by prior RAPM? Wil want to see where his shooting / scoring efficiency goes on his return and whether this rating goes higher or down with more 2013-14 data.

Channing Frye at +3 is a surprise. Previous best was a bit over +1, though the trend is basically up.

Rubio stil good on defense but near neutral on offensive impact.

[Crow]

So many stories of players and team evaluations (earlier and going forward):

Haslem is a bottom 10 in the league player on RAPM impact estimate now?

Aaron Brooks is bottom 5%, has had other bad estimates and got re-signed by Houston?

Jarrett Jack is also bottom 5% and got brought in by another “advanced analytics” team to provide veteran leadership… or go along their other big mistakes.

Ben McLemore young guy with star upside or just another over-hyped guy with a –3 rating or both?

Oladipo stuffing the boxscore and pulling a –3 impact estimate.

Valanciunas, coveted cornerstone but –2 and a long way to meaningful positive impact.

Brook Lopez sorely missed or we didn’t really need your –2?

What to think of next superstar Kyrie Irving and his –1.4? John Wall ok but short of +1 and short of his hype. Kemba Walker? Up thru the years to barely positive now.

Hard to ignore some of the good stats in Belinelli’s boxscore but his impact is near –1. Not that bad, but not really shooting brilliant match made in heaven either.

Big huge contract for Monta Ellis. RAPM slips to near –1. Yawn.

GMs drooling after Bradley Beal. +0.7

Pekovic, huge deal, cave to his final demands. Slightly negative impact estimate.

Kobe’s offensive help almost totally offset by his weak defense.

DeAndre Jordan’s offense mostly offset by a modestly weak defense. A pillar of their present and future?

Deron Williams and Paul Pierce +1s. Joe Johnson about +2, Garnett +3. (still? Is the prior influencing things too much? Just a question.)

Ibaka a shoe-in All-Star says Jalen? +1.2. Lee about +1.7. Cousins +0.2. Duncan +3.3. Dirk +5.
+2 or better players not on rookie contract and not making $8+ million. 2. N. Collison and Channing Frye. +2s on rookie contract (without a huge extension already) ready to help lead the league into a quality future? Umm, Rubio. That’s it.
Bring in Dwight Howard and end up with him replacing Asik? Their offensive and defensive estimates are nearly identical. Great move? Not yet, not really.

Pretty big minute Brandon Bass near – 4. Jeff Green? –1, eh.

Lots of underwhelming / disappointing former Tar Heels out there. I used to give those guys extra credit. Now I start out suspicious and have to be won over.

Steven Adams near –2. For a modest minute rookie I guess that is one step up the ladder.

Tyreke Evans –2. Not so smart move Pelicans.

Bynum near –1. Sounds about right. Could do worse.

Austin Rivers making good progress from year 1 up to –0.65.

Zach Randolph at near zero impact. Will Memphis listen more to the boxscore metrics or the RAPM assessment?

“Star or superstar on the verge” Kawhi Leonard? A near zero net impact estimate. Down from near +1.5 last season.

[coredump]

Thank you for posting these.

Two young 2nd-year players caught my eye in your numbers. Both are on teams clawing to make the playoffs this year. Maybe they get dealt away with future considerations for impact players.

1. Jae Crowder (+1.15) -- Stuck behind Marion (-2.03)
Based on the 5-man unit data on 82games.com, he seems to play very well alongside Nowitzki. Shot charts indicate he's spending too much time outside the 3pt line, and not near the basket where he's very effective.

2. Evan Fournier (+0.59) -- Stuck behind Foye (-1.17)
It looks like he almost exclusively gets minutes with (Robinson, Arthur, Mozgov) so probably difficult to isolate his impact. He looks like a solid 3pt shooter/spacer though, based on his shot charts.

[Crow]

I think the worst RAPM estimate for a named All=Star are for DeRozan at -1.2 and Irving -1.4. One could also object to Wall (+0.9) over Lowry (+2.3). The coach sure know who the 20+ point per game scorers are and apparently give that high weight. I wonder who many know any RAPM type estimate or give it any weight for this voting or anything actually important.

Wonder if a guy like Amir Johnson (+5) got any votes. Probably few if any and probably not based on RAPM.

[Mike G]

Joe Johnson?
If he were a perennial allstar-caliber player who was finally getting his due, then almost, maybe -- but not really.
Since he's been picked a number of times already, while never having a truly stellar season, it's almost nauseating.

[Crow]

Anthony Davis is only neutral on latest RAPM from talking practice that I checked. Not bad for a second year player but still worth noting and watching.

Noah is widely respected for his effort and impact but his current RAPM estimate is -0.4.

[talkingpractice]

Quick comment on some of the things mentioned here.

These prior-informed (by previous year) RAPMs are biased (ofc) to that selection of prior. The priors here are set very methodically (for purposes of this list we publish).

So for most of the 2014 rookies to be near -2 is not surprising, as most have < 1000 minutes played and are being regressed to the rookie prior. So this explains McLemore, Oladipo, Adams, etc. Players who obviously got better last season, like Wall or Brook Lopez (particularly with such low mp), can get somewhat "stuck" a bit in info from prior years, and are still being dragged down by prev year info. So the RAPM tables we are publishing are a function of the particular model used (the 'standard' RAPM-informed RAPM model that everyone asked for), and we need to take that into account when examining the results.

This is most affecting Anthony Davis. If we set his prior logically, based on what we think his actual prior is (and not based on last year's RAPM), then I'd have it at +15 and not -1.5 or whatever it is in this method.

[J.E.]

Excuse me for piggybacking

Vanilla RAPM numbers can now be accessed here http://stats-for-the-nba.appspot.com/va ... -2014.html

I'll probably update weekly - these numbers don't really change much from day to day

This is not using priors, but instead I'm doing one large regression while giving lower weight to observations from older seasons (the way Joe Sill did it)

[talkingpractice]

Excuse me for piggybacking

You're far from a piggyback. Happy to have you in any thread of ours.

I'll probably update weekly - these numbers don't really change much from day to day.

Ftr, we're only going to update once/week or so too now. Doing it 3 times/week isn't really needed, for the same reason that je just gave.

[Crow]

Thanks JE.

Look forward to comparing the two data sets.

[2016Champions]

A heavier use of priors definitely has its pros and cons. I think talkingpractice did a great job mentioning the cons. For me the two biggest pros would be:

Larger sample size
Passes the smell test better

What I mean by smell test is the impression people get at first glance. In order for RAPM to really take flight and become more recognized as a useful stat among general fans, the list needs to look agreeable. If there are numerous non-stars listed higher than the majority of All-Stars on the list then people will come take one sniff and sneeze all over it. So personally, I'm all for having a variation of RAPM with a heavy use of priors and box score stats, which I'm guessing is the case with these: http://stats-for-the-nba.appspot.com/ratings/2013.html

That's a much easier list to show someone completely new to RAPM as opposed to the vanilla version where Patrick Beverley is a boss. Just a thought.