APBRmetrics

Has anyone considered using RAPM but adjusting it for luck based on the current stats that we have? For example, some defenders are getting extra credit because their opponents are missing wide open shots. We could adjust that by using NBA.com shot data and applying an adjustment for ever shot that is logged as wide open (nearest defender 6+ feet away) and the shooter (less credit for leaving Curry or Korver open). Another idea would be to run a completely adjusted by shots RAPM. Adjust everything based on the shooters, feet from the basket, openness of shooter, etc. and the players whose teams take the best expected points shots (and the opposite for defense) should be the highest rated players.

In general, when a player performs over or under 'expected' over a significant statistical sample, we tend to think that there's something more substantial than just 'luck' going on. Any of this stuff you speculate about is predicated on having a data set that, AFAIK is not available to the public at large.

Nate wrote:In general, when a player performs over or under 'expected' over a significant statistical sample, we tend to think that there's something more substantial than just 'luck' going on. Any of this stuff you speculate about is predicated on having a data set that, AFAIK is not available to the public at large.

The data is available on NBA.com.

FWIW, on the back of my envelope, I estimate the standard deviation of this scoring randomness effect ranging from 0.13 (per 100 possessions) for a player putting in James Harden minutes (3000) to 0.41 for someone playing one tenth as much.

Assertion of luck seems to assume everyone responds to defense pressure, distance, shot clock time the same. There may be some "luck" but there is probably real variation of players on these things.

colts18 wrote:Has anyone considered using RAPM but adjusting it for luck based on the current stats that we have? For example, some defenders are getting extra credit because their opponents are missing wide open shots. We could adjust that by using NBA.com shot data and applying an adjustment for ever shot that is logged as wide open (nearest defender 6+ feet away) and the shooter (less credit for leaving Curry or Korver open). Another idea would be to run a completely adjusted by shots RAPM. Adjust everything based on the shooters, feet from the basket, openness of shooter, etc. and the players whose teams take the best expected points shots (and the opposite for defense) should be the highest rated players.

Could probably accomplish this with a mixture of KOBE (http://nyloncalculus.com/2015/09/28/int ... t-quality/) and RAPM. Have some ideas, some I've already explored.

There's WAY more than meets the eye going on in terms of luck. As I've said before and as we all know, not all points are created equally.
Making/missing open shots at absurd rates is definitely a starting point but not the easiest.

You might not think of this as luck, but essentially, points that come from less future-predictive events are more "lucky than others" -

For example, the points generated by defensive rebounds (dOR%) are WAY less predictive of future points than the points generated by offensive rebounds (OR%). http://kenpom.com/blog/index.php/weblog ... rebounding

I think Evan's four-factors RAPM back in the day accounted for the varying degrees of luck of the Four Factors (well, 8 factors really). Also, my NCAA ratings do this.

Anywho - Out-of-sample testing for individual stats like this would be huge for RAPM I think.

I've actually wanted to do this for a long time and I'll try to get around to this soon. I can actually tackle this with a counting stat metric instead of just RAPM.

bbstats wrote: Anywho - Out-of-sample testing for individual stats like this would be huge for RAPM I think.

RAPM, or at least the one that Jerry has published, uses cross validation/out-of-sample testing to adjust how much regression to the mean occurs. Is this not enough of an adjustment for luck? Or are you all calling for something different?

Starting now, I'll adjust RAPM/RPM for luck in one area:

In 'Free Throw 1 of 2' situations I'll replace the actual points scored {0, 1} with the FT% of the FT shooter

J.E. wrote:Starting now, I'll adjust RAPM/RPM for luck in one area:

In 'Free Throw 1 of 2' situations I'll replace the actual points scored {0, 1} with the FT% of the FT shooter

Whats the rationale for not including the 2nd FT?

How much of a difference does this make on player RPM? 0.2 points? 0.5? higher?

colts18 wrote:

J.E. wrote:Starting now, I'll adjust RAPM/RPM for luck in one area:

In 'Free Throw 1 of 2' situations I'll replace the actual points scored {0, 1} with the FT% of the FT shooter

Whats the rationale for not including the 2nd FT?

The 2nd FT can potentially be rebounded, influencing what comes next