APBRmetrics

DSMok1 wrote:Thanks a lot, J.E.!

Could you explain how players with ~0 possessions can have a variety of ratings in a pure RAPM format? Players below 30 possessions vary from -1 to -3. I would have expected them to converge to a single value (the prior) at that small a sample size.

Remember that it's quadratic shrinkage. There's not a whole lot of penalization going on for changing someone's estimate from 0 to 0.5, compared to, say, changing it from 4 to 4.5

J.E. wrote:

DSMok1 wrote:Thanks a lot, J.E.!

Could you explain how players with ~0 possessions can have a variety of ratings in a pure RAPM format? Players below 30 possessions vary from -1 to -3. I would have expected them to converge to a single value (the prior) at that small a sample size.

Remember that it's quadratic shrinkage. There's not a whole lot of penalization going on for changing someone's estimate from 0 to 0.5, compared to, say, changing it from 4 to 4.5

Ah yes. That makes sense. Thanks for the information!

J.E. wrote: As requested by DSMok1 here's RAPM done with all the data I have, adjusted for age when getting the estimates (to not penalize players that played with extremely old/young teammates) but aging again added in afterwards (otherwise the estimates would reflect the case of everyone being the same age).

J.E. according to your link, Kobe has played 143K possessions since 2001, yet B-R has him at 174K possessions. You have Jason Kidd at 129K possessions, but B-R has him at 153K possessions in that same time span. Why is there a huge discrepancy in # of possessions?

http://www.basketball-reference.com/pla ... er_by=poss

colts18 wrote:Why is there a huge discrepancy in # of possessions?

I was just listing regular season possessions

J.E. wrote:

colts18 wrote:Why is there a huge discrepancy in # of possessions?

I was just listing regular season possessions

This is an interesting can of worms.
Did you leave out playoffs because you don't know how much to weigh them?

There isn't much garbage time in playoffs. Teams tend to go with lineups that have 'worked' -- or they're perceived to have worked. Is it possible to get a poorer estimate of a player's value while including playoffs? Or is it only a question of how much better the estimates might be?

Mike G wrote:

J.E. wrote:

colts18 wrote:Why is there a huge discrepancy in # of possessions?

I was just listing regular season possessions

This is an interesting can of worms.
Did you leave out playoffs because you don't know how much to weigh them?

There isn't much garbage time in playoffs. Teams tend to go with lineups that have 'worked' -- or they're perceived to have worked. Is it possible to get a poorer estimate of a player's value while including playoffs? Or is it only a question of how much better the estimates might be?

I didn't say I didn't include playoffs when computing the estimates, I just didn't add each player's playoff possessions to the '# of possessions' that are listed in the file

J.E. wrote:I didn't say I didn't include playoffs when computing the estimates, I just didn't add each player's playoff possessions to the '# of possessions' that are listed in the file

Wait... you included the playoffs? That makes developing any SPM a lot harder (SoS is way harder to figure out). You've already been generous with your work here, but is there any way you could create a similar dataset for regular season only?

Thanks for your efforts!

DSMok1 wrote:

J.E. wrote:I didn't say I didn't include playoffs when computing the estimates, I just didn't add each player's playoff possessions to the '# of possessions' that are listed in the file

Wait... you included the playoffs? That makes developing any SPM a lot harder (SoS is way harder to figure out). You've already been generous with your work here, but is there any way you could create a similar dataset for regular season only?

Thanks for your efforts!

If I were you I'd want them included as the estimates get more precise. You shouldn't use playoff BoxScore data when developing ASPM but this APM data, with playoffs, definitely gives you better estimates of how good the players were in that time period

J.E. wrote:

DSMok1 wrote:

J.E. wrote:I didn't say I didn't include playoffs when computing the estimates, I just didn't add each player's playoff possessions to the '# of possessions' that are listed in the file

Wait... you included the playoffs? That makes developing any SPM a lot harder (SoS is way harder to figure out). You've already been generous with your work here, but is there any way you could create a similar dataset for regular season only?

Thanks for your efforts!

If I were you I'd want them included as the estimates get more precise. You shouldn't use playoff BoxScore data when developing ASPM but this APM data, with playoffs, definitely gives you better estimates of how good the players were in that time period

Yes, but it's not the same sample then.... an interesting quandry. How much more precise are the estimates when including playoff data? Does it make much difference at the 14 year level? I hesitate to use RAPM that doesn't exactly correspond to the stats I'm regressing. You may be right, but philosophically it sounds bad to me.

DSMok1 wrote:How much more precise are the estimates when including playoff data?

Can't really put a number on it at the moment, but more data is always better with these kind of APM-type analyses.

Does it make much difference at the 14 year level?

I'd guess not too much. Maybe in the 0.5-1 range for the extremes - which would be players that significantly over- or underperformed in the playoffs

DSMok1 wrote:... you included the playoffs? That makes developing any SPM a lot harder (SoS is way harder to figure out). ..

Unless I'm mistaken, "strength of schedule" would be irrelevant, since the possession data includes [strength of] opposing players on the floor.

In fact, you might mistakenly think that last year's OKC playoff team was better than they were if you used SOS; but since Westbrook was not playing, just factoring in the actual Thunder players is a more accurate representation of opponent strength, if you were the Grizzlies.

So, J.E., are playoffs given extra weight? Or are they just additional possessions vs quality opponents?

Mike G wrote:

DSMok1 wrote:... you included the playoffs? That makes developing any SPM a lot harder (SoS is way harder to figure out). ..

Unless I'm mistaken, "strength of schedule" would be irrelevant, since the possession data includes [strength of] opposing players on the floor.

In fact, you might mistakenly think that last year's OKC playoff team was better than they were if you used SOS; but since Westbrook was not playing, just factoring in the actual Thunder players is a more accurate representation of opponent strength, if you were the Grizzlies.

So, J.E., are playoffs given extra weight? Or are they just additional possessions vs quality opponents?

When regressing boxscore stats onto RAPM, I need to know the SoS under which those box score stats were compiled. It's trickier to get that info about playoffs.

Mike G wrote:So, J.E., are playoffs given extra weight? Or are they just additional possessions vs quality opponents?

The latter

It's trickier to get that info (SoS) about playoffs

How so?

J.E. wrote:

Mike G wrote:So, J.E., are playoffs given extra weight? Or are they just additional possessions vs quality opponents?

The latter

It's trickier to get that info (SoS) about playoffs

How so?

I can't just download this table: http://www.basketball-reference.com/lea ... tings.html



I'm okay with the 14 year dataset as it currently is. It'll get close enough.

Is this adjusted for margin? Would affect Ginobili's rating in a big way.