Anybody working on any exciting projects for the fall?
-
andylarsen
- Posts: 11
- Joined: Fri Jan 11, 2013 8:39 pm
Re: Anybody working on any exciting projects for the fall?
Thanks! And interesting!
What do others' ocular regressions suggest that the difference is between these two plots? It's not just that Kobe Bryant is way overpaid.
What do others' ocular regressions suggest that the difference is between these two plots? It's not just that Kobe Bryant is way overpaid.
Re: Anybody working on any exciting projects for the fall?
Andy, I now see that the dataset isn't pre-set up for team level analysis. Oh well. I might look at alternate performance / salary matchups or splits by salary level at a later time.
Re: Anybody working on any exciting projects for the fall?
For 2014-5, the correlation between overall RAPM performance estimates and salary WITHIN the subgroup paid over $10 million was a ridiculously weak .09. Even worse within the $5-10 million group. Better within the under $5 million but barely .2.
GMs have some ability to sort guys into these 3 groups to raise their overall performance-salary correlation to the still imo pretty weak number Andy reported but they had almost no success in ordering pay for performance within these broad, should be obvious? pay bands. The salary rules and the surge of new money make it harder; however, this doesn't look like particularly advanced, demonstrably expert matchup work.
Of course a lot depends on whether you give credence to the RAPM evaluations. A correlation with a box score metric would either put more heat on the GMs (if it generally agreed more with the RAPM correlation) or perhaps would put more scrutiny on one or both of the metrics, if one stood out far from the others or on all until otherwise determined if none were that near either of the others.
GMs have some ability to sort guys into these 3 groups to raise their overall performance-salary correlation to the still imo pretty weak number Andy reported but they had almost no success in ordering pay for performance within these broad, should be obvious? pay bands. The salary rules and the surge of new money make it harder; however, this doesn't look like particularly advanced, demonstrably expert matchup work.
Of course a lot depends on whether you give credence to the RAPM evaluations. A correlation with a box score metric would either put more heat on the GMs (if it generally agreed more with the RAPM correlation) or perhaps would put more scrutiny on one or both of the metrics, if one stood out far from the others or on all until otherwise determined if none were that near either of the others.
-
Kevin Pelton
- Site Admin
- Posts: 104
- Joined: Thu Apr 14, 2011 10:05 pm
Re: Anybody working on any exciting projects for the fall?
As others have noted, there are multiple reasons why a player's salary might not reflect his perceived value around the league. As a result, it might be worth looking strictly at free agents. There are still quirks in the rules like restricted vs. unrestricted, exceptions and so on but that gets us somewhat closer to how the market actually values players.
Re: Anybody working on any exciting projects for the fall?
Sure. Worth checking that subgroup.
Quickly (and perhaps imperfectly) eliminating rookie scale contracts from the 2014-15 dataset, I get a modest improvement in the overall correlation to .456 from .432 for everyone. Doesn't change things much.
Quickly (and perhaps imperfectly) eliminating rookie scale contracts from the 2014-15 dataset, I get a modest improvement in the overall correlation to .456 from .432 for everyone. Doesn't change things much.
Re: Anybody working on any exciting projects for the fall?
Crow, have you by chance checked the correlation for the 2009-10, for the data minus rookie contracts?
On a, perhaps, somewhat related point, what seems rather odd to me about the 2009-10 data compared to 2014-15 is that pertaining to +2 players.
2009-10 sees 26 such players with salaries about $8 million and above and 11 below, for 37 in all.
2014-15 by contrast sees 31 above $8 million and 33 below, a total (unless I've miscounted) of 64.
So, what we see is not only that there are gobs more very good players last year, but they are "unexpectedly" underpaid.
So, my first question pertains to the RAPM estimates used. Jeremias' RPM shows 74 +2 and higher last year, and unfortunately the http://stats-for-the-nba.appspot.com/ is no longer so we cannot access the more or less (?) comparable xRAPM for 2009-10. This said, I do not recall any large anomaly in the number of very good NBA players across years. If anything, my memory was one of enduring stability in the distribution.
The second question (returning to my request of Crow) is: were these grossly underpaid players last year mostly under rookie contracts, thereby explaining the deterioration of the correlation over the last five years?
On a, perhaps, somewhat related point, what seems rather odd to me about the 2009-10 data compared to 2014-15 is that pertaining to +2 players.
2009-10 sees 26 such players with salaries about $8 million and above and 11 below, for 37 in all.
2014-15 by contrast sees 31 above $8 million and 33 below, a total (unless I've miscounted) of 64.
So, what we see is not only that there are gobs more very good players last year, but they are "unexpectedly" underpaid.
So, my first question pertains to the RAPM estimates used. Jeremias' RPM shows 74 +2 and higher last year, and unfortunately the http://stats-for-the-nba.appspot.com/ is no longer so we cannot access the more or less (?) comparable xRAPM for 2009-10. This said, I do not recall any large anomaly in the number of very good NBA players across years. If anything, my memory was one of enduring stability in the distribution.
The second question (returning to my request of Crow) is: were these grossly underpaid players last year mostly under rookie contracts, thereby explaining the deterioration of the correlation over the last five years?
Re: Anybody working on any exciting projects for the fall?
Good questions. I'll see what I can do.
On the second question, I only count 6 guys over plus 2 on rookie contracts in 2014-15.
You are right that a low salary - good performance case hurts the correlation and the reputation of the global market but it is a success for one GM. So the story needs to be bigger than the general correlation of pay and performance.
On the second question, I only count 6 guys over plus 2 on rookie contracts in 2014-15.
You are right that a low salary - good performance case hurts the correlation and the reputation of the global market but it is a success for one GM. So the story needs to be bigger than the general correlation of pay and performance.
Re: Anybody working on any exciting projects for the fall?
Without rookie contracts the 2009-10 overall correlation falls from about .65 to about .57. The offensive correlation fell from .64 to .61. Defensive from .33 to .15.
So it appears that the correlation between performance and pay without rookie contracts between 2009-10 and 2014-15 fell significantly from .57 to about .46 overall.
But it improved noticeably relative to the correlation with rookies between 2009-10 and 2014-15.
How much do these changes involve chance, CBA rules and / or application or misapplication of analytics or various forms of subjectivity / emotion (to hold on to guys or grab them)? I dunno.
Is any or all of this a result of evolution in RAPM / RPM? I dunno.
I didn't do a separation correlation for rookie contracts but it appears from the trends that the rookie scale does a better job of paying rookies overall than the GMs do at paying vets when performance is measured by RAPM / RPM.
Could you construct a general guidance vet pay scale based on age and previous RAPM/RPM and perhaps other factors that would outperform the GMs? I am guessing yes. Probably pretty easily. Someone with adequate time & interest could retrodict this and then offer it as prediction going forward.
So it appears that the correlation between performance and pay without rookie contracts between 2009-10 and 2014-15 fell significantly from .57 to about .46 overall.
But it improved noticeably relative to the correlation with rookies between 2009-10 and 2014-15.
How much do these changes involve chance, CBA rules and / or application or misapplication of analytics or various forms of subjectivity / emotion (to hold on to guys or grab them)? I dunno.
Is any or all of this a result of evolution in RAPM / RPM? I dunno.
I didn't do a separation correlation for rookie contracts but it appears from the trends that the rookie scale does a better job of paying rookies overall than the GMs do at paying vets when performance is measured by RAPM / RPM.
Could you construct a general guidance vet pay scale based on age and previous RAPM/RPM and perhaps other factors that would outperform the GMs? I am guessing yes. Probably pretty easily. Someone with adequate time & interest could retrodict this and then offer it as prediction going forward.