Re: Trade Paul George
Posted: Mon Jul 03, 2017 1:10 am
If you look at the distribution of RAPM ratings for all players in a given season, they don't scale linearly. For instance, the difference between the 1st and 10th ranked players in J.E.'s 2017 single season RAPM was 3.44, whereas the difference between the 90th and 100th ranked players was only 0.11. In my mind, that means it becomes harder to improve your RAPM as the starting point becomes higher.
By my count, only 17 players since 2001 have had a +6 or greater single year NPI RAPM according to the collection I have from J.E. So if i was just guesstimating that player's RAPM for year 5 with that knowledge, I would probably put it somewhere around 4.5. If I just did a simple projection where i took 85% of his prior year RAPM and then added a simple age adjustment (assuming he was 23 entering his 5th year w/ an aging adj of (27 - age)*0.06), the projection would be 3.64. Using the 2.53 value as a prior and adding in the same aging adjustment gives me a rating of 2.77.
I think this shows the shortcomings of not applying an aging adjustment while actually running the RAPM. That would probably result in better predictive measurement since it wouldn't penalize the player as much for his first two years. Unfortunately I don't currently have age information included in my data so it's not something I can easily fix.
By my count, only 17 players since 2001 have had a +6 or greater single year NPI RAPM according to the collection I have from J.E. So if i was just guesstimating that player's RAPM for year 5 with that knowledge, I would probably put it somewhere around 4.5. If I just did a simple projection where i took 85% of his prior year RAPM and then added a simple age adjustment (assuming he was 23 entering his 5th year w/ an aging adj of (27 - age)*0.06), the projection would be 3.64. Using the 2.53 value as a prior and adding in the same aging adjustment gives me a rating of 2.77.
I think this shows the shortcomings of not applying an aging adjustment while actually running the RAPM. That would probably result in better predictive measurement since it wouldn't penalize the player as much for his first two years. Unfortunately I don't currently have age information included in my data so it's not something I can easily fix.