I was new to this community and is working on something related to player metrics. I happen to notice some discussions about the RAPM but could not find Sill's paper 'Improved NBA Adjusted +/- Using Regularization and Out-of-Sample Testing' Can anyone kindly direct me where I can find this?
I don't have it anymore but I know that a couple people on this board still do (I also have a slight recollection of it being on his site still). In the meantime, I know you can contact Joe and he would be happy to provide it for you.
Yep, anyone who wants the paper (which was presented in 2010, actually) can send me an email at joe underscore sill at yahoo and I'll send the pdf to you.
I have asked various people associated with the Sloan conference to put a link to the pdf on the conference page for the paper, without success:
I also love how it says 'VIDEO WAS NOT RECORDED FOR THIS TALK'. In the interest of accuracy, that should be edited to say 'VIDEO WAS RECORDED FOR THIS TALK, BUT WE LOST IT'.
Normal priors make for easy and familiar math, but they seem like an odd choice. The NBA represents the top millionth of the population with respect to basketball ability, so while NBA players are all borderline superhuman, we should expect most of them to be relatively close to 'replacement level' in talent.
I primarily am interested in how Joe did crossvalidation. In my paper I also did something similar. Also I think that ridge regerssion makes better sense because the data points do have different qualities. (I will not leave the explanation here but if anyone is interested I can write more on this)
