RAPM aging curve

[DSMok1]

I contacted Tango Tiger about the best article summarizing the aging and survivor bias issues, and he directed me to the following series by Michael Lichtman:

http://www.hardballtimes.com/how-do-bas ... ge-part-1/

and

http://www.hardballtimes.com/how-do-bas ... ge-part-2/

The second article is an absolutely excellent piece on the survivor bias issue and how to deal with it.

I hope this helps, J.E.!

[J.E.]

adding an Injury variable

That definitely seems useful. I wish we had more easily accessible injury data, but even then I'd have to go through most of the injuries and put them in certain 'injury'-groups. Then the question becomes how detailed you want to do things, which becomes a problem over over-simplifcation vs. sample size. Group ACL and MCL injuries together and you might be oversimplifying, don't group them together and your sample on MCL is very small

The real value is going to be found in adjusting for position/player type, as most empirical and anecdotal evidence says offense peaks earlier and defense (particularly big man defense) peaks later and declines more gradually.

I'm going to try and do that the next few days

can you post a graph w. O + D?

Thanks for all the links, DSMok1

http://www.hardballtimes.com/how-do-bas ... ge-part-2/

This seems weird to me because it seems they're arguing for less of a drop-off, while I'm actually expecting more of a drop-off once you try to account for survivor bias

[DSMok1]

Thanks for all the links, DSMok1

http://www.hardballtimes.com/how-do-bas ... ge-part-2/

This seems weird to me because it seems they're arguing for less of a drop-off, while I'm actually expecting more of a drop-off once you try to account for survivor bias

Well, run through it and we'll see what happens. I'm sure there IS a survivor bias; I'm just not sure what the effect is in basketball vs. baseball.

By the way, I did this all a while back with WS/48: http://godismyjudgeok.com/DStats/APBRme ... =2598.html .

[J.E.]

Ran the whole thing for the 5 different positions and here's what it looks like (aligned at 19)



I guess two things to notice is that SG's improve the least between 19 and 30 (lowest coefficient 10 out of 12 times). Something to keep in mind when drafting, I guess. PF's seem to improve the most but that definitely could be just noise. Everyone else is so close together that it's probably better to keep them in one group to increase sample size and counteract the noise

I used player weight to create 5 equally large player groups, with the players of the lowest weight being classified as PG and so on. If someone has a good argument to do things differently, let me hear it. Although I can't imagine that things would look very differently with official position designations

[colts18]

Did you use an aging curve on your 4 year and 10 year RAPM studies? I'd imagine its hard to do it because its difficult to adjust for guys who improved much faster than their aging curve (ex: Dragic).

[J.E.]

Did you use an aging curve on your 4 year and 10 year RAPM studies? I'd imagine its hard to do it because its difficult to adjust for guys who improved much faster than their aging curve (ex: Dragic).

I'm not exactly sure what you're asking. This is 13 years of matchupdata thrown into one pile, with player ages as dummy variables

[nbo2]

Did you use an aging curve on your 4 year and 10 year RAPM studies? I'd imagine its hard to do it because its difficult to adjust for guys who improved much faster than their aging curve (ex: Dragic).

I'm not exactly sure what you're asking. This is 13 years of matchupdata thrown into one pile, with player ages as dummy variables

So won't this be very influenced by sample bias, causing the ends of the curve to be inflated?

[J.E.]

So won't this be very influenced by sample bias, causing the ends of the curve to be inflated?

Yes - we were discussing this problem a few posts up