APBRmetrics

Hi All-

I recently drew on the work v-zero did in this thread viewtopic.php?f=2&t=8327 to create a version of Alternate Win Score that is adjusted for the Usage-Efficiency tradeoff. I rather uncreatively dubbed it Usage Adjusted Rating (UAR). Here are the results http://nbacouchside.com/2013/11/21/intr ... ed-rating/

One cool thing about it is that it tracks relatively closely with PER, which just confirms to me that Hollinger is an evil genius way ahead of his time.

So I just ran the correlation for UAR with PER and the correlation was .964. Wow.

nbacouchside wrote:So I just ran the correlation for UAR with PER and the correlation was .964. Wow.

Umm..that would tend to happen when the basis of the stat is the boxscore PER formula.

The underlying basis for it was Alternate Win Score, Jacob, not Game Score.

How are you incorporating this? As in, how does the equation look now?

So the way I did it was calculating the non scoring or turnover aspects of Alt Win Score. League average year over year for those was roughly 5.55. Then I calculated the league average Score Rating which is just 5*PPP+.076*Poss, where Poss = FGA+.44*FTA*TOV. Lg Avg Score Rating was roughly 6.2 year over year. So then I calculated Score Rating for everyone and subtracted out 6.2 to scale it so average scorers are 0 in Score Rating. I then multiplied Score Rating by 3.58 to get it on the same scale as the Non-Scoring or Turnover aspects of AWS and to weight it roughly 4 times as much as the rest, since Score Rating includes Efficiency, FT, and Turnovers which are 80% of the value by the four factors and rebounding makes up the remaining 20%. Initially, I used a factor of 3, but that wasn't really based on anything other than an eyeball guess. I like 3.58 better because it reflects the Four Factor weights, roughly, when you place Scoring and Non-Scoring on the same scale.

Because league average scoring rating is 0, the league average ranking for UAR is just the league average of the non-scoring aspects or roughly 5.55.

To calculate UARwins, you can just divide UARtotal by 55.9. That's just the divider that makes the numbers work with the particular scale of UAR. Kevin Love leads the league in UARwins with 2.5 wins, which is roughly equivalent to his 2.6 WinShares from basketball-reference.com, which also leads the league thus far.

The average UAR for all players in this set was 4.7. The scoring rating average -0.6, so it is basically rewarding the above average scorers only. The non-scoring rating averaged 5.3. So UAR really rewards those that are strong on non=scoring impacts. Including rebounders. Only 4 of the top 20 on UAR are under 6-8. Are scoring and rebounding being put on different scales? Appropriately or not? This is a late posting of thoughts on the topic so I am not sure what else I should have added.



The average PPP is just under .9

I need to fix some stuff with UAR. I tried to adjust so that scoring was actually valued 2.7 times the rate of the non-scoring stuff (mirroring the weight of the values in the scoring factor - fta, efg, tov, versus their weights among the four factors). I'm not sure why non-scorers are still being so heavily favored, unless I've done something wrong in my spreadsheet, which is possible. I think it might have to do with how I've subtracted out the average level of production so that below average scorers are negatives and above average scorers are positive. That might be a dumb way to do it. It might also just be that I've built something flawed and useless.

Take some time and you will probably figure it out and make it what you intended it to be or now want it to be.

I made some tweaks and came up with something better, I think.

http://nbacouchside.com/2013/12/17/anot ... -uarpm100/