Post by eric on Dec 12, 2018 19:47:35 GMT
I took the attributes of all active players to play at least 10,000 MP, scaled them by the wins per attribute described here, and regressed that against their actual win shares per 48. The result had an R^2 of .402 and a root mean squared error of .036 - that is, the prediction would be within .036 of the actual about two out of three times, so a player it thought was a solid .150 type (e.g. 9.4 wins in 3000 MP, or Mark Price) could plausibly be actually as good as .186 (11.6 / Shamorie Ponds) or as bad as .114 (7.1 / Skylar Diggins).
But these values are for how good the same player would be if they added a given amount of attribute, whereas our project is how good a player is who happens to have that much attribute. Finding the best fit to our 10,000 MP player set, we adjust our values from (table reordered and rounded to the nearest five)
to
Note that this is not saying that Shot Blocking for example is not as good of an upgrade as we thought. It is instead saying that it is not as good an indicator of how good a player overall is, which is an entirely different kind of analysis altogether. Still, there is very good correlation between the two analyses, which is nice.
Overall our regression rises to an R^2 of .662 and RMSE of .027. Now our predicted 9.4 Win Share player lies in a plausible band of 7.7 to 11.1 Win Shares.
.
So what?
.
So here are the currently unsigned free agents:
6 is league average level, so it's not like there are championship starter caliber players still out there. But there's plenty of overqualified backups.
Good luck!
But these values are for how good the same player would be if they added a given amount of attribute, whereas our project is how good a player is who happens to have that much attribute. Finding the best fit to our 10,000 MP player set, we adjust our values from (table reordered and rounded to the nearest five)
pg sg sf pf c attribute
50 70 70 80 80 InsideScoring
70 65 65 70 80 JumpShot
35 40 25 20 30 ThreePointShot
60 30 45 30 35 Handling
-20 -10 -10 -20 -20 Passing
20 10 0 5 -10 Quickness
-15 5 15 5 -5 PostDefense
5 5 10 5 0 PerimeterDefense
5 -5 5 15 5 DriveDefense
5 15 10 10 15 Stealing
25 35 40 55 55 ShotBlocking
25 30 20 25 25 OffenseRebound
20 30 25 35 25 DefenseRebound
10 20 35 35 20 Strength
-5 -5 0 0 -5 Jumpingto
pg sg sf pf c attribute
100 80 60 50 95 InsideScoring
70 65 65 70 85 JumpShot
40 35 25 20 25 ThreePointShot
55 40 45 35 35 Handling
-20 -10 -15 -10 -20 Passing
20 5 -5 -5 -5 Quickness
-10 5 20 5 -10 PostDefense
0 10 10 0 10 PerimeterDefense
5 -10 0 15 0 DriveDefense
15 20 15 20 20 Stealing
15 45 40 50 50 ShotBlocking
15 15 15 25 25 OffenseRebound
20 40 30 35 25 DefenseRebound
5 15 35 35 25 Strength
-5 -5 -5 0 -5 JumpingNote that this is not saying that Shot Blocking for example is not as good of an upgrade as we thought. It is instead saying that it is not as good an indicator of how good a player overall is, which is an entirely different kind of analysis altogether. Still, there is very good correlation between the two analyses, which is nice.
Overall our regression rises to an R^2 of .662 and RMSE of .027. Now our predicted 9.4 Win Share player lies in a plausible band of 7.7 to 11.1 Win Shares.
.
So what?
.
So here are the currently unsigned free agents:
ws amount
7 1
6 4
5 13
4 18
3 19
2 23
1 16
0 20
-1 7
-2 1
-3 3
-4 26 is league average level, so it's not like there are championship starter caliber players still out there. But there's plenty of overqualified backups.
Good luck!