Is "Win Score" a valid way to evaluate players?

[Wolves2011]

Some on this board have said win score isn't a good stat to look at.

The way to test any theories validity is to test it,.

Does it PREDICT?

lets look at win score for every team in the NBA in 2007-2008 and compare it to their record.
[I'm using 2007-2008 because the wages of wins journal hasn't posted all stats for all teams from last year yet.]

Keep in mind that the prediction are made by looking at things like points, rebounds, assists, steals, blocks etc and applying a "weighting" to them. That provides a win score for an individual player. It then adds up the win scores for all the players on the team and compares that to "reality".

As you can see below, all of the predictions except 4 were accurate within plus or minus 3.5 games. Most are within a game or two.

The model for predicting wins has validity.

The stats that the model says are important [and the model is "weighted" to reflect that] do predict wins.

Atlanta predicted wins 36.3, actual 37
Boston predicted wins 68.3, actual 66
Charlotte predicted wins 29.5, actual 32
Bulls predicted wins 32.8, actual 33
Cavs predicted wins 40.2, actual 45
Mavs predicted wins 53, actual 51
Nuggets predicted wins 51, actual 50
Pistons predicted 60.6, actual 59
Warriors predicted 46.8, actual 48
Rockets predicted 53.3, actual 55
Pacers predicted 37.2, actual 36
Clippers predicted 21.6, actual 23
Lakers predicted 60.4, actual 57
Grizzlies predicted 24.5, actual 22
Heat predicted 18.2, actual 15
Bucks predicted 22.8, actual 26
Wolves predicted 22.8, actual 22
Nets predicted 27.6, actual 34
Knicks predicted 23.6, actual 23
Hornets predicted 55.1, actual 56
Magic predicted 55.6, actual 52
76ers predicted 42, actual 40
Suns predicted 54.5, actual 66
Blazers predicted 38.7, actual 41
Kings, predicted 35.1, actual 38
Spurs, predicted 53.6, actual 56
sonics predicted 17.7, actual 20
Raptors predicted 48.8, actual 41
Jazz predicted 58.2, actual 54
Wizzards predicted 40.4, actual 43

An example of win score for a team is the 2007-2008 Celtics

The full team in 2007-2008
KG 17.9
Pierce 13.2
Rondo 10.5
Ray 8.1
Perkins 5.8
Posey 5.1
House 3.8
Powe 3.6
T.A. 1.3
PJ Brown 0.6
Pollard minus 0.1
Pruit minus 0.1
Cassell minus 0.1
Davis minus 0.2
Scalabrine minus 1.2

Total 68.3, actal wins 66

Lets also look at the 2008-2008 Celtics.

Rondo 17.2 wins
KG 11.6 wins
Ray 10.6 wins
Pierce 10.2 wins
Perk 5.0 wins
Powe 4.8 wins
House 4.7 wins
Tony 1.6 wins
Pruitt 0.2 wins
Giddens 0.1 wins
Walker 0.0 wins
O'bryant minus 0.1 wins
Moore minus 0.1 wins
Marbury minus 1.1 wins
Scalabrine minus 1.1 wins
Davis minus 2.5 wins

Total predicted wins 61.1, actual 62 wins

The model does a very good job of predicting wins [there are always statistical outliers in any model like this. ]

The model says "how much" to weight a stat = a point, rebound, assist, steal, block to get a win.

Its a good player evaluation tool.

The best available.

[captain green]

I also like this and use it quite a bit. " proven now could I interest you in a pair of zircon encrusted tweezers"?

[Bluewhale]

To predict team is quite different to predict the player's value. You cannot compare apple and orange.

[sully00]

I may or may not totally understand this thing, but as I understand the formula it undervalues points and overvalues steals. But beyond the formula it also doesn't pass the smell test. Most of the problem would seem to be in the positional weighting. But it also lacks consistency.

Rondo is not worth nearly twice as many wins as Paul Pierce. But even more importantly is Pierce, he has a better in year in '08-09 with a higher raw score and tallies a position adjusted score that is 3 points lower.

So besides the fact that it doesn't tell you what it claims to tell you and isn't consistent year to year it is great.