RealGM

Hello, sports fans. I wanted to get your thoughts on a stat I created for the purpose of comparing player values in an apples to apples way. The formula is simple enough:

2014/15 salary/1,000,000 / 2013/14 total win shares = Cost of a win share in millions of dollars (COWS)

NOTE: I used basketball reference data.

Here are the players who make greater than or equal to 8MM/year in 2014/15 ranked by COWS, from lowest cost to highest.

https://docs.google.com/spreadsheets/d/ ... KM/pubhtml

Top 20
Rank, Player, COWS
1 Stephen Curry, 0.793
2 Kevin Durant 0.989
3 George Hill 1.026
4 Kyle Lowry 1.026
5 DeAndre Jordan, 1.031
6 Trevor Ariza, 1.072
7 DeMar DeRozan, 1.080
8 Mike Conley, 1.085
9 Joakim Noah, 1.089
10 Kevin Love, 1.099
11 James Harden, 1.151
12 Lance Stephenson, 1.216
13 Serge Ibaka 1.286
14 Marcin Gortat, 1.288
15 LeBron James, 1.298
16 Tim Duncan, 1.351
17 Taj Gibson, 1.404
18 Paul Millsap, 1.418
19 Blake Griffin, 1.449
20 Paul George, 1.475

Here are the COWS for all players:

https://docs.google.com/spreadsheets/d/ ... 5I/pubhtml


Please let me know your thoughts. Obviously, there is no such thing as a perfect stat, but I think this one has some utility. Again, I would appreciate any insights or feedback. Thanks in advance.

WS and impact increase linearly and salaries don't. I don't think you'll see any correlation between the two. A Porsche 911 Turbo may go only 10% faster from 0-60, but may cost you 50% more. It may be a good way to rate players making the same money, but not across salary levels IMO.

BmanInBigD wrote:WS and impact increase linearly and salaries don't. I don't think you'll see any correlation between the two. A Porsche 911 Turbo may go only 10% faster from 0-60, but may cost you 50% more. It may be a good way to rate players making the same money, but not across salary levels IMO.

I agree that it's not a great way to compare players across salary levels. I'm working on a way to do that (because I'm also aware that WS and impact, as well as and salary and impact, have different functions).

I also agree that COWS is better for comparing players in similar salary bands.



One thing I have found is that the median COWS value of all players can be used as an effective gauge of a player's value.

The median COWS is between 2.55, almost regardless of pay band. If a player's COWS is below that median (less than 2.55), then that player is likely to be a good buy. If the player's COWS is above that median, then the player is likely not currently a good buy for one of two reasons: One, they're healthy and not producing output commensurate with their salary. And/or, two, they've lost significant time to injury. Healthy players with a COWS of 4.0 and above are very likely to be bad buys. The caveat to all of the above is that context matters.

Nice work! Have you tried fitting a model with the logarithm of salary as the explanatory variable?

The residuals for a model like this could be pretty informative:

WS = log(salary)