Tuesday, June 08, 2004

Money's Worth, part 2 

Last week I discussed a proposed new stat I called Ballplayer Value, which was a shot at evaluating whether or not a team was getting "value" from a ballplayer by taking that player's VORP and dividing it be ratio of their salary to the league median (about $800K in 2004). I provided non-pitcher BVs; here are D-Back pitchers' BVs (I assumed $300K salaries for Good, Bruney, and Service):

Pitcher Salary ($M) VORP BV
Randolph, S. .323 7.8 19.3
Good, A. .300 3.7 9.9
Bruney, B. .300 1.3 3.5
Service, S. .300 0.7 1.9
Koplove, M. .330 0.5 1.2
Johnson, R. 16.5 21.9 1.1
Webb, B. .335 0.0 0.0 (salary not updated, but it doesn't matter since VORP = 0)
Valverde, J. .320 0.0 0.0
Nance, S. .308 -1.4 -0.5
Choate, R. .326 -2.1 -0.9
Villarreal, O. .325 -2.9 -1.2
Fossum, C. .345 -4.4 -1.9
Daigle, C. .300 -10.3 -3.9
Sparks, S. .500 -8.2 -5.1
Dessens, E. 4.0 -4.5 -22.5
Mantei, M. 7.0 -8.4 -73.5

I don't have much to say about the chart other than it's incredibly sad. Only 3 of those VORPs are significantly positive numbers. Shudder.

Rob over at 6-4-2 kindly pointed out that my concern about VORP not taking into account playing time was, well, wrong (here are a couple links with some explanation). And so I think while the BV concept holds merit, it's going to need to be seriously overhauled. As a predictive concept, it's a good idea (if you're trying to build a starting lineup), but then I'd rather use PECOTA cards from BP or something like that. As a "hindsight-is-20/20" concept, I'm still not sure VORP is what I want to use... it's probably best to wait until the season is over.

Finally, in an unrelated note, congratulations to Luis Gonzalez, named NL Player of the Week, a nice change after what even he would admit was an abysmal May. Here's hoping he keeps it up.

And one of the pitchers responds this week like Gonzo did last week.


You'll probably want to ditch VORP for this study entirely. Since it doesn't factor in defense at all, it isn't of much use in determining a player's "true" value. SLWTS, WSAR or WARP3 are all better starting points since they incorporate a great deal more than VORP. SLWTS with heavy regression will probably get you pretty far.

Anyway, here are a couple of articles worth reading in case you haven't seen them already:



Post a Comment