This was supposed to be a response to an email about David Gassko's defensive metric "range" (read The Hardball Times here and here), but my reaction got way out of control. I admit ahead of time that I'm channeling a general dislike of defensive statistics into my argument against this particular one. Anyway, here we go...
1) Here's a list of the most egregious goofs, keeping in mind that Gassko only provided his top and bottom 5 for each position:
* Jeromy Burnitz, +15
* Kevin Millar, +13
* Andruw Jones, +2 (ish)
* Vernon Wells, -22
* Alex Rodriguez, -27
(I'm not even including the stuff that's off-the-wall in my personal opinion, such as Miguel Tejada and Jose Valentin (???) being the best shortstops of 2004... I'm sticking to generally-accepted-as-fact-by-knowledgeable-analysts assessments here.)
How can any of these results be considered authoritative when Gold Glove defenders are placed dead last at their positions? And how could Burnitz and Millar be above 0? Something's rotten in the state of Denmark. What kind of margin of error are we working with, +/- 50??? Gassko addresses Wells' bizarre -22 by saying, "he was +9 in 2004, won a Gold Glove and has a generally good reputation as a fielder. I don’t know what’s going on here." Well, I do... your formula doesn't work.
2) Let's take the example given in the initial article. In 2004, Derek Jeter made 392 assists, when Gassko estimates he should have had 419, good for a -27 rating raw, -16 converted to runs. But let's say there's a 5% margin of error. That's healthy, but not unreasonably big. A 5% swing would potentially bring Jeter within 392 of 398, putting him in mediocrity where he belongs, or down to 392 of 440, which makes him Edgar Renteria bad. (The ultimate insult!) Considering the sample-space problems that Gassko himself cites, how can he put any stock in his "playable" ball estimate? Given the wild data swings caused by a margin of error that's realistic for your average national poll, I consider the results unusable.
3) Defensive metrics don't work. Ever. Given the data available to civilians, defensive value is patently unquantifiable. Once the starting position and path of the fielder, speed and path of the batted ball, and speed of the baserunner down the path can be tracked, thus allowing "playability" of each ground ball to be calculated, then we can talk. Until then, any negative stat can be refuted by good old anecdotal evidence, as can any positive stat.
Defensive sabermetrics is therefore, in my mind, an intriguing, well-intentioned, utter waste of time.
Slap!
1) Here's a list of the most egregious goofs, keeping in mind that Gassko only provided his top and bottom 5 for each position:
* Jeromy Burnitz, +15
* Kevin Millar, +13
* Andruw Jones, +2 (ish)
* Vernon Wells, -22
* Alex Rodriguez, -27
(I'm not even including the stuff that's off-the-wall in my personal opinion, such as Miguel Tejada and Jose Valentin (???) being the best shortstops of 2004... I'm sticking to generally-accepted-as-fact-by-knowledgeable-analysts assessments here.)
How can any of these results be considered authoritative when Gold Glove defenders are placed dead last at their positions? And how could Burnitz and Millar be above 0? Something's rotten in the state of Denmark. What kind of margin of error are we working with, +/- 50??? Gassko addresses Wells' bizarre -22 by saying, "he was +9 in 2004, won a Gold Glove and has a generally good reputation as a fielder. I don’t know what’s going on here." Well, I do... your formula doesn't work.
2) Let's take the example given in the initial article. In 2004, Derek Jeter made 392 assists, when Gassko estimates he should have had 419, good for a -27 rating raw, -16 converted to runs. But let's say there's a 5% margin of error. That's healthy, but not unreasonably big. A 5% swing would potentially bring Jeter within 392 of 398, putting him in mediocrity where he belongs, or down to 392 of 440, which makes him Edgar Renteria bad. (The ultimate insult!) Considering the sample-space problems that Gassko himself cites, how can he put any stock in his "playable" ball estimate? Given the wild data swings caused by a margin of error that's realistic for your average national poll, I consider the results unusable.
3) Defensive metrics don't work. Ever. Given the data available to civilians, defensive value is patently unquantifiable. Once the starting position and path of the fielder, speed and path of the batted ball, and speed of the baserunner down the path can be tracked, thus allowing "playability" of each ground ball to be calculated, then we can talk. Until then, any negative stat can be refuted by good old anecdotal evidence, as can any positive stat.
Defensive sabermetrics is therefore, in my mind, an intriguing, well-intentioned, utter waste of time.
Slap!
posted by Jeff at
5:35 PM
0 Comments:
Post a Comment
<< Home