Big Pimpin wrote: noseeum wrote:
Big Pimpin wrote:
It's long been known that certain metrics (UZR especially) has had problems with LF in Fenway (and Enron/Minute Maid) because of the wall. And like I said before, I agree that we can't have complete confidence in any one number. But if you take other systems and multiple years and look at the entirety of the data and then put a range around it, I think you can be pretty confident in the result.
For instance, I may not agree that Manny is a -18 LF because that's his UZR number for 2007, but if I look at UZR and PMR and Chone (I didn't, I'm just making this argument) and come to the conclusion that he's roughly a -15 LF then I maybe I can reasonably think he's a -10 to -20 LF. Knowing the problems with Fenway and wanting being conservative, I would probably value him as a -5 to -15 LF. I guess my own personal opinion is simply that the defensive data and information is very useful, even if one particular metric or year or whatever isn't completely 100% reliable.
Yes, but to go from "this is useful" to "this definitely proves the Bradley + RP is more valuable than Dunn" is a bit of a stretch no?
Forgetting about what the stats say, I think you could also argue that defensive stats are meaningless for Bradley. He only played 20 games in the outfield last year, and he hasn't spent significant amounts of time in CF since 2005. He's suffered injuries every year since then. His admittedly stellar offensive output last year was as a DH. He's never matched it while playing a lot in the outfield. I think there's plenty of reason to be skeptical of relying on Bradley, and I think there's plenty reason to wonder how good he is defensively at this point in his career. He is of course way better than Dunn, but exactly how much better and what it means for value is very much in question IMO.
To be fair, I really didn't set out to prove that, it was a byproduct of "there's no way in hell 100 games of Bradley and 160 of Dunn are comparable" and I'm still confident I proved that. In fact I was surprised and just how much better Bradley rated. Anyway, it's especially true if Bradley's replacement is those 60 games is actually a guy who's pretty close to league average like Fukudome.
I also assumed significant offensive regression for Bradley (well I didn't, but James and Marcel did
). I think you're right about not knowing what Bradley's true defensive ability is, but I think there's as many questions about Dunn. Dunn rated very poorly last year, is that his true talent level? Is he fading as he gets older? Those things are very possible but weren't included in the numbers because I averaged the past 4 years.
Anyway, I didn't want to do a range and I didn't want to take the time to look at PMR because I didn't have the time, but I'm pretty confident that if I had I would have still found them pretty comparable in the same analysis.
Big Pimpin, I don't think you're quite getting my point about the metrics. You believe you've proven that Bradley+RP is better than Dunn.
What I'm saying is, all you've proven is that using one defensive metric and one offensive metric, as well as a certain way of combining those two metrics into a comprehensive number, Bradley + RP returns a higher number than Dunn. That, to me, does not prove anything because believing that result requires proof of a few other things first. In logic terms, you are treating as axioms things that need to be proven first, and I don't believe they have been proven.
In order for that to be a proof, you need several things:
1. Proof that the offensive metric is a reliable estimate of a player's relative offensive value (I would say we're pretty much covered here). We'll call offensive value "X"
2. Proof that the defensive metric is a reliable estimate of a player's relative defensive value (I think the case is less strong here. Yes, we can judge who's better and who's worse, but I don't think we can say that player A is definitively 5% better than player B, or even that in a past season, player A was 5% better than player B.). We'll call offensive value "Y"
3. Proof that offensive contributions from a specific position represent a certain percentage of a player's value and proof the defensive contributions from a specific position represent a certain percentage of a player's value (at this point we have a bunch of theories and attempts at quantifying this, but I don't think we have any proof that one system is better, worse, etc. There needs to be a whole lot more research before I can have any confidence on this one.), i.e. the sum of offense and defense = "Z".
So you're arguing, and fangraphs and other sites that come up with these sum total metrics, are arguing, that for every player, you can get X and why, and further X+Y=Z. And whichever player has the highest Z, that's the player you want. I just don't think it's anywhere near that precise yet.
The best performance prediction systems for baseball never get a higher correlation than about .6. Even for offense! Defense would be far worse I'm sure. So even if you accepted you had proof that points 1, 2, and 3 were accurate, all you can say with confidence is that over the past 4 years, 100 games per year of Bradley + 62 games of RP at LF would have been
better than Dunn. That doesn't mean that in 2009 that would be the case. But as I said, I don't even think we can be confident in that assessment even for the past four years.
Last edited by noseeum on Sun Jan 04, 2009 7:02 pm, edited 1 time in total.