Comments on: Revisiting Preseason Outfield Defense Expectations

By: Jason T

Jason T — Tue, 21 Jul 2009 12:58:42 +0000

Yes, because UZR doesn’t care for Adam Jones this season makes it worthless. Sometimes good systems makes mistakes. One could say the same thing about democracy on occasion.

By: UZR IS A JOKE

UZR IS A JOKE — Tue, 21 Jul 2009 09:30:08 +0000

Maybe Adam Jones’ UZR is so bad because UZR is a freakin’ joke. He has gotten to more balls than any other CF in the league. He has more putouts than expected outs, and I’m to believe the only other CF he is better than in RngR is Vernon Wells? Give me a break.

But keep pumping out this crap. Maybe enough GMs will start buying this garbage, and the GMs who realize the best way to judge a defenders ability is by watching him play the game, at least for now.

Maybe if the data on how the number was came by was released, it could be viewed by as many eyes as are needed to find it’s flaws. I really do hope that eventually there are some meaningful defensive metrics out there, but as for right now, UZR is not meaningful.

By: Grant

Grant — Sun, 19 Jul 2009 03:36:16 +0000

I have only see Reimold play a few times, but from what I saw he is a well below average defender. Just seeing the route’s he takes, his body type and instincts when actually making the catch, makes it hard for me to believe that he will ever be league average. I’d say -15 to -5 at a corner spot.

By: MGL

MGL — Sun, 19 Jul 2009 02:38:05 +0000

I didn’t mean to be snark in my first comment above – actually I did. The reason is that words like “stabilize” are meaningless in this context. Obviously (I think everyone who has ever heard of this website knows this) the more data you have, the more it represents a player’s true talent, assuming that the data is a reasonably accurate reflection of that talent. There is no magic number at which point it “stabilizes.”

And I have no idea what this means (the words “career trends”):

“One main problem test defensive metrics is that they are new and no good career trends can be made yet.”

But the poster states it as a fact, so I have no doubt that it is true, especially since that poster is actually a major league player…

By: MGL

MGL — Sun, 19 Jul 2009 02:31:38 +0000

It depends a little on the position and no one know for sure, but a good rule of thumb is that after a season’s worth of UZR, you can regress his UZR 50%, maybe a little less. Now that is based on back to back years of empirical UZR and does not necessarily mean that regressing one season 50% tells is a player’s true talent defensive value, but it is probably close. The reason that is the case is that if UZR is actually not an accurate stat it still could be reliable enough for only a 50% regression after 1 year. For example, if we give a child an IQ test once, it is likely that each child with a high or low score will regress – and let’s just say that it is 50% – if we give him another test – the same test. But if the test is a crappy one – i.e., it does not measure intelligence well, regressing that first IQ test 50% is not going to give us a good estimate of that child’s intelligence. It will only give us a good estimate of how he will score if he takes the test again.

By: Dave Allen

Dave Allen — Sat, 18 Jul 2009 03:57:46 +0000

After 100 coin flips I am over 95% confident I will be between 40 and 60 heads. Could one make a similar statement about UZR, after N innings I am 95% (or 50%) confident a player’s UZR/150 is within X runs of that player’s true talent.

By: MGL

MGL — Sat, 18 Jul 2009 03:39:56 +0000

How many coin flips does it take for the heads/tails percentage to stabilize? 631? 285? 1,318?

By: Dave Allen

Dave Allen — Sat, 18 Jul 2009 01:28:29 +0000

Yeah, thanks Jeff.

By: xeifrank

xeifrank — Sat, 18 Jul 2009 01:20:32 +0000

So around 900 or so innings for 50% regression?
vr, Xei

By: Jeff Zimmerman

Jeff Zimmerman — Fri, 17 Jul 2009 23:21:55 +0000

The number of games for a season is around 100-125 for a 50 regression. I got the number from MGL and running it against a few tests it seems correct. One main problem test defensive metrics is that they are new and no good career trends can be made yet