from Amazon
Monday, March 22, 2010
Mike Silva chronicles: UZR and Teix
I greatly enjoy discussing sabermetrics with Mike, and we had another Q & A this morning:
>
> 1. Do you agree that UZR has some flaws, especially with the zones between
> first and second? (Something I saw discussed on a site)
I would say that UZR has no “flaws”. If it had, MGL would correct them. There are design limitations or design assumptions. So, how the shared zones are handles is always an issue from creator to creator. MGL, Dewan, Pinto, Jensen, me… we all handle things differently in that regard. Those are always fair discussion points.
>
> 2. Does it take into account scoops (I believe there is a separate
> metric),
> outs, holding runners, etc.
UZR itself doesn’t handle scoops. However, MGL has created a separate metric to handle scoops that he has not incorporated into UZR. Rally at BAseballProjection.com also has done it (though I don’t know if it’s incorporated into his TotalZone metric). Rally generated the career best-list in “scoops” a couple of years back, and MGL replicated his study in that regard. Scoop is a skill that adds a couple of runs. Maybe 4 or 5 at the extreme case.
>
> 3. Do you think we undervalue a position that, ultimately, outside of the
> C
> and P, touches the ball more than anyone else on the field? Especially in
> WAR.
Number of touches really doesn’t matter. It touches times degree of difficulty. You can make the argument that catchers might be undervalued in WAR, given how unique the position is. But, you’ve got a huge segment of players who have played 1B, for the first time, in their 30s. It’s a hard case to make to say that the position is undervalued in WAR. Not to say I’m right and you’re wrong, but you’ll have to make the case.
>
> 4. Whats the best way we can improve UZR for accuracy?
GIGO (garbage in, garbage out). The lifeblood of any stat is the data, and how it’s recorded. You need to get more granular data, and we’ll be there in the next 2 years with HITf/x and FIELDf/x. We’ll finally be able to figure out exactly how hard a ball is hit, exactly the distance and spray angle and launch angle, and number of balls hopped, and how fast the batter-runner is, and where all the fielders are playing and how much they need to run to get to the path of the ball, etc.
Interestingly, the validation of UZR will come when we won’t need UZR in its current form. We’ll be able to see how good UZR was at capturing what it tried to capture with its limited data feed. We’ll finally be able to see how much all the extra data with its specific granularity adds to UZRf/x (whenever that comes out).