Reverend said:
Great post. I was going to post that I know more about the macro level stuff on fielding stats and the problems in using them for projecting especially in smaller samples but far less about how they work at the micro level and the variation involved in what they report, but you already addressed it. Good show.
To follow up, is there any sense of what the error bars would be at different samples, or does the methodology not allow for that?
Also: you should post more. Seriously. Some of us have literally been talking about this. But you might have a life or something, so, like, whatever.
Thanks. I don't know what the error bars are like, the UZR primer says that after a year, it's probably okay to regress halfway to zero and have a conservative estimate. However, because SS get more chances, the sample size is larger in a shorter amount of time.
Might be illustrative to look at the available data from Fangraphs. In 2013 at SS, Xander had 17 balls hit into his "zone", and made 16 plays on them. He also made plays on 5 balls out of zone. That gave him a 53.2 UZR/150 (compare to Andrelton Simmons' 24.9 during his time in the league). Meanwhile, he had 12 balls in his zone as a third baseman, made 6 of them, and made 1 play out of zone, for a UZR/150 of -47.3. Both of those numbers are absurd, and what's more they feel, on the face of it, to represent the opposite of what we observed: he was a reliable, if unspectacular, third baseman, and a rookie shortstop.
This year, Xander's had 84 balls in zone, made 65 plays (.774), and made 8 out of zone (0.09 / inning), good for a UZR/150 of -3.6 (that's split evenly between missing double plays, bad range, and errors). That's in 302.2 innings this year. Compare to Drew last year, who played 1093.1 innings at shortstop. He had 287 balls in zone, made 226 of them (.787), and got to 55 balls out of zone (0.05 / inning). Drew had a 6.7 UZR/150 last year (compared to 10.1 and 8.4 in his final two healthy seasons in Arizona). My gut is that after half a season, we can probably start to classify X into a few bins: "bad, below average, above average, or good."
As others have said, Drew is hopefully a good comp because he was really bad his first few years (-12.6, -13.9, -17.7), and then became an asset defensively. I don't know that I totally agree with the assessment that Xander's main problem has been his range. It's that he doesn't know how fast an average major leaguer gets to first, and so rushes plays that he has more time on. And he doesn't know physically what to do with himself when a play is a little off-kilter. He looks bad because he butchers a bunch of balls that he gets to, which Drew would have made comfortably.
I said this elsewhere, but Butterfield et al. mentioned that Xander is learning not just how to be a shortstop in general, but how to play on very different surfaces. I don't think it was a coincidence that there was a bunch of talk about his fielding when they were in Toronto (super fast artificial surface), basically quieted down when they were at Fenway, and is back now that they're on the road again.
EricFeczko said:
Thanks for the post. This answers my question of why UZR/DRS are less valid than more concrete measurements.
I'm confused why UZR would be scaled to an average player for a single season, given that such data may itself be unreliable. Why not scale across a six-year (or 10 yesr) span?
That seems like just a decision that Fangraphs made so that UZR would be comparable to league average. I agree that on the face of it, it probably makes the statistic less reliable.