I'm guessing UZR is adjusted to ballpark. I know that TZR is.

I always though the metrics never really captured what was going on at Wrigley properly either. Granted it may have changed since I haven't looked at in years but back in the day CFers always seemed to look great for the Cubs.

It's intriguing enough that I'd like to see it done for all OFs Camden vs. non-Camden (pace #1).

I've pointed this out on a number of occasions: Static-zone metrics have problems in the OF in oddly-shaped ballparks, because the OFs can't position themselves where they would typically play in a ballpark with normal dimensions.

It's really hard to factor in a ballpark adjustment. If the RF has to play 20 feet closer to home plate in Camden Yards because of the scoreboard, how would that affect the expectation for each zone? You can't figure it based on plays actually made, because then you're letting the thing that you are actually trying to measure affect the measurement.

It's also not clear to me that uncatchable wall balls are factored out of UZR - STATS didn't used to flag those, but I think they started doing so right before Dewan left.

-- MWE

Wasn't this an issue with LF in Fenway as well?

UZR is ballpark adjusted. I remember reading MGL's description of the ballpark adjustment methodology about two years ago, and I had some skepticism, but there is an adjustment that has a material effect on the OF results. As I recall, Dewan does not believe that a ballpark adjustment is necessary for the DRS method.

Mike (#4) is absolutely correct, and the reason is the one he gives in paragraph 1. And it's not just ballparks. Consider a team with a good RF, a good CF, and Greg Luzinski in LF. Their CF is going to play "out of position" towards LF, and the RF will be out a bit towards CF. That changes the positions of the zones for any zone-based method. Unless some system starts tracking where the OFs play at all times, they're going to suffer from this problem. Doesn't make them useless, but it does make them sometimes dicey. - Brock Hanke

As I recall, Dewan does not believe that a ballpark adjustment is necessary for the DRS method.

Dewan has started removing uncatchable wall balls from the calculations, however - which they didn't do for most of the time he was at STATS - and there are obviously a fair number of those balls at Fenway, for example.

-- MWE

Their CF is going to play "out of position" towards LF, and the RF will be out a bit towards CF. That changes the positions of the zones for any zone-based method.

But how much does this hurt? The CF should make fewer plays in his zone towards RCF but more plays out of his zone towards LF.

You can't figure it based on plays actually made, because then you're letting the thing that you are actually trying to measure affect the measurement.

Which is why you use previous years' data to do this. We have no problem treating new parks as neutral until we build up enough data to calculate a park effect. There's no reason this should be treated any differently. Surely we have enough data on Camden Yards to adjust for this.

But how much does this hurt? The CF should make fewer plays in his zone towards RCF but more plays out of his zone towards LF.

It depends on what the penalties are for not making plays in zone, vs the rewards for making plays out of zone. In most zone-based systems, the penalty for failing to make an in-zone play in the OF is generally larger than the reward for making an OOZ play, in large part because most of the OOZ plays that you make in an oddly-shaped park save singles, whereas most of the in-zone plays that you can't make in an oddly-shaped park go for extra bases. I had estimated this back around 2000 for Fenway LF as having an effect of around 7-10 runs per season - Fenway LF being the most extreme example of the type. At the time, STATS included wall balls as in-zone plays - outfielders were getting penalized for balls that they couldn't have caught without a net on a 10-foot pole, as well as some that they *might* have been able to catch where the outfielder didn't believe he should take the risk of going all the way back to the wall. Excluding wall balls (which I think started in 2006) has mitigated this effect somewhat, but you still have areas in left-center and right-center in Fenway where in-zone plays simply aren't going to be made because the contours of the ballpark don't let the fielder play the position that way.

You can't figure it based on plays actually made, because then you're letting the thing that you are actually trying to measure affect the measurement.

Which is why you use previous years' data to do this.

.. which still skews it, because the same fielder is often there for a number of years. Are fewer plays made in RF at Camden because Nick Markakis plays the majority of the innings there, or because the ballpark makes it hard to play RF there. or some combination of both?

Conceptually, zone-based ratings are sound. But until we can overlay positioning on top of the static zone matrix, especially in the outfield (hello Field f/X), we can't entirely sort out the wheat from the chaff.

-- MWE

Consider a team with a good RF, a good CF, and Greg Luzinski in LF. Their CF is going to play "out of position" towards LF, and the RF will be out a bit towards CF.

I remember watching a game in which I saw that very defense. Only it was in the Astrodome and the LF was named Bob Watson - with Cesar Cedeno in left center.

Mike hits the point bang on. These conversations won't exist once we crack Field f/x and get 5-10 years worth of positioning, speed, range, hang-time, first-step, end positioning for cut-off throws, etc... data from it.

Imagine a function of an OF's range based on hangtime, trajectory, and distance from the ball and adjusting for weather, righty/lefty, night/day, dome/grass, etc...Neat. One could make some cool-looking graphs out of that. Such a function might take a probabilistic form where the dependent variable is the probability that a given or average player will make a certain play. I imagine this could be an approach already taken by clubs for their proprietary use. It would give teams a better read on a player's potential performance in their own park. I know Boston has proprietary defensive numbers. I'm curious as to their methodology.

#4 Even if uncatchables are factored out there's still a problem with static zones. Boston LF always grade out poorly in static zone metrics like DA and ZR because you've turned a routine 1 chance, 1 putout to either 1 chance, double allowed (DA -- which didn't factor out uncatchables) or a 0 for 0 (ZR)

Yeah, park adjustments should catch most of this, but defensive park factors are bound to be pretty noisy.

#9 Scott Fischthal did a number of studies on the matter using DA. The most interesting of which (quoted from an rsba post) This bears following up, but we were just coming to the end of the DA period and interest in the matter pretty much died. I don't recall Scott doing any more work on the matter. (He had previously looked at such things as staff quality impact on DA and for evidence of the SS/3B interactions and SS/2B interactions.

(quoting now)
I was fooling around with DAs this morning and came across some interesting
preliminary results.

I was curious as to whether there was any correlation between CF defense
and corner OF defense. Could Ken Griffey's CF DA problems be related to
poor play by his mates in the Seattle outfield not "poaching" CF enough?

What I did was take the team OF DAs for every AL team from 1988-94 (98
teams). I then compared the CF DA to the combined DA in LF and RF for each
team. I rank ordered each.

Some of the interesting results:

1. The correlation (Pearson) between team CF DAs and team LF/RF DAs is a
somewhat disturbing .531 (Spearman's rho is .547). Furthermore, for team CF
DAs in the top and bottom quartiles, the correlation is .641 (Spearman:
651). Normalized to league average, the Pearson correlations are .573 and
617. The null hypothesis (actual correlation = 0) is easily rejected, with
effectively 100% certainty (via a t test; probit plots show normal
distributions).

2. There are NO cases in which a team's (e.g. "Toronto '92" or "Seattle
'94") CF DA is in the top quartile of performance over the period and the
corner OF DA is in the bottom quartile.

3. There are 3 cases of an upper quartile CF DA coinciding with a below
median LF/RF DA: Cleveland 1990, 7/62 (Mitch Webster!?); Toronto 1994, 18/53
(Devon White); Baltimore 1988, 21/74 (Ken Gerhart, et al.).

4. Conversely, there are only 5 cases of a lower quartile CF DA coinciding
with an above average LF/RF DA, including 3 cases of a lower quartile CF DA
coinciding with an upper quartile LF/RF DA. One of those cases is the Detroit
CF circus last year, with such comical selections as Kirk Gibson and Juan
Samuel playing there.

5. Seattle's CF DAs (primarily Ken Griffey, of course) in recent years have
been dreadful, including the worst performance of the period (1994).
However, their LF/RF DAs have *also* been uniformly dreadful. One of the
two times Griffey had a decent DA ('89), Seattle's LF/RF DA was also middle
of the pack -- much better than usual. The other time, Seattle's LF/RF DA
was as horrible as usual (remember, 98 teams):


I think it's fair to ask if the wildly spinning turnstile in LF combined with
the lumbering Buhner in RF makes Griffey's task in CF impossible. In the last
five years, the corner outfielders in Seattle haven't just been bad, they've
been brutal, never getting out of the bottom quartile.

6. Kenny Lofton's precipitous decline in DA last year coincided with a
large decline in the Indians' LF/RF DA, which dropped from good to mediocre
(25->51).

7. The year (1992) Willie Wilson had his fabulous DA (.641), the Oakland
corner outfielders had the second highest DA ever (.596).

Could Griffey's problems be caused by his mates in the Seattle outfield?
Could Seattle's other outfielders be suffering from *Griffey's* inability
to poach (STATS's Scoreboard showed that Griffey poaches less often than
many other outfielders)? Or maybe it really *is* the pitching staff's
fault?

However, my preliminary results appear to seriously call into question the
independence of OF DAs, and therefore their reliability as an estimate of a
player's defensive performance.