Saturday, January 31, 2015


Over at his site, Bill James is in the midst of what will be likely be a book devoted to a revamped version of his fielding method for Win Shares. Aside from Bill's valiant attempt to demystify and critique the work of those who've made overly aggressive claims about matters with the glove, it's fascinating to see how the people "on the inside" are jockeying for position. (We'll get to that in a bit.)

A shameless and entirely unrelated plug for our upcoming "International
Film Noir" series in San Francisco this March, where twelve of the fifteen
films in the series haven't been seen in the US for more than 50 years.
(Note to Bill James: resist the temptation to be a film/cultural critic.)
Bill is revising a series of models and value assessments about the various fielding positions, and there are some fascinating data perspectives that he's worked up as he attempts something like his own "big data" approach. (Not play-by-play data, but a more comprehensive agglomeration of traditional fielding data than what's been either envisioned or attempted previously.)

From looking at where he's at right now (up through first basemen), several things seem likely, and they  will be make for improvements in the earlier work. It's likely that the new method will at last rectify the ongoing modeling error that virtually every major fielding system has replicated--an exaggeration of the centerfielder's actual contribution to outs made. That should reduce the overall number of Win Shares assigned to that position. And this will help to eliminate a whole series of distortions that enter into other ranking systems (if those folks will pay attention to what's being said, that is).

It's also likely that Bill will stop short--in fact, not even mention--what is (and has been for some time) our view of what the most important missing factor in refining defensive evaluation. What's that? Simply, it's measuring how far fielders have to go from where they are positioned to get to the ball.

Now, clearly, one reason why Bill will at best only mention this in passing is that he wants to create a system that permits some kind of historical comparison, and the measurement above is simply not something that could be done without post-modern technology. It's clear that Bill is pretty much abdicating this aspect of things to the technocrats, who (unfortunately) are quite unlikely to ask the right questions about how to collect this data.

Coming soon! "Mini Tango Love Pies" with special containers that can
be used to turn any statistical argument into a blunt instrument.
If that data is collected properly, we will know much more about the effective defense-to-pitching ranges that exist but can't currently be measured. We'll know more about such concepts as "pitcher luck" because the distance to ball data will tell us much more than the overrated BABIP stat does.

But likely the biggest battle that will come up in this new discussion, and one that is already underway from the ongoing chatter (including side conversations at the gathering place where the Tango Love Pie™ continues to bake...) has to do with what the effective range between the best and worst fielders at a position is. In his current work, Bill suggests that this range is much lower than virtually everyone else in the field. That has spawned some dubious modeling exercises elsewhere that try to force-fit a link between the gap in best-to-worst and the overall modeling inference about the overall importance of fielding in run prevention.

Those models are ideological holdovers from earlier, flawed representations of data and they persist in the thinking. The flawed result is that the effective range from best to worst is accepted as existing across all teams, when that gap is mitigated by the fact that in real life, team defense is never existing at anything like the individual positional extremes. In short, if you think there's a 25-run difference at a position, you can't just add up the seven positions and claim that the effective impact range for team defense is 175 runs. You have to have temper that "greatest possible gap" to reflect that no team--even using a team-based method that builds in the assumption that bad teams have lousier fielders (an assumption that is a modeler's compromise), there's no way that even the worst team can have all of the worst fielders on the list. It would be tantamount to multiplying your replacement level value seven times and then applying it to the data set--the result would be to make the fielders look far worse than is actually the case.
Somewhere in actual "effective value range" for the
team-aggregated run prevention effects from the Defensive Spectrum.

It's clear that the answer about the "effective quality range" for teams as a whole is at least half of what the "additive approach" claims. It looks like Bill's method overcorrects a bit for this, and the early chatter suggests that this discussion will become one of the key skirmishes in the ongoing "Fielding Wars."

At any rate, it's good to see Bill focusing himself on these issues again--and it's also heartening to see our old pal Charlie Saeger, who was actually ahead of the fielding curve in the late 90s when his Context-Adjusted Defense (CAD) method (one of the proudest moments in the flamboyant history of BBBA) shifted the ground on which these discussions originate, right in the middle of the ongoing responses, providing his usual tongue-in-cheek sanity check.