Add new comment

As usual, Sandy, a thought provoking reply.
I agree completely with your point #1, obviously DNRA was engineered with that paradigm in mind.
Your point #2 I agree with as well in the absolute sense. You're correct that Ks and BBs may not be the most valuable parts of a pitchers game (in the absolute run reference frame) despite the larger predictability they possess...afterall, the vast bulk of a pitcher's DNRA comes from the hits he allows, not the walks. I would caution you, however, that the variation in hit rates among pitchers is significantly smaller than the variation in K and BB rates...McCracken's studies revealed how using only K, BB, HR and the average fielding assumption, he could essentially completely reproduce the typical range of ERAs for a league. In other words, BABIP is a very minor contributor to the spread in pitching performance.
Numerically speaking, your point #3 begins with a position that the data doesn't currently support. Even if we assume that HR, BB and K are 100% pitching and everything else is 100% fielding, we get that pitchers are contributing about 53% of the linear weight runs created and about 57% of the variability in team LW Runs Allowed. That's the baseline you have to start with...after which point any influence pithcers do have on things like baserunning, BABIP and GB/FB pushes pitching control slightly higher. Leaving me pretty happy with my Pitching/Fielding split of 59-61 for the pitchers and 39-41 for the fielders. Now that was not the case as recently as the 1950s when K and HR rates were far lower, but in the modern game, I don't see how pitching could be responsible for less than 57% of the defensive game.
On the rest of your point #3, the standard deviation of PCA defensive wins created per 162 games is about as large as the mean win scoring rate for most positions, for example, center fielders average 2.52 wins per 162 with a standard deviation of 1.6 wins/162. But before you conclude that fielding stats are more useless than other stats, think again. The standard deviation for offensive wins created by PCA is ALSO about as large as the mean win scoring rate (4.15 +/- 3.9!). Baseball is played by humans...and humans have chaotic performance tendencies in everything...not just sports. If you stare at the defensive win cards produced by PCA, as I have done, you come to appreciate how well even a very basic uberstat model like PCA does at separating good fielders from bad ones and picking out the bad seasons sensibly (almost always, when I see a really low win figure produced by a good fielder, it comes with some story about how that player was injured that year or had legal troubles or also struggled offensively or some other obviously red flag that explains it). I do not believe that a top-down defensive analysis that starts with team data and divides credit among the players will be prone to the kinds of wild fluctuations that you seem to think are endemic to fielding stats...at least...no more so than offensive stats.
On your point #4, I think we're atrociously bad at forecasting defensive performance solely because we're not using team analysis. All of today's modern sabermetricians have taken the bait on one of the worst mistakes in the history of the analysis of this game...that being the decision to believe that zone based fielding analysis works without a team context. It doesn't. The high variability we observe in UZR comes partially from the system's failure to place players in their proper context.
As to your stated belief that fielding performance is led mostly by motivation and coordination between fielders and pitchers, that's possible, but it's not merely "a few defensive whizzes" who look good no matter where they play. It turns out that a top-down defensive analysis finds the very good players just as well as it finds the great ones. If you start at the top of any position's all-time defensive win list and work your way down, the top 15 will be very logical (the better a player is, the more he stands out)...but even once you get into the merely good gloves, you're still going to be nodding your head more than scratching it. I could bring up countless hundreds of fielding profiles for merely-good players, and you're not going to see any more inherent wobble in their win scoring rates than you do for the top fielders.