Add new comment

He started WS with team run data. Modern saberists like starting with all of the individual plays because they (rightfully) point out that runs include luck factors. The Mariners, in 2009, for example, scored about 50 runs fewer than their batting line says they should have. A deep analysis of the number of plate appearances with RISP that went to each hitter (given their normal production) would tell you whether that was luck or bad skill, but the point is that runs, being the outer result stat aren't perfect.
But I would counter with the note that if you ignore the reality, you can't measure the reality. I think if you start with the reality (R) and then figure out how to apportion it using the play-specific data...your answers will be way more stable than UZR. I've shown this with PCA already. UZR fielding data is ALL OVER THE PLACE...PCA is, by comparison, like an elite marksman shooting at a target. The target might not be exactly the truth, but at least the reading is relatively stable.