Rob Neyer attended the PITCHf/x Summit, and talks about some ideas for measuring fielding with Hit Tracker:
As Greg pointed out, all of our current “new” defensive metrics are “zone-based”; that is, they begin by separating the field into distinct zones, noting in which zone a play has been made, and then apportioning credit (or not) when a fielder makes a play (or doesn’t) in that zone. I’m simplifying, of course, but essentially every reputable system now in use shares two significant defects: the “zones” don’t fit neatly into today’s highly variable outfield dimensions, and the systems don’t have any way to account for the fielder’s starting position.
What’s more, Greg argued that the great majority of plays made by outfielders are essentially irrelevant. If an outfielder doesn’t have to move to catch a line drive, that play still counts in his favor. If he can’t catch that same fly ball because he was pulled way in with the potential winning run on third base in the ninth inning, that play still counts against him.
What Greg proposes is stripping out the plays that any outfielder could make, and the plays that perhaps no outfielder could make, and look instead at only the plays that might be made. And this might be done given FIELDf/x’s ability to record an outfielder’s starting position, the hang time of the batted ball, and the baseball’s landing point.
No, nothing needs to be stripped out. The starting position of the fielder, the hang time and the landing position of the ball become part of the probabilistic model. It’s going to be very interesting to see how the two systems measure fielding.
Note that one isn’t necessarily better than the other. As I’ve noted before, my Probabilistic Model of Range doesn’t measure just range, but the ability of a fielder to position himself well. From the description above, True Range would measure the ability of a fielder to move to a batted ball, his actual range. If the fielder doesn’t position himself well, there should be a penalty for that also, and the traditional zone based system catches those.
If we see a player who does well in a zone system but poorly in a range system, we might conclude he (or his coaches) position him well.
David, I agree that your “Probabilistic Model of Range doesn’t measure just range, but the ability of a fielder to position himself well,” and is a valuable defensive measurement, but how would you separate the fielder’s positioning of himself and the positioning called for a manager or the situation? It’s problematic to penalize an outfielder for letting a ball fall in front of him if he is in a no doubles defense, or allowing a ball over his head in a situation like the one in Rob’s post. I’m unsure how many such chances a player would have in a season but the worry would be that it could skew the data, especially if a player happened to have lots of these chances.
@Andrew: That, of course, is a big problem. That’s why I like to measure the team’s defense at a position as a whole, to see if one player is very different from the rest of his teammates. If the subs are good but the starter is bad, it seems less likely that the manager is moving him to the wrong position.
Your measure is going to be the better one for looking at a large collection of data, esp. if a player moves from team to team, stadium to stadium. The advantage Rybarczyk’s has is that it cleans out some noise, which means it might give better information with smaller data sets. The disadvantage, of course, is that it washes out some important information with the noise.