A number of Yankee fans were upset last night and this morning at home plate umpire Gerry Davis’ perceived squeezing of C. C. Sabathia’s strike zone in the early innings and a few generous calls given to Jose Valverde.
Some of the calls did seem puzzling. Before calling for Davis’ head on a platter, however, you may wish to read Colin Wyers’ Baseball Prospectus piece from last October about the dangers of relying too much on Pitch f/x data (e.g., TBS PitchTrax), let alone the human eye.
Now, we’ve already talked about the problems with using cameras to track pitches — but assuming you have good data on where the cameras were located relative to the field and the optical qualities of the lenses being used, using some advanced math you can get to a pretty accurate reconing of the pitched ball. Sportvision, makers of PITCHf/x, report that a properly calibrated PITCHf/x system is accurate to within a half an inch at the front of home plate.
But is the data we’re seeing coming from a properly calibrated system? Remember — the accuracy of the system is based on having accurate data on the location of the cameras and the properties of the lenses. In order to do this, PITCHf/x operators calibrate the cameras based on a set of landmarks placed upon the field before the start of the game. The key here is “before the start of the game.”
The game in question, for instance, had an announced attendance of 42,035 — that’s a tickets sold count, not a turnstile count, but for a playoff game I imagine they’re pretty close. The average adult weighs something like 175 pounds (at least, that’s what they claim). What this means is that between the time the PITCHf/x system was calibrated and the time the first pitch was thrown, roughly 3,680 tons of baseball fan was introduced into Target Field, not counting seat cushions, signs and rally towels. And each one of those baseball fans was somewhere around 98.6 degrees Farenheit and radiating heat. All that weight and heat causes the stadium to actually move, and the cameras move with it.
In this case, I asked Mike Fast to look at the data and come up with an estimate of how far out of alignment the cameras at Target Field might have been that night. Estimating the error is difficult — in smaller samples, you have to contend with noise; in larger samples, you may miss changes that happen out of time.
PITCHf/x reported the pitch at .67 feet away from the center of home plate as it crossed the front of the plate; according to Mike’s corrections, it was probably .72 feet away, with a margin of error of .06 feet (accounting for the random error in pitch location measurement, plus the estimated error in the correction.) The edge of the zone (in other words, the edge of the plate) extends to .71 feet from the center of the plate in either direction. Now, PITCHf/x is giving us the center of the ball — if any part of the ball catches the plate, it’s a strike. So the effective zone extends to roughly .83 feet from the center of the plate.
So we think that pitch was probably a strike, given what we’ve seen with PITCHf/x — but we’re not entirely certain. (Remember, standard error means a 68 percent chance outcomes occur within the MOE.) It is, essentially, a borderline pitch.
He adds that judging the location of a pitch vertically is even trickier, since no two batters are perfectly identical in height.
This is why it is expressly unhelpful to go to your favorite PITCHf/x website, pull up a scatterplot from a single game, and use it as evidence the umpire did a bad job. The responsible thing to do (and this is what MLB does when using PITCHf/x to grade umpires) is to correct for these calibration errors and to look at a larger sample of data.
This is also one reason it’s infeasible right now to use PITCHf/x to call balls and strikes in a live game. (There are others — timeliness is one, of course. Another is operator error — game scorers are just as human as umpires, and they sometimes make mistakes in associating the PITCHf/x data with the right pitch in the game, for instance.)