Measuring Baseball Defensive Value Using Statcast Data

Abstract

Multiple methods of measuring the defensive value of baseball players have been developed. These methods commonly rely on human batted ball charters, which inherently introduces the possibility of measurement error and lack of objectivity to these metrics. Using newly available Statcast data, we construct a new metric, SAFE 2.0, that utilizes Bayesian hierarchical logistic regression to calculate the probability that a given batted ball will be caught by a fielder. We use kernel density estimation to approximate the relative frequency of each batted ball in our data. We also incorporate the run consequence of each batted ball. Combining the catch probability, the relative frequency, and the run consequence of batted balls over a grid, we arrive at our new metric, SAFE 2.0. We apply our method to all batted balls hit to centerfield in the 2016 Major League Baseball season, and rank all centerfielders according to their relative performance for the 2016 season as measured by SAFE 2.0. We then compare these rankings to the rankings of the most commonly used measure of defensive value, Ultimate Zone Rating.