An Empirical Assessment of MLB Park Factors

Abstract

This paper empirically tests the accuracy of Major League Baseball park factors. It does so via three studies, the samples for which consist of the plate appearances for all AL and NL hitters over the 1998-2024 seasons. The first study evaluates how accurately park factors predict batters' weighted on-base average (wOBA) after hitters change teams. The second uses park factors to forecast how player road wOBA performances diverge from home performances. And the third uses quasi-experimental, propensity-matching methods to compare batter wOBA performances across diversely configured parks. The studies uniformly find that park factors (including the MLB's own Statcast ones) substantially overestimate the impact of park differences. This inference is reinforced by a simulation that identifies in each study the maximum degree of accuracy reasonably attainable by valid park factors. The paper also examines how park-factor predictions fare (poorly) in relation to a park-factor-free best linear unbiased prediction (BLUP) measure of hitter ability derived from a multi-level regression model. The paper concludes by weighing the benefits of park-factor-informed analyses of hitter offensive value against the costs park factors impose by distorting raw performance metrics.