Personnel-adjustment for home run park effects in Major League Baseball

TL;DR

This paper develops a statistical model to accurately estimate and compare home run frequencies across different baseball ballparks by adjusting for player personnel and handedness effects, revealing more accurate park rankings.

Contribution

It introduces a generalized linear mixed effects model that accounts for personnel and handedness asymmetries to better estimate park effects on home runs.

Findings

Adjusted home run frequencies differ significantly from raw data.

Ballpark rankings change when accounting for personnel effects.

Model provides more accurate comparisons of park effects.

Abstract

In Major League Baseball, every ballpark is different, with different dimensions and climates. These differences make some ballparks more conducive to hitting home runs than others. Several factors conspire to make estimation of these differences challenging. Home runs are relatively rare, occurring in roughly 3\% of plate appearances. The quality of personnel and the frequency of batter-pitcher handedness combinations that appear in the thirty ballparks vary considerably. Because of asymmetries, effects due to ballpark can depend strongly on hitter handedness. We consider generalized linear mixed effects models based on the Poisson distribution for home runs. We use as our observational unit the combination of game and handedness-matchup. Our model allows for four theoretical mean home run frequency functions for each ballpark. We control for variation in personnel across games by constructing ``elsewhere'' measures of batter ability to hit home runs and pitcher tendency to give them up, using data from parks other than the one in which the response is observed. We analyze 13 seasons of data and find that the estimated home run frequencies adjusted to average personnel are substantially different from observed home run frequencies, leading to considerably different ballpark rankings than often appear in the media.