[3G1-OS-14a] OS-14

10:00 AM - 10:15 AM JST(1:00 AM - 1:15 AM UTC)

[3G1-OS-14a-05] Evaluation of Catcher’s Pitching Strategies in Baseball Using Causal Inference with Pitch Sequences within the Plate Appearance

〇Takimi Miura¹, Keisuke Fujii^1,2 (1. Nagoya University, 2. RIKEN)

Keywords:

Sports,Time series data analysis,Causal inference

Pitching strategies in baseball refer to the decision-making process by which pitchers select pitch types and locations to retire batters. However, evaluating pitching strategies requires controlling for player- and game-specific factors while accounting for the sequential nature of pitches within a plate appearance. Methods for quantitatively evaluating their causal effects are limited. This study proposes combining G-computation, a causal inference framework capable of modeling pitch sequences, with a Transformer-based model to evaluate pitching strategies. Using Major League Baseball game data, we modeled the generative process of catcher-specific pitching strategies, pitch characteristics, and plate appearance outcomes. We quantified plate appearance outcomes based on expected runs and estimated counterfactual outcomes for each catcher via Monte Carlo simulation. Then, we estimated the average causal effects between catchers. Through a case study focusing on two catchers, we discuss considerations for interpreting the estimates and their potential integration with WAR. This study provides an analytical framework for multifaceted catcher performance evaluation.