Modeling the probability of a batter/pitcher matchup event: A Bayesian approach

Area of Science:

• Sports Analytics
• Statistical Modeling
• Probability Theory

Background:

• The standard log5 model offers a simple approach to predicting batter/pitcher matchups but lacks flexibility.
• Generalized log5 models enhance flexibility by estimating coefficients from data, but often require extensive historical data, which is rarely available.
• Existing models struggle with data scarcity for specific player matchups.

Purpose of the Study:

• To develop a novel Bayesian hierarchical log5 model that accurately predicts batter/pitcher matchup probabilities.
• To address the limitations of existing log5 models, particularly the challenge of data scarcity.
• To improve predictive performance by integrating prior knowledge with data-driven coefficient estimation.

Main Methods:

• Developed a Bayesian hierarchical log5 model, extending the traditional log5 approach.
• Utilized fixed coefficients from the standard log5 model as prior information for estimating unknown coefficients.
• Incorporated a variable for pitcher's team defensive ability in an extended model version.

Main Results:

• The proposed Bayesian hierarchical log5 model demonstrated superior predictive performance compared to standard and generalized log5 models.
• The model effectively estimates matchup probabilities even with limited historical player data.
• Including defensive ability further enhanced the predictive accuracy of the extended model.

Conclusions:

• The Bayesian hierarchical log5 model offers a robust and flexible solution for predicting baseball matchup probabilities.
• This approach effectively leverages prior knowledge to overcome data limitations in statistical modeling.
• The model provides a valuable tool for enhancing predictive accuracy in sports analytics.