Causal Effect Analysis of Serving Performance Using Double Machine Learning

Abstract Serving performance is widely recognized as a critical factor influencing match outcomes in professional tennis. To assess its true contribution to winning probability, this study applies Double Machine Learning (DML) to 2013–2024 ATP men’s singles match data, estimating the causal effects of four key serve-related indicators: ace rate, first serve win rate, first serve in rate, and double fault rate. The analysis identifies both Average Treatment Effects (ATE) and Conditional Average Treatment Effects (CATE). Results show that ace rate exhibits a consistent negative causal effect, suggesting potential drawbacks of ace-based strategies. First serve win rate displays strong positive effects on grass courts and among lower-ranked players, while first serve in rate has a stable positive impact, especially on clay surfaces and in mid-tier tournaments. Double fault rate effects are generally insignificant. The robustness of the estimates is confirmed through placebo tests and subsample validations. These findings provide data-driven insights for optimizing serve strategies in professional tennis under varied competitive conditions.