This study proposes a novel machine learning-based approach for assessing leadership capability by quantifying the season-level impact of head coaches in the National Basketball Association (NBA). Harnessing 24 seasons of NBA data (1999-2023), we estimate each team's theoretical win probability for every game using only the prior season's player statistics, deliberately excluding coaching effects. The discrepancy between these predictions and actual outcomes is interpreted as the coach's marginal contribution. To validate the robustness of this framework, we applied multiple machine learning algorithms, with LightGBM achieving the highest prediction accuracy at 68.50%. Although the improvement over the baseline accuracy is modest (1.25%), this finding carries nontrivial implications in professional sports, where small performance margins can yield substantial competitive and economic benefits. In contrast to traditional win-loss or tenure-based metrics, our method establishes a performance-adjusted baseline for leadership evaluation applicable to both sports and non-sports contexts. Furthermore, the study advances leadership assessment by providing benchmarks that transcend conventional win-rate metrics, thereby offering scalable, data-driven tools to measure managerial effectiveness in high-performance settings. Overall, this framework contributes to both sports analytics and organizational leadership by furnishing an interpretable model for evaluating leadership capability.
© Copyright 2025 Journal of Sports Analytics. IOS Press. Alle Rechte vorbehalten.