Abstract
We analyze how coaching strategies and in-game tactics can possibly affect goal difference and home-win probabilities using hand-coded Serie A match commentary (2011/12-2013/14). We analyze home-away goal difference, home-win probability, and home points using generalized linear, logit, and proportional-odds models. By coupling such models with robust and bootstrap inference, we find stable results: aggressive opening schemes are positively associated with performance, shots and goal-kicks exhibit consistent positive effects, crosses display a persistent negative association, and referee decisions-especially red cards and penalties-strongly shape outcomes. These relationships remain robust to alternative specifications, extensive subset selection, and model-averaging procedures that account for model uncertainty. Beyond confirming home advantage, we show that coaching decisions are systematically associated with match performance conditional on pre-match strength and market expectations. Our approach sheds light on the rationale of real-time coaching and introduces a scalable, text-based framework for studying decision-making under uncertainty in competitive environments.