Perspectives on data analytics for gaining a competitive advantage in football: harnessing data for decision support

In this issue, Olthof and Davis highlight the potential of computational methods for gaining a competitive advantage in football and call for a closer collaboration between football and data science experts to fully leverage these opportunities. While we agree in principle with both aspects, we would like to amend some considerations that may contribute nuances to this perspective.Sustained success in a competitive environment results from a stream of good, well-informed decisions. Computational methods may support decision making in football by alleviating information overload, time constraints, unintended variation, and human biases. The advantages for data management and automated feature extraction are beyond doubt. However, as also emphasized by Olthof and Davis, the critical part of decision making is a prediction task: Forecasting the potential outcome of the available options and choosing between options based on limited amounts of information. Over the last two decades, the use of large amounts of data for the construction of sophisticated metrics and predictive models has gained widespread use in elite football. However, high-dimensional, data-driven algorithms don't necessarily provide the most accurate and helpful predictions. Rather, deliberately sparse, interpretable models that leverage data-driven modelling as well as domain expertise have repeatedly shown to have competitive predictive performance while at the same time avoiding the downsides of black-box algorithms for decision support. We illustrate this 'less-can-be-more' effect with two worked examples based on real-world data. Finally, predictability of an outcome can be low even in principle, putting hard limits to predictive accuracy regardless of modelling strategy.