Underlying Performance Indicators in Football: Quantifying the Quality of Chances in Football

Abstract

The ability to evaluate performance accurately and objectively has profound impact on the success of organizations, as it impacts effective decision-making outcomes. In football, decisionmaking is often influenced by recent results, which can result in outcome bias. However, using results to evaluate performance has been found insufficient due to the inherent nature of football results being influenced by randomness and luck, thus calls for a more granular and objective underlying performance indicator. This research demonstrates how such a performance indicator can be modelled by applying the concept of expected goals (xG), which in effect quantifies the quality of chances in football. Through statistical validations, it was found that xG outperforms many of the traditional performance metrics, in terms of objectivity and accuracy in evaluating the quality of chances produced. Furthermore, it was found that an XGBoost machine learning technique yielded the best performing model, with an AUC-ROC score of 0.79. However, it is argued that additional data volume, variables, and data mining iterations are necessary to fully exploit the potential that lies in the predictability of the model regarding future results. The value of an xG-model depends on its predictive capability and its objectivity, which are evaluated and discussed. Additionally, perspectives are made to its potential for reducing outcome biases, which it is hypothesized will lead to more effective strategic decision-making in football, however, needs validation in future studies