Is Football/Soccer Purely Stochastic, Made Out of Luck, or Maybe Predictable? How Does Bayesian Reasoning Assess Sports?

Abstract

Predicting the game score is a well-explored duty, using mathematical/statistical models. Nonetheless, by adopting a Bayesian methodology, this study aimed to estimate probabilistically the Chilean Premier League teams’ position, considering them a hierarchical structure. This approach enabled the evaluation of the main Chilean championship that provides the major soccer players for the national team. Thus, a countable (Poisson) regression structure was considered to explain each match as a combination of home advantage, added to the power of attack and defense of each team and considering their performance in the championship as an independent game. We were able to quantify the relationship across the defense and attack of each team and, in addition, were able to group/verify the performance of the entirety of the 2020 Chilean Premier League. For the model validation, we saved the last five games for the model prediction and we found that, in this league, the teams presented a statistical significance in the attack factors, which influences the scores (goals); however, all the teams showed low defense power and we have also found that playing at home or away did not present a game advantage. Our model was able to predict the Chilean league position table, with precision on the top five positions, and from the 6–11 positions there was a small shift (close performance in the championship) caused by the similarity of the expected number of goals, which implied the same position on the rank. This type of model has been shown to be very competitive for the soccer championship prediction.