Dissecting Poisson based prediction models in association football: A comprehensive look at methodology, assumptions, and accuracy using data from the main European Leagues (2011 – 2022)
MetadataShow full item record
- Master Thesis 
As the access to broader and better data increases, data analytics, statistical modeling, and data science generally find ever-growing interest in sports analytics, including association football. It is no secret that both clubs and even higher governing bodies in the sport implement data-driven strategies to give them insights and a competitive advantage in play. Recognizing the importance of the sport as a fan and from the point of view of an analyst, this work seeks to contribute to the current body of literature by offering a thorough investigation of one of the most elegant approaches to sports analytics in association football; The Poisson goal model. Based on the simple and intuitive idea that goals in football are rare discrete events that follow the Poisson distribution while conditional on team performance, the concept has been appealing to many researchers. At the same time, a simplistic idea at its core, its application to realworld data, has been met with much discussion regarding underlying assumptions and methodology. Much of the discussion in the last 40 years since the idea was formalized concerns addressing assumptions such as the applicability of the Poisson distribution, score interdependence, overdispersion, and parameter stability. In the present work, we take a step back and reexamine the idea, methodology, and assumptions in the light of the most recent data from Europe’s major leagues. Furthermore, we examine sone novel concept such as considering xG (expected goals). Overall, some changing dynamics are revealed and some of the propositions made for the model do not hold given the recent developments in the sport.