Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 27: Sports Studies: Football/Soccer
SOE 27.2: Vortrag
Freitag, 22. März 2024, 12:30–12:45, MA 001
"Expected Goals" and other KPIs to characterize team performance in soccer matches: how to quantify their quality? — •Andreas Heuer1 and Fabian Wunderlich2 — 1Institut für Physikalische Chemie, Universität Münster, 48149 Münster — 2Institut für Trainingswissenschaft und Sportinformatik, Deutsche Sporthochschule Köln, 50933 Köln
A variety of so-called Key Performance Indicators (KPIs) are available to characterize the performance of teams in soccer matches. A relatively novel and very popular KPI is expected goals (xG), which is derived by weighting each shot with an empirical probability of scoring a goal from that position on the pitch. How informative are KPIs to estimate the team strength and predict future results?
This question is analysed within an appropriate statistical framework aiming to answer two questions: (i) How well does the estimation process work when the statistical noise due to finite information is absent? The associated score directly expresses how well the chosen KPI reflects the underlying team strength. (ii) How much is the estimation process affected by statistical noise? Both pieces of information can be used to construct a normalized score that is a direct measure of the overall forecast quality of a KPI.
This general formalism is applied to the five biggest European leagues for a variety of KPIs. From this analysis, the quality of xG compared to other KPIs as well as possible differences across leagues can be clearly quantified. Implications for the prediction of individual soccer match results are discussed.
Keywords: key performance indicators; sports statistics; soccer; Baysian inference; regression dilution