Team Coordination Dynamics of Winning NBA Teams

Predicting sports games outcomes is an endless pursuit shared by stakeholders ranging from fans to coaches to data scientists. We have begun investigating the value of positional data recorded during basketball gameplay with the goal of predicting outcomes from team dynamics as they emerge. We approached this problem by analyzing the “shape” of team movements on the court and investigated whether team dynamics in NBA games mimicked long-range correlated patterns observed in other team contexts. We analyzed 622 NBA games from an archival data set, including all area time series obtained for each of the four quarters. We fit a linear mixed-effects model with normalized α or percent determinism, as the outcome variable, and a fixed effect of win/loss and random team effects (i.e., random intercepts). These preliminary results suggest that analyzing positional data using time series data may provide meaningful information relating to game outcomes and team coordination dynamics