Dynamic Duos: Investigating the Composition of Powerful Pairs in Basketball with Network Analysis

With the increase of analytics in basketball, research has started to focus on team chemistry via novel player roles that dynamically emerge within teams. According to (Fewell et al., 2012), basketball teams can be represented as networks and explored to find relationships between the individual players and team chemistry. In addition, related research (Hedquist, 2022) also challenges the ability of traditional player roles (such as point guards, shooting guards, etc.) to capture the essence of the roles of players in a team. As a result, traditional views of team composition are limited and don't provide enough insight to managers when optimizing for team dynamics. This paper examines the composition of high-performing duos to better capture the nuanced view of player importance from a team perspective. The methodology of this paper consisted of creating network diagrams (players as nodes, passes as edges) for 14 out of the 16 playoff teams for the 2021-2022 NBA season using NetworkX. Our data was sourced using the NBA API and SportsReference. Once the networks were created, we computed a weighted centrality measure for each player, taking into account a player's betweenness centrality (a measure of a player's impact on the flow through the team), a player's player efficiency rating (a common statistical measure for a player's performance) and a player's assist ratio (measuring their indirect contribution to the team). With these measures, we ordered the players to select the player with the highest weighted centrality and their neighbor with the next highest centrality measure. We called these pairs high-performing duos. We then used k-means clustering to identify the broad player roles predominant in these duos. Our findings show that while 50% of duos consisted of a high-value player (evaluated using Value Over Replacement Player) and strong assist ratios as well as shooting abilities, 75% of duos were characterized by the presence of an agile support player. This demonstrates that examining individual high-performers does not provide as nuanced a view as taking into account their dynamic with other support players. Examining duos is one way to provide insight into the team dynamics that may exist within team networks. Findings from this research can be used by managers as well as analysts who are looking to better understand and estimate player contributions and importance from the perspective of team dynamics.