Syahmi
Muhammad Syahmi Nor Mokles is a PhD candidate in Sports Science at La Trobe University, Melbourne. His research examines team cohesion, player interaction, and performance in team sports using social network analysis. His current work focuses on constructing and analysing player interaction networks from National Basketball League play-by-play data to understand how patterns of collaboration relate to team performance. His broader research interests include social network analysis, team dynamics, and computational approaches to sports performance.
Session
Basketball depends on players working together. Statistics such as points, assists, shooting efficiency, rebounds, and turnovers tell us what players and teams produced, but they do not show who worked with whom. Two teams can record the same number of assists through different patterns of interaction. One may repeatedly use a few player combinations, while another spreads assisted scoring across many pairs. Examining these relationships may help us understand aspects of team play that aggregate statistics leave out.
This study asks how assist-scorer interactions are organised within professional basketball teams and how these patterns relate to game performance. It also examines whether the findings change when assisted scoring is measured as a total count or relative to the points a team scores. This exploratory study reports initial findings from the 2025–26 Australian National Basketball League season as the first stage of a planned multi-season analysis.
Play-by-play data were used to construct 351 complete team-game networks from 10 teams across 176 matches. Each network represents one team in one game. Players involved in recorded assist-scorer interactions are the nodes. An assist from Player A leading to a made score by Player B creates a directed connection from A to B, weighted by how often that interaction occurred. These networks capture recorded assisted scoring rather than all passing or teamwork.
We measured interaction volume, directed density, reciprocity, and the number of players involved. Density describes the proportion of possible directed connections that occurred among involved players; reciprocity describes whether players assisted one another in both directions. Linear regression modelled point differential, and logistic regression modelled win/loss. Both models included the network measures, team and opponent as categorical fixed effects, and home/away status as a covariate. Match-clustered robust standard errors accounted for dependence between opposing observations from each match.
Preliminary analyses found that greater interaction volume was associated with a higher point differential and higher odds of winning. This positive association remained when all matches involving each team were excluded in turn. However, every interaction already involves a made score, so this finding alone does not establish a performance advantage associated with a particular network structure.
Structural findings were less consistent. Density was negatively associated with winning when the model included total interactions, but positively associated when the total was replaced by interactions per 100 team points. This interaction rate showed no clear association in simpler adjusted models and became negative when density and player involvement were included together. Previous-game network measures also showed no clear associations with next-game outcomes.
These findings highlight the need to distinguish how much assisted scoring occurs from how it is organised across teammates. Networks describe player-pair relationships that team totals omit, but the current analyses do not identify a consistently successful pattern of play. Following abstract submission, the analysis will be extended to additional seasons to assess whether the observed relationships remain consistent.