Network analysis of kick-in possession chains in elite Australian football

Abstract

The study aim was to investigate ball movement patterns using network analysis techniques, to compare between successful and unsuccessful outcomes and teams in the Australian Football League (AFL). This analysis focused on possession chains starting from a kick-in (n = 1,720), drawn from all games played in the 2015 AFL Premiership season (18 teams, 206 games). Player interactions were quantified using four network metrics: cluster coefficient, degree centrality, network density, and entropy. Three-way ANOVA with Tukey post hoc and omega(2) effect sizes were calculated to assess whether differences existed between kick-in outcomes, ladder brackets, and match outcomes for each network metric. No significant differences were observed between ladder brackets or match outcomes for any network metric. More successful kick-in chains were characterised by lower density (omega(2) = 0.26, large effect; F(9, 1678) = 66.6, p < 0.00) and higher entropy (omega(2) = 0.17, large effect; F(9, 1678) = 39.6, p < 0.00). This suggests that chains resulting in successful kick-in outcomes exhibited lower interconnectedness, with a high number of players involved, and lower predictability in ball movement patterns. These findings have practical value for coaches and performance analysts and support further applications of network analysis in Australian football.