Substitution timing and within-match changes in team performance in professional soccer: a passing network analysis of the 2023–2024 La Liga season

Substitutions are the most direct tactical intervention in a soccer match, and the five-substitution rule has widened the timing decisions. This study compared within-match performance changes across substitution timings from technical-tactical, physical (event-derived proxies), passing-network, and spatial perspectives, and examined moderation by match status, position, and strategy type. All 380 matches of the 2023–2024 La Liga season yielded 2,077 substitution batches, categorised as strategic (45–60 min, \(N = 445\) ), regular (61–75 min, \(N = 829\) ), or late (>75 min, \(N = 803\) ). In a within-team pre-post design, net changes in 26 indicators were computed between the 15 minutes before and after each batch. Group differences were assessed with Kruskal-Wallis tests under Benjamini-Hochberg correction, and linear mixed models with cluster-bootstrap intervals tested the timing \(×\) match-status interaction. Sensitivity analyses addressed window overlap, incoming position, in-window red cards and goals, and halftime versus early second-half substitutions. Late substitutions were followed by a fall in pass completion (−5.9%, strategic versus late d = 0.47), a decline in passing-network clustering (−18.4%, d = 0.41), and contraction of convex hull area (−4.7%) and team width (−3.2%, d = 0.45). Strategic substitutions left them unchanged or slightly improved. Pitch-passing-network changes exceeded player-network changes (zonal density −23.4%, d = 0.71). Expected goals and net On-Ball Value rose after strategic substitutions (+16.4% and +22.7%), but the standardised effects were trivial (d = 0.10 and 0.11). Late substitutions were followed by faster forward progression (+6.8%, d = −0.35), and the timing associations held across sensitivity analyses. A leading status was associated with declines in eight attacking, circulation, and cohesion indicators but no change in connectivity or spatial deployment, and six timing \(×\) status interactions were robust. Position and strategy type mattered less than timing, and halftime substitutions were followed by more favourable changes than early second-half ones (d up to 0.59). Earlier substitutions were accompanied by stable or slightly improved collective structure, whereas later substitutions were followed by lower pass completion, looser passing networks, and spatial contraction, with only trivial differences in attacking output. These associations may inform in-match substitution-timing decisions.