Abstract
This study examined context-dependent behavioural variation and concurrent attacking output in professional football. Wyscout data covered the 2023/2024 EFL Championship. After a data audit, 546 matches contributed 1092 team-match observations and 2182 team-score-state observations covering at least 5 min. Principal component analysis summarised eight attacking indicators; the defensive-pressure measure combined standardised advanced regains and reversed PPDA. Duration-weighted team fixed-effects models used two-way team/match cluster-robust covariance and t-based inference. Behavioural deviations were calculated relative to leave-one-match-out team-season means. Two attacking components explained 78.7% of variance. Winning was associated with more direct play, deeper loss locations and lower pressure; losing showed the opposite pattern. Deeper loss-location deviation was associated with lower xG/90 (β = −0.139, 95% CI −0.210 to −0.068), lower shot rates and lower penalty-area entry rates. Higher pressure was associated with higher xG/90 (β = 0.133, 95% CI 0.052 to 0.214), higher shot rates and higher entry rates. The pressure–xG association depended on the composite weighting and was mainly attributable to advanced regains. With team fixed effects, deviation models offered negligible explanatory gain over models using original scores. These retrospective associations describe behavioural variation; they do not establish purposeful adaptation, causal benefits or prospective predictive value.