Using Bayesian hierarchical modeling to decompose a team's observed defensive behavior into three independent components: (1) baseline tactical tendency (what the team does on average against an average opponent), (2) home/away adjustment (how they modify their approach based on venue), and (3) opponent strength effect (how much the opposing team's quality forces a change). This separates a team's true tactical identity from the confounding effects of schedule and opponent quality. Without this decomposition, a team like Burnley looks like a pure low-block team — but after removing opponent strength effects, they're actually the 4th most aggressive pressing team because they push up against weaker opponents.
(1) Use the GNN model to produce per-possession-state short possession probabilities for every match over a season. (2) Fit a Bayesian hierarchical model: Y = baseline_tendency + home_away_effect + opponent_strength_effect + noise. Use the in-possession team's xGD minus the out-of-possession team's xGD as the opponent strength covariate. (3) Extract posterior distributions for each component — the Bayesian approach gives uncertainty estimates, not just point estimates. (4) Interpret: baseline tendency shows the team's true defensive philosophy; home/away effect shows tactical venue adaptation; the opponent strength coefficient shows how much the meta-game (adapting to stronger/weaker opponents) drives what we see.
Key findings:
After Bayesian decomposition removes opponent quality and home/away effects, Burnley is the 4th most aggressive pressing team. Their low-block reputation comes from playing mostly against much stronger opponents. Man City's pressing intensity is partly an artifact of their talent advantage.