Evaluating Three-Match Change Scores as Performance Measures: Reproducibility Materials for Cross-Tournament Consistency, Criterion-Related Evidence, and Context Sensitivity in International Football

Version 2.0.0 provides the complete reproducibility materials supporting the study “Evaluating Three-Match Change Scores as Performance Measures: Cross-Tournament Consistency, Criterion-Related Evidence, and Context Sensitivity in International Football”. The archive contains derived team-match and team-tournament datasets, dated pre-tournament World Football Elo ratings, data-processing and feature-engineering scripts, leave-one-tournament-out modelling, group-cluster bootstrap analyses, cross-tournament consistency estimates, indicator-specific analyses, mathematical-coupling analyses, matchday-3 competitive-state analyses, probabilistic scoring metrics, machine-readable analytical outputs, figure-generation code, software requirements, and integrity hashes. The sample comprises six major international men’s football tournaments across three confederations, representing 152 team-tournaments, 456 team-match observations, and 38 competition groups. The study evaluates the measurement properties of short-sequence change scores derived from expected goals, possession, passes per defensive action (PPDA), and duel success. The analyses distinguish opening-match level, early change from match 1 to match 2, three-match endpoint slope, and three-match variability. Cross-tournament relative consistency is estimated from teams observed across multiple tournaments. Criterion-related evidence is evaluated using fully held-out tournament validation, with discrimination and probabilistic scoring assessed separately. The study also examines the mathematical dependence of endpoint slopes on their constituent observations and the sensitivity of late change to competitive state. Endpoint slopes showed substantial mathematical coupling with their opening and final values and reduced held-out ranking discrimination when added to opening-match indicators across all six external tournament folds. Cross-tournament consistency estimates were generally imprecise and compatible with a broad range of values. Early change produced more heterogeneous and indicator-specific evidence, including favourable probabilistic-scoring estimates for expected goals in some analyses. These findings indicate that the measurement properties of short-sequence temporal scores depend on the indicator, temporal representation, comparator, criterion, and intended use. The results should not be interpreted as evidence that football teams cannot adapt. Rather, they show that short-sequence aggregate change scores require explicit measurement evaluation before being interpreted as stable performance characteristics or labelled as adaptation, momentum, growth, or development. Raw third-party Elo snapshot files are not redistributed. The archive provides the dated analytical Elo values, source information, retrieval documentation, and reproduction code required for the reported analyses. StatsBomb and World Football Elo materials remain subject to their original terms. The CC BY 4.0 licence applies to the authors’ original code, derived analytical materials, documentation, and figures.