Big Five representation, Elo and FIFA World Cup knockout progression, 1998–2026: audited data and reproducibility materials

Audited research compendium supporting the evaluation of Big Five league representation as a candidate squad-composition performance indicator for FIFA World Cup knockout progression from 1998 to 2026. The release contains 1,584 first-knockout starting-eleven records nested within 144 team-tournaments across eight men’s World Cups and 48 national teams. The analytical framework evaluates three linked properties: discrimination between levels of format-harmonised knockout progression; incremental information beyond pre-tournament World Football Elo; and transportability of any increment to World Cup editions and national teams excluded from model estimation. The 1998–2010 historical extension preserves the frozen exposure, outcome and model definitions established for 2014–2026. Historical starter status derives from the Fjelstul World Cup Database and was cross-checked against RSSSF match details. Club affiliation and top-division status were adjudicated using tournament squad records and season-specific domestic-league reference files. The 2014–2026 subset retains the completed provenance audit, including 505 re-checked starter records, 13 starter-name corrections, 45 club-affiliation checks and seven Big Five reclassifications. Analyses include tournament-adjusted ordinal and cut-point models, AICc comparison with Akaike weights, 5,000 national-team cluster-bootstrap samples with bias-corrected and accelerated intervals, leave-one-edition-out validation, leave-one-national-team-out validation, a strict 2026 temporal and tournament-format holdout, and a predictive cluster bootstrap that refits both competing models within every fold and resample. Additional checks address collinearity, edition adjustment and dependence through within-edition permutation testing and an edition-level block bootstrap. Big Five representation was associated with deeper knockout progression before adjustment for Elo, but this association attenuated after adjustment. Model comparisons and held-out evaluations did not establish a stable, transportable increment beyond prior team strength. These findings concern the predictive information provided by current league representation; they do not distinguish the mechanisms through which players enter elite leagues or develop within them. Version 5.2.0 updates presentation and documentation only. Figure 1 presents the relationship between Big Five representation and Elo alongside the four primary ordinal-model estimates. Figure 2 displays the held-out difference in ignorance score between the joint Big Five + Elo model and Elo alone for each World Cup edition, together with the equal-edition mean and its 95% BCa interval. The former supplementary interval figure was removed; its underlying estimates remain available in the deposited machine-readable results. No data, frozen definition, model specification, inferential estimate, predictive score, random seed or substantive conclusion changed from v5.1.0. The estimand is knockout progression conditional on qualification for the knockout stage. These observational data do not identify a causal effect of Big Five employment, estimate the long-term developmental return to elite-league exposure or address group-stage qualification. The compendium includes analytical data, provenance records, analysis code, machine-readable results and publication-ready figures. The derivative dataset is distributed under CC BY-SA 4.0, with attribution to the Fjelstul World Cup Database.