Supplement

This supplement supports “Using model-based evaluation to interpret variation in infectious disease forecast performance”. It is organised as the analysis runs: how forecasts and observations were sourced and who took part; how the covariates were derived; how the model was specified and fitted; the full set of estimates behind the main-text figures; and the sensitivity analyses. Every number and figure here is generated from the same fitted model as the main text, output/log/results.rds.

• Data and participation: how forecasts and observations were sourced, who took part, and how models met the study’s eligibility criteria.
• Covariate derivation: how model structure, epidemic trend and variant phase were classified and derived.
• Model specification and fitting: the causal diagram behind covariate selection, why structure has no separate main effect, and how the model was fitted and checked.
• Estimates: the full set of partial effects behind the main-text figures, on both the reported and raw scale, and the model ranking comparison.
• Sensitivity analyses: alternative scales, covariate sets and error families tried during model development.

The full supplement is also available as a single combined document: supplement.pdf.