Bayesian Virtual Survey for Consumer Acceptance Forecasting: Computational Outputs

Bayesian Virtual Survey (BVS): Computational Outputs This deposit contains the computational outputs of a Bayesian Virtual Survey (BVS) framework for forecasting consumer acceptance of recycled-content food-contact packaging. The framework fuses Spanish census microdata (CIS Study 3391, 2023 ISSP Environment IV, n=2,254) with informative priors from 13 international peer-reviewed studies, generating 100,000 synthetic consumers via a Dirichlet-Multinomial conjugate Bayesian network with block-wise MC³ structure learning. Companion to a manuscript currently under double-anonymised peer review. Authorship, institutional affiliations, and funding information have been withheld in compliance with the journal's review policy and will be added upon acceptance. Contents 01_priors/ — literature meta-analysis: Dirichlet alpha vectors, I² heterogeneity diagnostics, pooled regression coefficients 02_structure/ — DAG learning and Bayesian fusion: empirical and fused causal graphs, conditional probability tables (CPTs) 03_synthetic_data/ — synthetic populations S1 (full Bayes) and S2 (CIS-only baseline) 04_validation/ — five-stage validation: three-level framework, per-variable diagnostics (TVD, χ², correlation), held-out Bayesian validation, baseline comparison against BN-MLE / Synthpop / CTGAN / TVAE, external holdout against Ipsos 2022 Spain 05_sensitivity/ — robustness analyses for the heterogeneity discount parameter λ and the discretisation granularity K 06_figures/ — validation figures (PDF and PNG) MANIFEST.csv — SHA256 hashes, sizes, and per-file descriptions Key results External validation against Ipsos 2022 Spain holdout: RMSE = 0.033 Three-level validation: median TVD = 0.013 (machinery check), fusion coherence > 60% across variables Baseline comparison: lower TVD and correlation MAE than CTGAN, TVAE, and independent baselines Robustness: main qualitative findings invariant to λ ∈ {0, 0.5, 1, 2} and K ∈ {3, 5, 7}