Replication package for "Do Human and Artificial-Intelligence Avatar Presenters Persuade Differently? Trust and Engagement Pathways to Purchase Intention in Five European Markets"

Version 2.0.0 revises the coding of presentation order for the pilot wave (fixed human-first, n = 154), which was miscoded in version 1.0.0, and re-estimates all outputs accordingly. Version 1.0.0 should not be used. Version 2.0.1 corrects the file paths inside the Python scripts; data and outputs are unchanged. Replication package for the study "Do Human and Artificial-Intelligence Avatar Presenters Persuade Differently? Trust and Engagement Pathways to Purchase Intention in Five European Markets." A within-subjects experiment in which 630 consumers across five European markets (Portugal, Spain, France, Germany, Poland) each evaluated a human presenter and a script-identical AI avatar delivering the same product demonstration, on five constructs: presenter authority, presenter authenticity, consumer trust, consumer engagement, and purchase intention. The two presenters are demographically matched; script, product, wardrobe, setting, framing and production quality are held constant, and the avatar's voice was calibrated to the human presenter's. This deposit contains the anonymised item-level dataset, the SmartPLS 4 import files, the complete estimation output, the reproduction guide, and the reconciliation between the SmartPLS and Python pipelines. Full documentation is in README.md. CONTENTS Dataset- Dataset_Wide_630_v2.csv — the primary dataset. Anonymised item-level data, one row per participant (630 rows): PID, Country, Origin (Pilot / Batch2), Order (AvatarFirst / HumanFirst), demographics, and the 110 item responses prefixed HU_ (human) and AV_ (avatar). This single file contains all the empirical data underlying the article.- SmartPLS_WIDE_630_v2.csv — same content, SmartPLS-ready.- SmartPLS_LONG_1260_MGA_v2.csv — long format (630 participants × 2 presenter conditions) with a Presenter column.- SmartPLS_STAGE2_WIDE_subscores_v2.csv — the 14 dimension means per condition (item averages). SmartPLS 4 import files- SmartPLS_Stage1_Human_v2.csv and SmartPLS_Stage1_Avatar_v2.csv — first-order measurement, 630 rows × 55 items each- SmartPLS_Stage2_Human_v2.csv and SmartPLS_Stage2_Avatar_v2.csv — second-order and structural stage, 13 first-order latent-variable scores plus the 5 purchase-intention items- SmartPLS_Stage2_STACKED_MICOM_v2.csv — stacked file for MICOM permutation testing (groups Human / Avatar)- README_SmartPLS_Import.txt — import instructions, construct-to-item assignment, and the exact estimation settings- SmartPLS4_Reproduction_Guide.docx — step-by-step reproduction guide SmartPLS 4 outputs (complete exports)- SmartPLS_Results_Human_Algorithm_v2.xlsx and SmartPLS_Results_Avatar_Algorithm_v2.xlsx — PLS-SEM algorithm output per condition- SmartPLS_Results_Human_Bootstrap_v2.xlsx and SmartPLS_Results_Avatar_Bootstrap_v2.xlsx — bootstrapping (5,000 subsamples, BCa, two-tailed)- SmartPLS_Results_Human_PLSpredict_v2.xlsx and SmartPLS_Results_Avatar_PLSpredict_v2.xlsx — PLSpredict (10 folds × 10 repetitions)- SmartPLS_Results_MICOM_Permutation_v2.xlsx — MICOM permutation test (5,000 permutations) across conditions- Stage1_Reliability_v2.xlsx — reliability of the 14 first-order dimensions- SmartPLS_Reconciliation_v1_v2.xlsx — every estimate, version 1.0.0 vs 2.0.0 Python scripts (Python 3.10; pandas, numpy, scipy, statsmodels, pingouin, scikit-learn, shap)- plspm.py — PLS path modelling (path weighting, Mode A) reproducing SmartPLS- layer2.py — repeated-measures ANOVA, order-adjusted mixed model, TOST, by-country differences- layer345.py — Stage 2 estimation, Source Credibility second-order model, paired bootstrap of paths, indirect effects and between-condition differences, paths by country- layer5_shap.py — gradient boosting and SHAP importance- verify_pilot_order.py — checks the pilot-wave presentation order against the raw survey export (the raw export is not deposited because it contains panel identifiers) Stimuli- The twelve stimulus videos (six human, six avatar, one per language) are hosted on YouTube as unlisted clips; the URLs are listed in Appendix A of the article. NOTE ON THE PAIRED DESIGN Each participant contributes both a human and an avatar observation, so the human-versus-avatar comparison must preserve participant-level pairing. SmartPLS's native PLS-MGA should not be used for this contrast, as it treats the two conditions as independent groups. The equivalence tests (TOST), the specific indirect effects and the between-condition path differences were estimated outside SmartPLS from the long-format dataset, keyed on PID; plspm.py reproduces the SmartPLS point estimates and layer345.py performs the paired bootstrap; SmartPLS_Reconciliation_v1_v2.xlsx aligns every estimate across versions. ETHICS AND RIGHTS Participants were recruited through Prolific Academic and gave informed consent under a GDPR-compliant procedure; all were debriefed. The dataset carries no recruitment identifier, no IP address, no free-text response and no timestamp; PID is an arbitrary sequence number and age is released as a two-level band. The stimulus videos are not redistributed in this deposit and are not covered by the CC BY 4.0 licence. The human presenter's likeness and voice were cleared by written consent for use as research stimuli only. The videos remain hosted on YouTube as unlisted clips, reachable by anyone holding the direct URL without a login, and are listed in Appendix A of the article. RELATION TO OTHER DEPOSITS This deposit is unrelated to Zenodo record 10.5281/zenodo.20381330, which is the replication package for a separate, earlier study by the same authors (a pilot of n = 145 plus a synthetic sample of N = 500). The two datasets are distinct and must not be treated as versions of one another. FUNDING Funded by national funds through FCT — Fundação para a Ciência e a Tecnologia, I.P., under Programme Contract UID/05105/2025, attributed to REMIT — Research on Economics, Management and Information Technologies, Universidade Portucalense Infante D. Henrique.