How Robust Are Political Differences in Trust in Public Service Broadcasters? A Same-Data Reproduction and Threshold Analysis of BBC and NHK Surveys

Abstract

This preprint and technical supplement report a same-data reproduction and distribution/threshold analysis of BBC and NHK trust surveys, extending Pickering, Sunahara and Hansen (2025), DOI 10.1515/commun-2024-0184. The study is exploratory, not preregistered, and its model, dependence and survey-design limits are retained. It is not a claim of causal or nationally representative inference. Human contribution and accountability. Nguyen Minh Quan refined the aims and the focus on the complete distribution of trust, directed revisions and completion gates, and specified three methodological safeguards: no model selection for favourable p-values; treating wave sensitivity as assumption-based rather than as verified survey-design inference; and keeping conclusions within the evidence. His confirmed CRediT roles are Conceptualization, bounded Methodology, Project administration, and Writing – review & editing. The author has confirmed personal review and understanding of the methods, results tables and confidence intervals, the ability to explain them, and review of the approved scientific and reproducibility dossier with responsibility for its accuracy, honesty and content rights within the scope he can confirm. This does not establish that he personally wrote or ran the code, independently recalculated every result, or performed a separate line-by-line human code audit. Substantive AI use. OpenAI ChatGPT and ChatGPT Work were used substantially for source reading and checking, technical methodological implementation, Python code development and execution, statistical calculations and numerical verification, tables and figures, initial drafting, editing and internal adversarial review. The technical implementation, computation and recorded computational checks were carried out through this AI-supported workflow; they are not credited as the author's personal Software, Formal analysis, Validation or Writing – original draft. Recorded workflow labels identify GPT-6/ChatGPT where documented; exact model versions for all earlier sessions cannot be independently recovered, and a single version throughout is not asserted. The computations use the original empirical survey inputs identified by Pickering (2024), Harvard Dataverse DOI 10.7910/DVN/AUE81C, version 1.0; the recorded checks connect outputs to those inputs. The aggregate results are computed analyses of those inputs, not synthetic respondent observations or invented empirical findings. Computational checks through AI are not external human peer review or journal certification. No AI tool is an author, creator or contributor; tools are disclosed here, not entered in those metadata fields. The preprint/supplement are proposed under CC BY 4.0. Original respondent-level data and original R are not redistributed. Source citations, limitations, full CRediT and AI disclosures are also in the manuscript. This description identifies documented human engagement and technical AI execution without asserting repository acceptance or certification. Public reproduction materials. Companion software, aggregate outputs, figures and reproducibility instructions were published on 12 October 2026 as Nguyen Minh Quan (2026), BBC and NHK trust reproduction code and aggregate materials, version 1.0.0, Zenodo, https://doi.org/10.5281/zenodo.23307214; record: https://zenodo.org/records/23307214. This is the existing software DOI. It is not the DOI of this preprint. Version 1.0.1 of the preprint incorporates the actual software availability and citation; scientific results, methods, figures and conclusions are unchanged from the approved version 1.0.0.