Statistical analysis plan for hypothesis 2 of a PhD thesis: media and Twitter 5C framing indicators of vaccination, Slovenia 2021 (frozen version 1.0)

Pre-specified statistical analysis plan for hypothesis 2 of the PhD thesis of Mario Fafangel (University of Ljubljana, Faculty of Medicine). Hypothesis 2 states that the association between vaccination-framing indicators from a model trained on mainstream media and those from a model trained on Twitter data is high. The plan tests it as a proof of concept on the event window of the article deposited under 10.5281/zenodo.22912892 (2 August to 31 December 2021). It compares the weekly share of COVID-19 items with undermining Confidence framing (5C framework) between two series: Slovenian mainstream media, measured with the adapted open-weight extractor of that article (model M); Slovenian-language posts from a Twitter stream collected by the author in 2020–2023, measured with a classifier trained on Twitter posts outside the test window (model T). The plan also specifies the validation against human coding, the verdict rule, and the sensitivity and secondary analyses. Pre-specification. The plan was frozen on 19 October 2026, before any model output on the Twitter data existed; section 0.1 of the plan lists everything that had been seen. The files are identical to the private project repository at git tag sap-h2-freeze-v1, commit 2d425ea692c711cd03de99edbe305a1a10cb7d4b (19 October 2026, 11:07 CEST). SHA-256 of the frozen plan (thesis/h2_twitter_analysis_plan.md): 5605d6348a97fafbb50d79a973458899a828af4a2791abb5b5ae5ee8444e9828. The manifest file gives the checksum of every frozen file. The files contain no post text, identifiers or account names. File access is restricted until the thesis defence.