Gaze on IPCC climate maps carries behaviour-specific signals for intended low-carbon action

Résumé fourni par la source

De-identified gaze data and the complete analysis code for the manuscript "Gaze on IPCC climate maps carries behaviour-specific signals for intended low-carbon action". 381 participants at three sites viewed IPCC climate maps while their gaze was recorded with a head-mounted eye tracker, and completed questionnaires on low-carbon behaviour, climate anxiety and audience segmentation. The deposit contains one row per fixation for every participant, the per-stimulus gaze and saliency measures, the participant-level feature table that is the exact input to the prediction pipeline, the cached per-fold cross-validation results behind every reported number, a data dictionary, and the full wording of all 23 questionnaire items in Spanish and English. Start with notebooks/Reproduce_paper_results.ipynb. It reproduces the cohort, the eye-tracking gain for all six models and all three outcomes, the permutation null and the posterior grid, from the files in this deposit alone, in under a minute, with no raw recordings and no network access. It ships with its outputs already filled in, and it is executed as part of building the archive, so an archive that cannot reproduce its own numbers never gets packaged. To run it: unzip climate-et-ml-manuscript-v1.0.0.zip, then from inside the unpacked folder, pip install -r requirements.txt and open notebooks/Reproduce_paper_results.ipynb. The source questionnaire workbook is not deposited: it records self-reported nationality and free text alongside age and gender. Every item any reported analysis uses is in the deposited feature table under coded participant IDs. Pupil Cloud recording identifiers are removed and wall-clock timestamps are replaced by milliseconds from the start of each participant's own viewing, so every within-session interval is preserved. The figure rasters are published with the article and are not duplicated here. README.md is attached separately and lists every path. Code is MIT; data are CC BY 4.0.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.