Measuring Normative Preferences in Sustainability Policy: A Multi-Model Audit of Large Language Models — Data and Code

Dataset and analysis code for the article Measuring Normative Preferences in Sustainability Policy: A Multi-Model Audit of Large Language Models (Fonseca & Klink). The study audits seven large language models on the economic policies advanced to drive decarbonisation and ecological transition, using a 14-item binary forced-choice questionnaire under a full factorial design (institutional persona × language × temperature × presentation order × repetitions). Data were collected between 1 and 3 May 2026. The deposit contains the SQLite database with all 31,196 valid responses, the data-collection engine, the analysis pipeline (binomial tests with FDR correction, logistic regressions, and an ICC variance decomposition) and the scripts that generate the paper's figures. Update Log: v1.1 corrects the heatmap figure script (reversed row labels) and enlarges the variance-decomposition figure; data and analysis unchanged