Data and code for "Hidden Biases: Textbook Wording and Data Contamination in Behavioral Economics Experiments with Large Language Models"

English responses, analysis code and output tables for a study presented at INFTEC 2026 (Nicosia, 22-24 October 2026). Four open-weight language models (GPT-OSS-120B, GPT-OSS-20B, Qwen3.8-27B and ALLaM-2-7B, Groq API) answered six classic behavioral economics experiments (risky-choice framing, mental accounting, the conjunction fallacy, anchoring, mixed-gamble acceptance and the ultimatum game), each in its published wording and in a disguised version with the same decision structure. The package contains 3,520 responses collected on 18 and 19 September 2026, the Python code that computes bias indices, stratified bootstrap confidence intervals, tests and pooled regressions, and the resulting tables and figures. Running the code on the included data reproduces the results of the paper. Only the English part of a bilingual data collection is included.