Search papers, labs, and topics across Lattice.
This study introduces the Stereotypes-to-Decisions (S2D) framework to systematically evaluate regional bias in large language models (LLMs) across all 34 provincial-level regions of China. By analyzing stereotype ratings of Warmth and Competence alongside paired-choice tasks in Education, Occupation, and Social Interaction, the research uncovers significant regional disparities that correlate with economic and digital development indicators. The findings highlight the stability of these biases across different languages and the necessity for more nuanced evaluations and mitigation strategies in LLMs.
Regional bias in LLMs is not just abstract; it significantly influences social decisions and reflects real-world economic disparities.
Regional bias in large language models (LLMs) may shape both perceptions of regional groups and decisions about individuals from different regions. Yet existing studies often examine these manifestations separately, leaving their structure and consequences unclear. We introduce Stereotypes-to-Decisions (S2D), a systematic framework evaluating regional bias from abstract stereotypes to concrete social decisions. Covering all 34 provincial-level administrative regions of China, S2D evaluates six LLMs using stereotype ratings of Warmth (perceived friendliness and trustworthiness) and Competence (perceived capability and intelligence), along with paired-choice tasks across Education, Occupation, and Social Interaction. Results reveal substantial regional differences in regional scores, with considerable agreement across models, especially for Competence and Occupation decisions. Furthermore, these patterns are associated with regional economic and digital development indicators and display mixed human-like stereotypes, with some regions rated highly on one dimension but poorly on the other. They also remain largely stable across Chinese and English prompts. Overall, our findings show that regional bias in LLMs is prevalent, systematic, and consequential, motivating more regionally aware evaluation and mitigation.
