Matching-LLM：Predicting Public Opinions Using Large Language Models

魏立佳，武汉大学经济与管理学院教授、博士生导师，数理经济与数理金融系主任，行为科学研究实验中心执行主任。研究领域是行为经济学、数字经济。现兼任教育部经济学“101计划”《行为与实验经济学》课程联合牵头人、China Economic Review等国际期刊的客座编辑等职务。论文发表于Marketing Science、Econometric Theory、Experimental Economics、European Economic Review、Journal of Economic Behavior & Organization、AEA: Papers and Proceedings等国际一流期刊，以及《中国工业经济》《经济学（季刊）》《系统工程理论与实践》等中文权威期刊。主持国家自然科学基金重点、面上、青年项目，教育部人文社会科学基金，教育部产学研创新基金，科技部高端专家引进项目等国家级、省部级项目。

In recent years, large language models (LLMs) have attracted attention due to their ability to generate human-like text. As surveys and opinion polls remain key tools for gauging public attitudes, there is increasing interest in assessing whether LLMs can accurately replicate human responses. This study examines the potential of LLMs, specifically ChatGPT-4o, to replicate human responses in large-scale surveys and to predict election outcomes based on demographic data. Employing data from the World Values Survey (WVS), the American National Election Studies (ANES), and the German Longitudinal Election Study (GLES), we assess the LLM’s performance in two key tasks: predicting human survey responses and both U.S. and German election results. In survey tasks, the LLM was tasked with generating synthetic responses for various socio-cultural and trust-related questions, demonstrating notable alignment with human response patterns across U.S.-China samples, though with some limitations on value-sensitive topics. In voting tasks, the LLM was mainly used to simulate voting behavior in past U.S. elections and predict the outcomes of the 2024 U.S. election and the 2025 German federal election. Our findings show that the LLM replicates cultural differences effectively, exhibits in-sample predictive validity, and provides plausible out-of-sample forecasts, suggesting potential as a cost-effective supplement for survey-based research.