FlockVote: LLM-Empowered Agent-Based Modeling for Simulating U.S. Presidential Elections

Published as a conference paper at ICAIS 2025 FLOCKVOTE: LLM-EMPOWERED AGENT-BASED MODELING FOR SIMULATING U.S. PRESIDENTIAL ELECTIONS Lingfeng Zhou1, Yi Xu4, Zhenyu Wang3, Dequan Wang1,2∗ 1Shanghai Jiao Tong University 2Shanghai Innovation Institute 3Shanghai Academy of Social Sciences 4Nanjing University ABSTRACT Modeling complex human behavior, such as voter decisions in national elections, is a long-standing challenge for computational social science. Traditional agent- based models (ABMs) are limited by oversimplified rules, while large-scale statis- tical models often lack interpretability. We introduce FlockVote1, a novel frame- work that uses Large Language Models (LLMs) to build a “computational labo- ratory” of LLM agents for political simulation. Each agent is instantiated with a high-fidelity demographic profile and dynamic contextual information (e.g. can- didate policies), enabling it to perform nuanced, generative reasoning to simulate a voting decision. We deploy this framework as a testbed on the 2024 U.S. Presi- dential Election, focusing on seven key swing states. Our simulation’s macro-level results successfully replicate the real-world outcome, demonstrating the high fi- delity of our “virtual society”. The primary contribution is not only the prediction, but also the framework’s utility as an interpretable research tool. FlockVote moves beyond black-box outputs, allowing researchers to probe agent-level rationale and analyze the stability and sensitivity of LLM-driven social simulations. 1 INTRODUCTION Understanding and modeling complex social phenomena, such as election outcomes, is a crucial endeavor in political and computational social science. Traditionally, research in this domain has relied on statistical models built from polling data (Brito et al., 2021; Coletto et al., 2015; Chauhan et al., 2021) or conventional agent-based modeling (ABM) approaches (Gao et al., 2022; De Marchi & Page, 2014; Qiu & Phang, 2020; Fowler & Smirnov, 2005). However, these methods often face limitations: statistical models struggle with dynamic, real-time contexts, while traditional ABMs tend to oversimplify complex human decision-making into predefined rules or heuristic functions. This simplification hinders their ability to capture the full complexity and heterogeneity of real- world voter populations, limiting their utility as true “computational laboratories” for social science. The emergence of Large Language Models (LLMs) presents a new paradigm, offering unprece- dented capabilities to model human behavior with greater nuance and realism (Argyle et al., 2023; Zhang et al., 2024; Yu et al., 2024b; Grossmann et al., 2023; Gujral et al., 2024) (Figure 1). This work leverages this potential to propose FlockVote, an agent-based simulation framework designed as a computational laboratory for social science research. Instead of merely forecasting, FlockVote uses LLM agents to construct a high-fidelity “virtual society” for in silico experimentation. Each agent in the FlockVote framework is instantiated with a detailed voter profile, mirroring the demographic composition of U.S. states based on data from the 2023 American Community Sur- vey (ACS)2 and the 2020 U.S. Religion Census by the Association of Statisticians of American Religious Bodies (ASARB)3. These profiles include key attributes such as state, race, sex, age, oc- cupation, industry, education level, and religion. Crucially, agents are not static; they are provided ∗Corresponding author: dequanwang@sjtu.edu.cn 1https://github.com/maple-zhou/FlockVote 2https://www.census.gov/programs-surveys/acs 3https://usreligioncensus.org/ 1 arXiv:2512.05982v1 [physics.soc-ph] 27 Nov 2025 Published as a conference paper at ICAIS 2025 Statistical Models Data Factors Traditional Agent-Based Modeling Predefined Rules Constrained Agents: Only vote based on rules Flexible Agents: Vote and interact with humans Agent Profile State/Race/Sex/Age/ Occupation/Industry/ Education/Religion LLM-Empowered Agent-Based Modeling Model Data Data Donald Trump: 0.51 Kamala Harris: 0.49 … Context Info. Economy/Immi- gration/Abortion/… Vote: Donald Trump Vote: Donald Trump Reason: I prefer his policies … Figure 1: Comparison of Social Simulation Methodologies. (Left) Conventional statistical models are often “black boxes” that correlate data-driven factors with outcomes but lack causal or behav- ioral interpretability. (Center) Traditional agent-based modeling (ABM) relies on agents that follow predefined, heuristic rules. This limits their behavioral realism and ability to adapt to new contex- tual information. (Right) LLM-powered agent-based modeling, the approach used in our FlockVote framework, serves as a “computational laboratory”. It empowers autonomous agents with demo- graphic profiles and dynamic context, enabling them to simulate complex, human-like reasoning. This provides a flexible, nuanced, and interpretable simulation essential for social science inquiry. w

…(Full text truncated)…

This content is AI-processed based on ArXiv data.