“What Would Happen If We Did This?” Using LLMs to Anticipate Policy Impacts

Session Abstract: Institutional research often asks: “What would happen if we implemented a new policy?” Traditional causal inference explains past outcomes but cannot anticipate future actions. This study integrates Large Language Models (LLMs) with Agent-Based Modeling (ABM) to enhance predictive policy analysis. Using Taiwan’s 2021–2022 reform—expanding student applications from three to six choices—as a case, we simulate nationwide exam and program selection data. By capturing student decision patterns, our AI-driven model compares enrollment outcomes under both policies. Findings show how LLM-ABM integration offers institutional leaders a powerful tool to forecast consequences and improve higher education decision-making.

Keywords: Agent-Based Policy Simulation, Behavioral Modeling with LLMs, Predictive Analytics in IR