Methodological Approaches to Studying Group Behavior: From Human Interaction to Agent-Based Models

Understanding group behavior demands a careful balance between ecological validity and experimental control, yet progress has often been constrained by methodological limitations. Despite these challenges, researchers have developed rigorous paradigms beyond traditional economic games—for example, collective foraging—particularly by leveraging new technologies such as virtual reality and large language models. To make sense of these advancements, this article proposes a conceptual and methodological framework for studying group behavior across three complementary approaches: (1) human–human interaction with controlled agents, (2) human–agent interaction with algorithm-driven behavior, and (3) agent–agent interaction through simulation. Leveraging advances in digital platforms, computational modeling, and large language model–driven agents, these approaches enable experimental probing of how local interactions give rise to group-level patterns. Each approach presents distinct trade-offs in realism, control, and scalability, and their integration allows for triangulation of findings and deeper theoretical insight. We advocate for integrating these approaches, where humans and intelligent agents interact within controlled yet realistic scenarios, designing algorithmic agents with more humanlike parameters, and combining them with advanced data collection techniques. This framework supports a richer understanding of human group behavior in an increasingly hybrid human–machine social landscape.