A PLS-SEM Comparison of Human and LLM-Generated Responses

Recent advances in large language models (LLMs) have prompted interest in their potential to simulate human survey responses, but their ability to replicate theory-driven causal structures remains underexplored. This study investigates whether LLMs can generate survey data that structurally reproduces a social science model integrating Protection Motivation Theory (PMT) and Motivated Consumer Innovativeness (MCI). Using partial least squares structural equation modeling (PLS-SEM), we compared human responses with LLM-generated responses created through persona-based prompts reflecting real panelists' demographics and psychological traits. The analysis reveals that LLMs can reproduce several key path coefficients found in the human data, demonstrating structural consistency in core theoretical relationships. While not uniformly consistent across all constructs, these findings suggest that LLMs may function as a complementary pre-testing instrument to refine survey design and assess model feasibility prior to full-scale data collection.