Psychometric Item Validation Using Virtual Respondents with Trait-Response Mediators

Researchers developed a framework using large language models to simulate virtual respondents for validating psychometric survey items, addressing the challenge of ensuring construct validity without costly human data collection. The approach uses trait-response mediators to identify survey items that robustly measure intended psychological traits across three major trait theories.

• →New framework enables cost-effective psychometric survey validation using LLM-simulated virtual respondents instead of expensive human studies.
• →The method accounts for mediators that influence how traits translate to survey responses, improving item validity assessment.
• →Testing across Big5, Schwartz, and VIA psychological trait theories demonstrated the framework's effectiveness.
• →LLMs showed capability to generate plausible mediators and simulate realistic human survey response behavior.
• →Researchers publicly released dataset and code to support future psychometric research applications.