Abstract

Purpose: This study aims to validate the potential of generative AI–based virtual data as a methodological alternative for testing talent models in military organizations, where data collection is restricted. Methodology: Using the Army Excellent Talent Characteristics Model as a framework, virtual data were generated through Conditional Tabular GAN (CTGAN) and evaluated using confirmatory factor analysis, independent sample t-tests, and multi-group structural equation modeling. Results: The first virtual dataset failed to meet convergent validity, but the regenerated dataset achieved acceptable construct reliability and average variance extracted values. The analysis showed that virtual data could replicate the overall measurement structure of actual data, yet differences emerged in the strength of structural paths. Specifically, leadership and communication satisfaction had stronger effects on trust in the actual dataset, while creativity was significant only in the virtual dataset. Trust influenced organizational citizenship behavior in both groups, but the explanatory power was higher in the real dataset. Conclusion: These findings suggest that virtual data can serve as a useful tool for validating latent structures, pre-testing survey designs, and enhancing reproducibility. However, its role should be considered complementary rather than a full substitute for real-world data when examining causal relationships in military human resource research.