Construct homonyms – constructs with the same label but different meanings – are pervasive in social science and pose a significant challenge to theory development, scale creation, and the accumulation of robust empirical findings. Despite their impact, these subtle inconsistencies often remain undetected, hindering progress in the field. Our paper introduces a novel, AI-aided methodology that leverages latent semantic analysis (LSA) and large language models (LLMs) to uncover these hidden construct homonyms with unprecedented efficiency. We demonstrate the power of this approach through an in-depth analysis of three widely recognized scales of paternalistic leadership: Cheng et al.’s (2000) 3-factor model, Aycan’s (2006) 5-factor model, and Pellegrini and Scandura’s (2008) 1-factor model. Our findings reveal seven major themes, yet only two themes – “assert leaders’ authoritative rights” and “expect loyalty from followers” – are shared across all three models. Intriguingly, each scale also captures unique dimensions: Cheng et al.’s model highlights “moral leadership” and “the combination of benevolence and discipline,” while Aycan’s and Pellegrini & Scandura’s scales emphasize themes like “creating a family atmosphere” and “emotional involvement in followers’ personal lives.” These results illuminate the nuanced operationalization of paternalistic leadership across cultural contexts, offering researchers actionable insights for selecting the most appropriate instruments for their work. By partnering with AI, our methodology provides a powerful tool for identifying and addressing construct homonyms, paving the way for greater conceptual clarity and theoretical advancement.
