Abstract
This research introduces a structured methodological framework for identifying and interpreting latent heterogeneity in variance-based structural equation models using multigroup analysis (MGA). Although partial least squares structural equation modelling (PLS-SEM) is widely used in business research, most applications assume parameter homogeneity by interpreting pooled path coefficients as invariant across heterogeneous firms. Such assumptions can obscure context-specific mechanisms and produce misleading theoretical conclusions. The framework integrates theoretical justification for subgroup formation, data preparation, measurement invariance testing, multigroup estimation, and interpretive guidance. An empirical illustration using survey data from 642 SMEs shows that pooled models conceal systematic differences in the relationships among engagement, innovation search, and performance across firm sizes. Results reveal scale-contingent mechanisms that redefine mediation patterns. Methodologically, the study advances transparent guidelines for MGA in PLS-SEM. Substantively, it demonstrates how explicit heterogeneity assessment sharpens theoretical boundary conditions and strengthens causal interpretation in business research.