A Genetic Algorithm Segmentation Approach for Uncovering and Separating Groups of Data in PLS Path Modeling

Segmentation is a critical issue for PLS path modeling. Group specific model estimates can significantly differ from other groups or the overall model estimates (Ringle, Wende, & Will, 2007). Sequential segmentation strategies on the manifest data level, such as K-means or tree clustering, usually fail to identify groups of data with distinctive inner path model estimates (Ringle, 2006). The authors of this paper pursue three important objectives in contributing to segmentation in PLS path modeling. First, we present a new kind of PLS Path Model based Clustering approach (PLS-PMC) that uses a Genetic Algorithm (GA) to account for heterogeneity in the estimates for inner and outer path model relationships. Second, we develop an alternative and novel segmentation method for PLS-PMC that builds on preliminary subset partitions and demands less computation than the GA approach. Third, we introduce a data generation procedure for pre-specified PLS path model parameters to facilitate a simulation study. The study allows primary assessment of the GA segmentation approach and reveals its capabilities and effectiveness compared to our alterative method. Both pioneering techniques address the requirement to uncover and deal with heterogeneity of data with adequate means of segmentation in PLS path modeling. The findings are important for researchers and practitioners because they permit further differentiation of PLS estimates on the aggregate data level in terms of creating interpretations that are more precise for the results for each segment.