CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1680
Title: Comparing methods for discovering unobserved heterogeneity in PLS-PM: A Monte Carlo study Authors:  Laura Trinchera - NEOMA Business School (France)
Giorgio Russolillo - CNAM (France) [presenting]
Abstract: PLS Path Modeling assumes homogeneity among the observed units. In particular a unique model, i.e. the global model, is estimated for the whole dataset. Real data are often affected by unobserved heterogeneity: A unique model may hide important differences. Looking for homogeneous groups is primordial for such datasets. Response-based Unit Segmentation (REBUS-PLS) and PLS prediction-oriented segmentation (PLS-POS) are two recent approaches for dealing with unobserved heterogeneity in PLS path models. Those two methods share a common idea: both aim at identifying group-specific models with an higher predicting capability compared to the global model. We present a Monte Carlo simulation study for comparing REBUS-PLS and PLS-POS in terms of both prediction and parameter recovering. Moreover we asses their pertinence for high-dimensional data in terms of computational time and robustness of the results.