Model-Free Measurement of Case Influence in Structural Equation Modeling

• In Event: Methods for Multilevel, Categorical, and Multidimensional Data

Abstract

This study introduces a model-free case influence measure (DOCR) to SEM and evaluates its performance compared to that of Mahalanobis Distance and Generalized Cook’s Distance when the sample size, proportion of target cases to non-target cases, and type of model used to generate the data are manipulated. The findings suggest DOCR generally performed better than the other measures in identifying target cases across all simulated conditions. However, the performance of DOCR under small sample size was not satisfactory, and results suggested DOCR is sensitive to sample size. Recommendations for current use of DOCR and for future research are provided.

Authors

• Fathima Mohrrag Jaffari, National center for assessment

• Jennifer Koran, Southern Illinois University Carbondale