A Comparison of Four Approaches to Estimating Structural Coefficients in Small Samples With Multidimensional Items

• In Event: Methodological Innovations in Structural Equation Modeling

Abstract

This study compares four approaches to estimating structural coefficients in small sample sizes when items are multidimensional. Data are generated based on a structural regression (SR) model. Design factors for data generation include sample size, reliability coefficient, and magnitude of structural coefficient. Data are analyzed using the SR model, path analysis, and the single-indicator approach with maximum likelihood method, and the SR model with Bayesian approach. Different types of reliability coefficients are used for correction for measurement errors for the single-indicator approach. Preliminary results show that both the single-indicator and Bayesian approaches produced satisfactory parameter estimates but lacked power in most conditions, compared to the other two approaches.

Authors

• Yiting Yao, Florida State University

• Yanyun Yang, Florida State University