Standardized estimates for latent interaction effects: Method comparison and selection strategy

Abstract

Standardized estimation plays an important role in model interpretation and effect size comparison. The appropriate standardized estimation of latent interaction effects, appeared over 10 years ago, has been widely adopted. However, there is no systematic comparison of different methods in their appropriate standardized estimation. Through a simulation study in different conditions, we compared the latent interaction effects of three methods: The product indicator approach, Latent Moderated Structural Equations (LMS), Bayesian methods with or without prior information. The results showed that, when normality assumption was satisfied, LMS and the informative Bayesian method performed best in standardized estimation. When normality assumption was violated, however, the product indicator approach was more robust than the other methods, but a large sample size (not less than 500) was necessary. For small samples, when the correlation between exogenous latent variables was close to zero, the noninformative Bayesian method could be used.