Modeling Data with Measurement Errors but without Predefined Metrics: Fact versus Fallacy

Abstract

Data in social and behavioral sciences typically contain measurement errors and also do not have predefined metrics. Structural equation modeling (SEM) is commonly used to analyze such data. This article discuss issues in latent-variable modeling as compared to regression analysis with composite-scores. Via logical reasoning and analytical results as well as the analyses of two real datasets, several misconceptions related to bias and accuracy of parameter estimates, standardization of variables, and result interpretation are clarified. The results are expected to facilitate better understanding of the strength and limitations of SEM and regression analysis with weighted composites, and to advance social and behavioral data science.