Construct Validity of Likert Scales through Confirmatory Factor Analysis: A Simulation Study Comparing Different Methods of Estimation Based on Pearson and Polychoric Correlations

Abstract

The widespread use of Pearson correlations and, by extension, the Maximum Likelihood estimation method, does not take into account the measurement properties of Likert scales observed variables when carrying out a construct validity process through Confirmatory Factor Analysis (CFA). This simulation study compares four estimation methods (Maximum Likelihood –ML-, Robust Maximum Likelihood –RML-, Robust Unweighted Least Squares –, RULS) according to two of the assumptions CFA is supposed to fulfil: multivariate normality and, especially, the continuous measurement nature of both latent and observed variables. Goodness of fit is diagnosed by X² Likelihood Ratio Test and RMSEA indices. Results suggest ULS and RULS are preferable as polychoric correlations help to overcome grouping and transformation errors produced when using Pearson correlations for ordinal observed variables. Data measurement scale consideration enhances the ability of hypothesized models to reproduce accurately construct variables relationships.