Using Structural Equation Modeling to Test for Multidimensionality

Abstract

Abstract Unidimensionality is an assumption of many statistical techniques that examine the psychometric properties of measures. In fact, unidimensionality is necessary for construct validity. This article illustrates how structural equation modeling can be used to test the multidimensionality of a measure. By using data collected on a multidimensional measure, we compare an oblique factor model with a higher order factor model. Results indicate that the oblique factor model fit the data significantly better, because the measure is multidimensional. If the measure was unidimensional, then the higher order factor should explain the correlation between the latent variables. Implications for researchers and measurement development are discussed.