Online Program

Partial Least Squares Structural Equation Modeling as an Analysis Tool in Epidemiological Studies (303019)

Keywords: Partial least squares structural equation modelling, SmartPLS

Epidemiologic studies often have variables that are not directly measured. A method to capture these factors is latent variable analysis. While, traditionally, covariance-based SEM (CB-SEM) models have been used in this context, we present an alternative approach: partial least square structural equation modeling (PLS-SEM). This method is advantageous, since it allows for a formative measurement model, does not have as many distributional assumptions as CB-SEM, and works well for small sample sizes. Our motivating example is from a study of emergency department (ED) use in children. We use PLS-SEM to first construct a formative latent variable to measure how busy the ED is when a child presents with an injury or illness and then use this construct in a formal analysis to explore its relationship with wait time. We also used bootstrapping algorithm to test the significance of the variables included in the model. The SmartPLS (3.1.3) we used for this analysis was user friendly and included several other algorithms such as confirmatory tetrad analysis, finite mixture segmentation, and multi-group analysis.