A note on using scale sum scores in path analysis

Sum scores are widely used in the social sciences, yet their appropriateness remains a topic of considerable debate in the psychometric literature. A recent article by Raykov and Zhang (2025, Struct. Equ. Model.) has cautioned against employing sum scores as predictor variables in subsequent analyses, as this practice may lead to biased estimates of regression coefficients. As an alternative, structural equation modeling (SEM) based on a unidimensional factor model—where the latent factor replaces the sum score—has been advocated. The present article argues that reliability adjustments can also be implemented without resorting to SEM, using reliability-corrected regression models designed for measurement error correction. Furthermore, it is demonstrated that the SEM approach becomes inferior to measurement error correction methods when the assumption of a unidimensional measurement model is violated or when design-based reliability indices, such as Cronbach’s alpha, are preferred over model-based alternatives like McDonald’s omega. The article concludes that a fully integrated SEM approach, combining both measurement and structural components, is advantageous over measurement error correction approaches with reliability adjustment only under specific and limited conditions.