The effects of parceling unidimensional scales on structural parameter estimates in structural equation modeling

Structural equation modeling allows several methods of estimating the disattermated association between 2 or more latent variables (i.e., the measurement model). In one common approach, measurement models are specified using item parcels as indicators of latent constructs. Item parcels versus original items are often used as indicators in these contexts to avoid estimation problems or solve issues associated with multivariate normality of the data. One concern associated with the use of item parceling is that no single "correct" approach exists to construct the parcels. Despite the controversy associated with selecting the most appropriate parceling method, less is understood with regard to how these methods influence the structural or path coefficients. By means of simulated and empirical data, this article addresses some commonly used strategies to model disattenuated structural coefficients between latent variables. Results revealed that when a single unidimensional scale is used to represent a latent construct, the use of individual items, item parcels, or an appropriate representation of measurement error through a single observed variable all will result in identical disattenuated structural coefficient estimates. Implications for the future of item parceling are discussed.