Item Parceling in Structural Equation Modeling: A Primer

Abstract

This article provides a primer of the statistical technique called parceling, or aggregating items and using those aggregates as indicators of latent constructs, for structural equation modeling (SEM). First, two major types of parceling (subset-item-parcel and all-item-parcel approaches), alongside the traditional item-based approach, are illustrated. Second, both pros and cons of parceling are explicated. Parceling provides psychometric and modeling-related benefits. Risks associated with parceling are also noted. Particularly, potential induction of estimation bias and model misspecification are pointed out. In relation to the latter problem, unidimensionality of the scale is highlighted as an important prerequisite for the use of parceling. Finally, issues of the number of parcels per factor and parcel-building algorithms are discussed. Forming three parcels per factor by the random algorithm is recommended. Suggestions and general guidelines for the use of item parceling for SEM are also provided.