Parceling in Structural Equation Modeling

Abstract

Parceling is pre-modeling strategy to create fewer and more reliable indicators of constructs for use with latent variable models. Parceling is particularly useful for developmental scientists because longitudinal models can become quite complex and even intractable when measurement models of items are fit. In this Element the authors provide a detailed account of the advantages of using parcels, their potential pitfalls, as well as the techniques for creating them for conducting latent variable structural equation modeling (SEM) in the context of the developmental sciences. They finish with a review of the recent use of parcels in developmental journals. Although they focus on developmental applications of parceling, parceling is also highly applicable to any discipline that uses latent variable SEM.

Adaptation of the CAD Scale to Japanese Parents: The Domains for Moral Violations in the CAD Triad Hypothesis

“…A parcel is a sum of two or more items to create more reliable indicators in the same construct (Kishton & Wideaman, 1994;Little et al, 2002;Little et al, 2013;Matsunaga, 2008). The technique of using parcels as an indicator of constructs was accompanied by several controversies for a couple of decades (Meade & Kroustalis, 2006;Little et al, 2013;Little et al, 2022;Matsunaga, 2008;Rioux et al, 2020). The arguments against item parcelling are the issues of estimation bias, model misspecification, (Bandalos & Finney, 2001;Hall et al, 1999), and masking differences between groups (Meade & Kroustalis, 2006).…”

“…The random algorithm is a method in which items are allocated to parcels randomly, with average numbers of items across parcels, without repetition (Matsunaga, 2008;Little et al, 2002;Little et al, 2013). The random algorithm is reasonable as a stratagem because it can be a good strategy if there are a large number of samples (N > 200), many items, high item communality, and item diversity (Little et al, 2013;Little et al, 2022). After creating parcels, we examined parcelled CFA and compared goodness-of-fit indices of those models with item-level models.…”

The important role of mothers during displacement: Direct and indirect effects of the refugee context on Syrian refugee children's mental health

“…To reduce the complexity and improve the stability of our models, we conducted subset item parceling to create latent factors capturing war exposure, the physical environment, social environment, negative parenting, supportive parenting, and child PTSD, depression, and externalizing. We created three parcels for each latent construct to maximize stability, in line with recommendations from the literature (Little et al., 2022). We used item loadings from exploratory factor analyses as a guide for which parcel to assign each item to, created parcels via sums of items, and then tested the fit of each latent measure at each wave with confirmatory factor analysis to see if it adequately replicated the underlying covariance structure (Hu & Bentler, 1999).…”

Remote, Disconnected, or Detached? Examining the Effects of Psychological Disconnectedness and Cynicism on Employee Performance, Wellbeing, and Work–Family Interface

“…In order to maximize the reliability and parsimony of our structural equation model, item parcels were created for construct measures with more than three items (i.e., disconnectedness, cynicism, mental health, and cognitive failures). We followed Little, Cunningham, Shahar, and Widaman's [71] and Little, Rioux, Odejimi, and Stickley's [72] recommendations and sequentially assigned items on the basis of the highest to lowest corrected item-to-scale correlations to create three item parcels per construct. Subsequent analyses were conducted using the parcels as manifest indicators of the latent variables with Mplus 8.0 [73].…”