Problems with Rationales for Parceling that Fail to Consider Parcel-Allocation Variability.

Area of Science:

• Structural Equation Modeling (SEM)
• Psychometrics
• Statistical Methodology

Background:

• Parcels (averages or sums of item scores) are frequently used as indicators for latent constructs in SEM.
• Parcel-allocation variability (PAV) refers to result variations arising from different ways items are grouped into parcels within a sample.
• PAV can impact model fit, parameter estimates, standard errors, and inferential decisions, raising concerns about representativeness and replicability.

Purpose of the Study:

• To explain and demonstrate the problems associated with current rationales for parceling in SEM that do not account for PAV.
• To challenge the notion that specific parceling strategies or unidimensionality tests inherently avoid PAV.
• To propose an alternative approach for improving power in detecting misspecification without the risks of PAV.

Main Methods:

• Explanation and demonstration of parcel-allocation variability (PAV) in structural equation modeling.
• Critique of existing methodological literature justifying parceling without addressing PAV.
• Comparison of item-level models with varying structural constraints to assess power and avoid PAV.

Main Results:

• Purposive parceling algorithms for multidimensional constructs do not eliminate PAV.
• Passing unidimensionality tests at the item level does not prevent PAV in parcel-level models.
• Addressing PAV is necessary when parceling; alternatively, comparing item-level models offers higher power for detecting misspecification.

Conclusions:

• Current justifications for parceling in SEM are often flawed as they neglect parcel-allocation variability (PAV).
• Researchers must quantify and account for PAV to ensure the representativeness and replicability of their findings.
• Comparing item-level models with differing structural constraints is a superior method for enhancing statistical power while avoiding PAV.