Appropriateness of composites in structural equation models

Area of Science:

• Psychometrics
• Health Services Research
• Structural Equation Modeling

Background:

• Assessing physical health often involves complex multiple indicator models.
• These models can be impractical for real-world applications.
• Simplifying these models is crucial for broader usability.

Purpose of the Study:

• To evaluate the appropriateness of using composite variables instead of multiple indicators in physical health structural models.
• To develop a practical strategy for simplifying complex health measurement models.
• To determine if composite variables can accurately represent latent constructs of physical health.

Main Methods:

• A two-stage strategy was employed to simplify the model.
• Stage one involved estimating reliability for each composite to fix measurement error variance and composite-to-latent variable regression.
• Stage two reestimated the model with these constrained parameters, followed by regression analyses.

Main Results:

• Parameter estimates for causal linkages and residual error variances using composites closely matched those from the multiple indicators approach.
• The proposed simplification strategy demonstrated high fidelity to the original, more complex model.
• Reliability estimation and parameter constraint effectively managed measurement error.

Conclusions:

• Using composite variables is an appropriate and practical simplification for structural models of physical health.
• The proposed two-stage strategy provides a valid method for enhancing the practicality of health measurement models.
• This approach supports the use of composites in structural modeling without significant loss of information or accuracy.