The practice of aggregating lower level predictors in clustered data: A reflection on reflective variables

Area of Science:

• Psychometrics
• Multilevel Modeling
• Statistical Methods

Background:

• Multilevel modeling (MLM) with manifest aggregation of person and item scores can bias regression coefficients for formative variables.
• Latent variable modeling offers solutions but can present convergence and identification challenges.

Purpose of the Study:

• To investigate the appropriateness of manifest aggregation for reflective variables in multilevel modeling.
• To identify conditions where manifest aggregation is a viable alternative to latent variable modeling for L2 reflective variables.

Main Methods:

• Utilized population formula-based computations.
• Employed Monte Carlo simulations to evaluate conditions for manifest aggregation.

Main Results:

• Demonstrated specific conditions under which manifest aggregation is suitable for L2 reflective variables.
• Showcased how manifest aggregation can circumvent convergence and identification issues associated with latent variable models.

Conclusions:

• Manifest aggregation can be reliably used for L2 reflective variables under certain conditions, simplifying analysis.
• Researchers should consider latent aggregations when specific theoretical or data structures warrant them.