A Note on Conditional AIC for Linear Mixed-Effects Models.

Area of Science:

• Statistics
• Mixed-Effects Models
• Model Selection

Background:

• The Akaike Information Criterion (AIC) is commonly used for model selection in mixed-effects models.
• Marginal AIC is inappropriate for cluster-focused analyses, as shown by Vaida and Blanchard (2005).
• Conditional AIC was proposed but relies on strong assumptions about the random effects variance-covariance matrix.

Purpose of the Study:

• To develop a generalized conditional AIC that removes restrictive assumptions.
• To enhance the applicability of conditional AIC in mixed-effects modeling.
• To provide a more robust model selection criterion for clustered data.

Main Methods:

• Development of a generalized conditional AIC formula.
• Relaxation of assumptions regarding the known variance-covariance matrix of random effects.
• Evaluation through simulation studies.

Main Results:

• The proposed generalized conditional AIC is shown to be promising in simulation studies.
• The new method expands the applicability of conditional AIC.
• The developed criterion offers a more flexible approach to model selection.

Conclusions:

• The generalized conditional AIC provides a valuable tool for model selection in mixed-effects models, particularly for clustered data.
• This approach overcomes limitations of previous conditional AIC methods.
• Further research and application are warranted to validate its performance across diverse datasets.