Multiple imputation of missing covariate values in multilevel models with random slopes: a cautionary note.

Area of Science:

• Statistics
• Multilevel Modeling
• Data Science

Background:

• Multiple imputation (MI) is a key method for handling missing data.
• Existing MI guidelines are challenging to apply to multilevel research.
• Random slope models present unique difficulties for MI.

Purpose of the Study:

• To explore MI applications in multilevel random-coefficient models.
• To address theoretical challenges of slope variation in MI.
• To identify limitations of current MI software for complex models.

Main Methods:

• Discussed MI applications in multilevel random-coefficient models.
• Investigated theoretical challenges of slope variation.
• Analyzed limitations of standard MI software.
• Conducted three simulation studies.

Main Results:

• MI effectively recovers most parameters but has limitations with slope variation when covariates are missing.
• MI provides reasonable parameter estimates, even in smaller samples or when assumptions are violated.
• Current MI software has limitations for fully capturing slope variation.

Conclusions:

• MI is a valuable tool for missing data in multilevel models, but requires careful consideration for slope variation.
• Researchers should be aware of MI's current limitations in complex multilevel models.
• Further software development is needed to fully address slope variation in MI.