Multiple imputation of missing data in multilevel models with the R package mdmb: a flexible sequential modeling...

Area of Science:

• Statistics
• Multilevel Modeling
• Bayesian Inference

Background:

• Multilevel models frequently incorporate nonlinear effects like random slopes and interactions.
• Missing data in these models pose estimation challenges.
• Existing multiple imputation (MI) methods for multilevel data may not adequately address nonlinear variable associations.

Purpose of the Study:

• To propose a novel sequential modeling approach for handling missing data in multilevel models with nonlinear effects.
• To ensure imputations are compatible with the substantive analysis model.
• To provide a flexible and accurate method for complex multilevel data.

Main Methods:

• A sequential modeling approach utilizing Bayesian estimation techniques.
• Decomposition of the joint data distribution into parts corresponding to outcome and explanatory variables.
• Implementation in the R package 'mdmb' for practical application.

Main Results:

• The sequential modeling approach effectively handles missing data in multilevel models with nonlinear effects.
• Comparison with conventional and other compatible MI approaches demonstrated its efficacy.
• Simulation studies validated the performance of the proposed method.

Conclusions:

• The proposed Bayesian sequential modeling approach offers a robust solution for missing data in complex multilevel models.
• This method properly accounts for nonlinear associations, enhancing imputation accuracy.
• The 'mdmb' R package facilitates the application of this advanced technique.