The effect of missing levels of nesting in multilevel analysis

Area of Science:

• Biostatistics
• Statistical Modeling
• Data Analysis

Background:

• Multilevel analysis is crucial for hierarchical data across public health and genomics.
• Existing methods may lose information by ignoring data hierarchy or missing cluster levels.
• Accurate analysis requires methods that account for all hierarchy levels, especially with missing data.

Purpose of the Study:

• To introduce and evaluate a multilevel linear mixed effect model (LMM) with single imputation.
• To assess the model's ability to incorporate all data hierarchy levels despite missing clusters.
• To compare its performance against models that ignore hierarchy or missing intermediate clusters.

Main Methods:

• Developed a multilevel linear mixed effect model (LMM) incorporating single imputation for missing cluster levels.
• Applied the proposed LMM and comparative models to simulated data with varying cluster sizes and missing rates.
• Utilized hierarchically structured cohort data with missing intermediate levels for validation.

Main Results:

• The multilevel LMM with single imputation demonstrated superior accuracy in estimating fixed coefficients and variance components.
• This model showed lower mean squared error and better coverage probability compared to models ignoring hierarchy or missing clusters.
• Models ignoring data hierarchy or missing clusters tended to overestimate variance components of random effects.

Conclusions:

• Multilevel LMM with single imputation is a robust method for analyzing hierarchical data with missing cluster levels.
• This approach provides more accurate parameter estimates and avoids overestimation of random effects variance.
• The findings are supported by both simulation studies and real-world cohort data analysis.