Combination rules for homoscedastic and heteroscedastic MANOVA models from multiply imputed datasets.

Area of Science:

• Statistics
• Multivariate Analysis
• Data Science

Background:

• Classical MANOVA and MBF procedures require complete data.
• Missing data can compromise the validity of these statistical tests.
• Existing methods for handling missing data may reduce statistical power.

Purpose of the Study:

• To develop and evaluate combination rules for MANOVA and MBF procedures with multiply imputed datasets.
• To assess the performance of these new methods in terms of type I error rates and statistical power.
• To compare the proposed methods against traditional approaches with complete data and listwise deletion.

Main Methods:

• Development of combination rules for MANOVA and MBF procedures applied to multiply imputed datasets.
• Illustration using a two-factor multivariate design with both equal and unequal covariance matrices (MI-MANOVA and MI-MBF).
• Monte-Carlo simulations to compare proposed methods with complete data and listwise deletion (LD-MANOVA, LD-MBF).

Main Results:

• Type I error rates were well-controlled across all analyses, including those with missing data and imputation.
• The MI-MANOVA approach demonstrated significantly higher statistical power than LD-MANOVA.
• The power of MI-MANOVA was comparable to complete data MANOVA, and MI-MBF performed similarly with unequal covariance matrices.

Conclusions:

• The proposed combination rules for MANOVA and MBF with multiply imputed data perform well.
• These methods offer a practical solution for handling missing data in multivariate analyses.
• The MI-MANOVA and MI-MBF procedures provide robust and powerful alternatives to traditional methods when data are incomplete.