Imputation of Mixed Data with Multilevel Singular Value Decomposition

We present a single imputation method for multilevel data containing mixed quantitative and qualitative variables with missing values. The approach is based on multilevel singular value decomposition, which decomposes data variability into between- and within-group components and applies SVD to each. The method is computationally efficient, handles diverse data types, and is demonstrated on simulations and medical data from multiple hospitals.

PDF Preprint