Communications for Statistical Applications and Methods

Indexed in /covered by CAS, KoreaScience & DOI/Crossref:eISSN 2383-4757   pISSN 2287-7843

Fig. 1.

Download original image
Fig. 1. Iterative integrated imputation with a toy example. X is a data matrix of 5 observation vectors with missing values that do not form a rectangle. Iteratively for each column with missing elements xi (i = 3, 4, 5), we apply structured matrix completion to the matrix χi, consisting of all complete data vectors; xi; and any columns with missing values that are a subset of those missing from xi. Imputation is performed on the entire rectangular submatrix of all features missing in xi, with any observed elements not from xi treated as missing. So, imputing x3 and x4 uses x5, whereas x5 is imputed using only the complete data vectors.
Communications for Statistical Applications and Methods 2019;26:411-30 https://doi.org/10.29220/CSAM.2019.26.4.411
© 2019 Communications for Statistical Applications and Methods