Projective Resampling Imputation Estimation Method for Missing Covariates Problems
In this work, motivated by the spirit of projection resampling/random projection, we propose a Projection Resampling Imputation Mean Estimation (PRIME) to project the covariates along randomly sampled directions to obtain samples of scalar-valued predictors (dimension reduction). And then a simple geometric average is taken on the scalar predictors to impute the missing parts (using all-sided information). In general, several advantages of our method can be concluded as follows. First, the PRIME is capable of dealing with missing data containing no complete observations, while most existed methods require a fraction of subjects to have fully completed observations. Second, we can integrate the averaging of the imputation estimation of multiple projection directions and fully utilize the available information to reach a more reliable and insightful result.
Reproduce code for case1 in our paper
-
help_func.Rincludes some helpful functions about PRIME -
example.Ris the reproducing code for case1 in our paper
#you should install some depedency packages, such as MASS,misaem,ggplot2,gridExtra
source("./help_func.R")
source("./example.R")Qustions
Please contact 13023988263@163.com or ele717@163.com with any questions about the PRIME method.