DMCEM: DMCEM

In DIRMR: Distributed Imputation for Random Effects Models with Missing Responses

DMCEM

Description

The DMCEM method uses the sample mean to approximate the integral in step E, rather than performing a single Monte Carlo sampling on the entire dataset. This enables DMCEM to significantly reduce the computation time and memory consumption when processing large datasets, while updating model parameters by calculating the sample mean of each subset.

Usage

DMCEM(data,df1,M,maxiter)

Arguments

data	The real data sets with missing data used in the method
df1	The real data sets used in the method
M	The number of Blocks
maxiter	The maximum number of iterations

Value

Y011	The response variable value after projection for each block
Yhat	The estimated response variable value after projection for each block
Ymean	The mean of response variable value after projection for each block
Yhatmean	The mean of response variable value after projection for each block

Author(s)

Guangbao Guo,Yu Li

Examples

set.seed(99)
library(MASS)
library(mvtnorm)
n=50;p=6;q=5;M=2;omega=0.15;ratio=0.1;maxiter=15;nob=round(n-(n*ratio))
dd.start=1;sigma2_e.start=1
X0=matrix(runif(n*p,0,2),ncol=p)
beta=matrix(rnorm(p*1,0,3),nrow=p)
Z0=matrix(runif(n*q,2,3),ncol=q)
e=matrix(rnorm(n*1,0,sigma2_e.start),n,1)
b=matrix(rnorm(q*1,0,1),q,1)
Y0=X0
df1=data.frame(Y=Y0,X=X0,Z=Z0)
misra=function(data,ratio){
  nob=round(n-(n*ratio))
  data[sample(n,n-nob),1]=NA
  return(data)}
data=misra(data=df1,ratio=0.1)
DMCEM(data,df1,M=2,maxiter=15)

DIRMR documentation built on April 3, 2025, 6:03 p.m.