lab:zhang:missing_data_mediation_analysis
Table of Contents
Simulation results
Data simulation
Listwise deletion
Pairwise deletion
BMEM
- Results from BMEM as of April 06, 2010 (no missing in X)
Multiple imputation method
Results
Files for simulation
Some codes
FIML in Mplus
- Simulation results as of April 07 2010 (no missing in X)
A simple example looking at the influence of MNAR auxiliary variables on regression coefficients
## a simple simulation proc.res<-function(out.file){ output <- suppressWarnings(readLines(out.file)) lnumber <- grep(" X", output) est<- output[lnumber[2]] sub('X', '', est) } res1<-NULL res2<-NULL for (i in 1:1000){ N<-500 x<-rnorm(N) y<-0*x + rnorm(N) z<-(x+y)+rnorm(N) cor(cbind(x,y,z)) x[sample(1:N, N*.3)]<-NA y[sample(1:N, N*.3)]<-NA z[z>quantile(z, .7)]<-NA write.table(cbind(x,y,z), 'testreg.txt', row.names=F, col.names=F, na='.') system("c:/programs/mplus/mplus reg.inp",show.output.on.console=F) temp1<-proc.res('reg.out') res1<-cbind(res1, temp1) system("c:/programs/mplus/mplus reg-aux.inp",show.output.on.console=F) temp2<-proc.res('reg-aux.out') res2<-cbind(res2, temp2) cat(i, '\n') } write.table(res1, 'res1.txt', row.names=F, col.names=F, quote=F) write.table(res2, 'res2.txt', row.names=F, col.names=F, quote=F) res1<-scan('res1.txt') res1<-array(res1, dim=c(4,1000)) res2<-scan('res2.txt') res2<-array(res2, dim=c(4,1000)) apply(res1,1,mean) apply(res2,1,mean)
Using EM in R
library(norm) library(MASS) library(mvtnorm) N<-500 r<-.9 set.seed(0) allres<-NULL for (i in 1:1000){ sigma <- matrix(c(1,0.3,0.3,1),2,2) yx <- mvrnorm(N, rep(0,2),sigma) yx[sample(1:N, N*r), 1]<-NA yx[yx[, 2]>quantile(yx[,2], 1-r), 2]<-NA m<-prelim.norm(yx) est<-em.norm(m) res<-getparam.norm(m, est) res<-c(unlist(res), mean(yx[,1], na.rm=T), var(yx[,1], na.rm=T)) allres<-rbind(allres, res) } apply(allres, 2, mean) ## function to get the likelihood function llik<-function(x, par){ #par[1]: mu1 #par[2]: mu2 #par[3]: s1 #par[4]: s2 #par[5]: s12 mu<-par[1:2] sigma <- array(c(par[3], par[5], par[5], par[4]), dim=c(2,2)) n<-dim(x)[1] l<-0 for (i in 1:n){ if ((!is.na(x[i,1])) & (!is.na(x[i,2]))){ l<-l + dmvnorm(x[i,], mu, sigma, log=T) } if ((!is.na(x[i,1])) & (is.na(x[i,2]))){ l<-l + dnorm(x[i,1], par[1], sqrt(par[3]), log=T) } if ((is.na(x[i,1])) & (!is.na(x[i,2]))){ l<-l + dnorm(x[i,2], par[2], sqrt(par[4]), log=T) } } l } nlm(llik, c(0,0,1,1,0), x=yx)
lab/zhang/missing_data_mediation_analysis.txt · Last modified: 2016/01/24 09:48 by 127.0.0.1