par4 <- 'P1 P5 Q1 Q3 P95 P99' par3 <- '5' par2 <- '12' par1 <- '50' par1 <- as.numeric(par1) par2 <- as.numeric(par2) par3 <- as.numeric(par3) if (par1 < 10) par1 = 10 if (par1 > 5000) par1 = 5000 if (par2 < 3) par2 = 3 if (par2 > length(x)) par2 = length(x) library(modeest) library(lattice) library(boot) boot.stat <- function(s) { s.mean <- mean(s) s.median <- median(s) s.midrange <- (max(s) + min(s)) / 2 s.mode <- mlv(s,method='mfv')$M s.kernelmode <- mlv(s, method='kernel')$M c(s.mean, s.median, s.midrange, s.mode, s.kernelmode) } (r <- tsboot(x, boot.stat, R=par1, l=12, sim='fixed')) bitmap(file='plot1.png') plot(r$t[,1],type='p',ylab='simulated values',main='Simulation of Mean') grid() dev.off() bitmap(file='plot2.png') plot(r$t[,2],type='p',ylab='simulated values',main='Simulation of Median') grid() dev.off() bitmap(file='plot3.png') plot(r$t[,3],type='p',ylab='simulated values',main='Simulation of Midrange') grid() dev.off() bitmap(file='plot7a.png') plot(r$t[,4],type='p',ylab='simulated values',main='Simulation of Mode') grid() dev.off() bitmap(file='plot8a.png') plot(r$t[,5],type='p',ylab='simulated values',main='Simulation of Mode of Kernel Density') grid() dev.off() bitmap(file='plot4.png') densityplot(~r$t[,1],col='black',main='Density Plot',xlab='mean') dev.off() bitmap(file='plot5.png') densityplot(~r$t[,2],col='black',main='Density Plot',xlab='median') dev.off() bitmap(file='plot6.png') densityplot(~r$t[,3],col='black',main='Density Plot',xlab='midrange') dev.off() z <- data.frame(cbind(r$t[,1],r$t[,2],r$t[,3],r$t[,4],r$t[,5]) ) colnames(z) <- list('mean','median','midrange','mode','mode.k.dens') bitmap(file='plot7.png') boxplot(z,notch=TRUE,ylab='simulated values',main='Bootstrap Simulation - Central Tendency') grid() dev.off() if (par4 == 'P1 P5 Q1 Q3 P95 P99') { myq.1 <- 0.01 myq.2 <- 0.05 myq.3 <- 0.95 myq.4 <- 0.99 myl.1 <- 'P1' myl.2 <- 'P5' myl.3 <- 'P95' myl.4 <- 'P99' } if (par4 == 'P0.5 P2.5 Q1 Q3 P97.5 P99.5') { myq.1 <- 0.005 myq.2 <- 0.025 myq.3 <- 0.975 myq.4 <- 0.995 myl.1 <- 'P0.5' myl.2 <- 'P2.5' myl.3 <- 'P97.5' myl.4 <- 'P99.5' } if (par4 == 'P10 P20 Q1 Q3 P80 P90') { myq.1 <- 0.10 myq.2 <- 0.20 myq.3 <- 0.80 myq.4 <- 0.90 myl.1 <- 'P10' myl.2 <- 'P20' myl.3 <- 'P80' myl.4 <- 'P90' } load(file='createtable') a<-table.start() a<-table.row.start(a) a<-table.element(a,'Estimation Results of Blocked Bootstrap',10,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'statistic',header=TRUE) a<-table.element(a,myl.1,header=TRUE) a<-table.element(a,myl.2,header=TRUE) a<-table.element(a,'Q1',header=TRUE) a<-table.element(a,'Estimate',header=TRUE) a<-table.element(a,'Q3',header=TRUE) a<-table.element(a,myl.3,header=TRUE) a<-table.element(a,myl.4,header=TRUE) a<-table.element(a,'S.D.',header=TRUE) a<-table.element(a,'IQR',header=TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'mean',header=TRUE) q1 <- quantile(r$t[,1],0.25)[[1]] q3 <- quantile(r$t[,1],0.75)[[1]] p01 <- quantile(r$t[,1],myq.1)[[1]] p05 <- quantile(r$t[,1],myq.2)[[1]] p95 <- quantile(r$t[,1],myq.3)[[1]] p99 <- quantile(r$t[,1],myq.4)[[1]] a<-table.element(a,signif(p01,par3)) a<-table.element(a,signif(p05,par3)) a<-table.element(a,signif(q1,par3)) a<-table.element(a,signif(r$t0[1],par3)) a<-table.element(a,signif(q3,par3)) a<-table.element(a,signif(p95,par3)) a<-table.element(a,signif(p99,par3)) a<-table.element( a,signif( sqrt(var(r$t[,1])),par3 ) ) a<-table.element(a,signif(q3-q1,par3)) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'median',header=TRUE) q1 <- quantile(r$t[,2],0.25)[[1]] q3 <- quantile(r$t[,2],0.75)[[1]] p01 <- quantile(r$t[,2],myq.1)[[1]] p05 <- quantile(r$t[,2],myq.2)[[1]] p95 <- quantile(r$t[,2],myq.3)[[1]] p99 <- quantile(r$t[,2],myq.4)[[1]] a<-table.element(a,signif(p01,par3)) a<-table.element(a,signif(p05,par3)) a<-table.element(a,signif(q1,par3)) a<-table.element(a,signif(r$t0[2],par3)) a<-table.element(a,signif(q3,par3)) a<-table.element(a,signif(p95,par3)) a<-table.element(a,signif(p99,par3)) a<-table.element(a,signif(sqrt(var(r$t[,2])),par3)) a<-table.element(a,signif(q3-q1,par3)) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'midrange',header=TRUE) q1 <- quantile(r$t[,3],0.25)[[1]] q3 <- quantile(r$t[,3],0.75)[[1]] p01 <- quantile(r$t[,3],myq.1)[[1]] p05 <- quantile(r$t[,3],myq.2)[[1]] p95 <- quantile(r$t[,3],myq.3)[[1]] p99 <- quantile(r$t[,3],myq.4)[[1]] a<-table.element(a,signif(p01,par3)) a<-table.element(a,signif(p05,par3)) a<-table.element(a,signif(q1,par3)) a<-table.element(a,signif(r$t0[3],par3)) a<-table.element(a,signif(q3,par3)) a<-table.element(a,signif(p95,par3)) a<-table.element(a,signif(p99,par3)) a<-table.element(a,signif(sqrt(var(r$t[,3])),par3)) a<-table.element(a,signif(q3-q1,par3)) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'mode',header=TRUE) q1 <- quantile(r$t[,4],0.25)[[1]] q3 <- quantile(r$t[,4],0.75)[[1]] p01 <- quantile(r$t[,4],myq.1)[[1]] p05 <- quantile(r$t[,4],myq.2)[[1]] p95 <- quantile(r$t[,4],myq.3)[[1]] p99 <- quantile(r$t[,4],myq.4)[[1]] a<-table.element(a,signif(p01,par3)) a<-table.element(a,signif(p05,par3)) a<-table.element(a,signif(q1,par3)) a<-table.element(a,signif(r$t0[4],par3)) a<-table.element(a,signif(q3,par3)) a<-table.element(a,signif(p95,par3)) a<-table.element(a,signif(p99,par3)) a<-table.element(a,signif(sqrt(var(r$t[,4])),par3)) a<-table.element(a,signif(q3-q1,par3)) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'mode k.dens',header=TRUE) q1 <- quantile(r$t[,5],0.25)[[1]] q3 <- quantile(r$t[,5],0.75)[[1]] p01 <- quantile(r$t[,5],myq.1)[[1]] p05 <- quantile(r$t[,5],myq.2)[[1]] p95 <- quantile(r$t[,5],myq.3)[[1]] p99 <- quantile(r$t[,5],myq.4)[[1]] a<-table.element(a,signif(p01,par3)) a<-table.element(a,signif(p05,par3)) a<-table.element(a,signif(q1,par3)) a<-table.element(a,signif(r$t0[5],par3)) a<-table.element(a,signif(q3,par3)) a<-table.element(a,signif(p95,par3)) a<-table.element(a,signif(p99,par3)) a<-table.element(a,signif(sqrt(var(r$t[,5])),par3)) a<-table.element(a,signif(q3-q1,par3)) a<-table.row.end(a) a<-table.end(a) table.save(a,file='mytable.tab')
|