[R] Speed up studentized confidence intervals ?

varin sacha v@r|n@@ch@ @end|ng |rom y@hoo@|r
Wed Dec 29 20:08:33 CET 2021


Dear David,
Dear Rui,

Many thanks for your response. It perfectly works for the mean. Now I have a problem with my R code for the median. Because I always get 1 (100%) coverage probability that is more than very strange. Indeed, considering that an interval whose lower limit is the smallest value in the sample and whose upper limit is the largest value has 1/32 + 1/32 = 1/16 probability of non-coverage, implying that the confidence of such an interval is 15/16 rather than 1 (100%), I suspect that the confidence interval I use for the median is not correctly defined for n=5 observations, and likely contains all observations in the sample ? What is wrong with my R code ?

########################################
library(boot)

s=rgamma(n=100000,shape=2,rate=5)
median(s)

N <- 100
out <- replicate(N, {
a<- sample(s,size=5)
median(a) 

dat<-data.frame(a)
med<-function(d,i) {
temp<-d[i,]
median(temp)
}

  boot.out <- boot(data = dat, statistic = med, R = 10000)
  boot.ci(boot.out, type = "bca")$bca[, 4:5]
})

#coverage probability
median(out[1, ] < median(s) & median(s) < out[2, ])
########################################




Le jeudi 23 décembre 2021, 14:10:36 UTC+1, Rui Barradas <ruipbarradas using sapo.pt> a écrit : 





Hello,

The code is running very slowly because you are recreating the function 
in the replicate() loop and because you are creating a data.frame also 
in the loop.

And because in the bootstrap statistic function med() you are computing 
the variance of yet another loop. This is probably statistically wrong 
but like David says, without a problem description it's hard to say.

Also, why compute variances if they are never used?

Here is complete code executing in much less than 2:00 hours. Note that 
it passes the vector a directly to med(), not a df with just one column.


library(boot)

set.seed(2021)
s <- sample(178:798, 100000, replace = TRUE)
mean(s)

med <- function(d, i) {
  temp <- d[i]
  f <- mean(temp)
  g <- var(temp)
  c(Mean = f, Var = g)
}

N <- 1000
out <- replicate(N, {
  a <- sample(s, size = 5)
  boot.out <- boot(data = a, statistic = med, R = 10000)
  boot.ci(boot.out, type = "stud")$stud[, 4:5]
})
mean(out[1, ] < mean(s) & mean(s) < out[2, ])
#[1] 0.952



Hope this helps,

Rui Barradas

Às 11:45 de 19/12/21, varin sacha via R-help escreveu:
> Dear R-experts,
> 
> Here below my R code working but really really slowly ! I need 2 hours with my computer to finally get an answer ! Is there a way to improve my R code to speed it up ? At least to win 1 hour ;=)
> 
> Many thanks
> 
> ########################################################
> library(boot)
> 
> s<- sample(178:798, 100000, replace=TRUE)
> mean(s)
> 
> N <- 1000
> out <- replicate(N, {
> a<- sample(s,size=5)
> mean(a)
> dat<-data.frame(a)
> 
> med<-function(d,i) {
> temp<-d[i,]
> f<-mean(temp)
> g<-var(replicate(50,mean(sample(temp,replace=T))))
> return(c(f,g))
> 
> }
> 
>    boot.out <- boot(data = dat, statistic = med, R = 10000)
>    boot.ci(boot.out, type = "stud")$stud[, 4:5]
> })
> mean(out[1,] < mean(s) & mean(s) < out[2,])
> ########################################################
> 
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 



More information about the R-help mailing list