[R] boot and variances of the bootstrap replicates of the variable of interest?

Dan Bolser dmb at mrc-dunn.cam.ac.uk
Tue Jan 4 17:20:19 CET 2005


I want to use boot.ci to generate confidence intervals over the
bootstrapped mean(s) of a group of observations (i.e. I have 10
observations and I want to know how confident I can be on the value for
the mean).

I don't know (or want to know) the details of bootstrapping - I just have
the simplistic idea of taking samples, measuring a statistic on the
sample, and getting some confidence in the result based on the variance of
the sample statistics.

I am not sure which "type of interval [is] required", so I wanted to try
"studentized intervals", because some of my groups have a small number of
samples (like one or two). However, "variance estimates are required for
studentized intervals".

I can generate the variance for the statistic like this..


# Shame the 'mean' function dosn't work with boot
mean.x <- function(xx,ii) mean(xx[ii])

# Don't know how (or why) to pick a value for this.
nboot <- 1000 

# The 10 or so observations (in this case)
boot.sub <- boot(data=shabby.sub$a, statistic=mean.x, R=nboot)

# Variance of the statistic over the 1000 replicates
my.var.t <- var(boot.sub$t)

# t: A matrix with 'R' rows each of which is a bootstrap replicate
#    of 'statistic'. 

However, I need the variance for each R statistic... I think...

It is a shame that boot and boot.ci cant work more together better,
without me (a patent idiot) having to pass values (correctly) from the
results of boot into boot.ci.


Any help with the above would be appreciated, as I am about to weep
horrible tears (once more) at the alter of R.

(not really :)

Dan.




More information about the R-help mailing list