[R] build boot object for boot.ci

Daniel Metzler daniel.metzler at gmx.net
Thu Jan 5 21:36:40 CET 2006


Hello all,

I have a dataset with several variables v1...vn and five groups f1...fn.
For each group I took a subset of the data

f1<-subset(data,f==1)
f2<-subset(data,f==2)

and bootstrapped the weighted mean for v1...vn, which worked
nicely.

data.boot<-boot(f1$v1,mymean,R=2000)

Afterwards I combined all boot.out$t for each group:

boot$f1v1<-data.boot$t

data.boot<-boot(f1$v2,mymean,R=2000)
boot$f1v2<-data.boot$t
...
data.boot<-boot(f2$v1,mymean,R=2000)
boot$f2v1<-data.boot$t
...

Each row of boot is used as input in a model resulting in boot$W.

My idea is to find (bootstrap) CIs for W.
Sorry, I'm new to bootstrap (and R...): Does it make sense to treat  
the Ws like a bootstrapped variable (and calculate e.g. BCa CIs),  
because only v1..vn
are bootstrapped? If yes, how is a boot object build on the basis of  
boot$W?

I'm aware that boot() handles strata, weights and complex functions,  
but I didn't manage to get my model bootstrapped right. Hopefully, I  
can circumvent turning my
script upside down.


Daniel




More information about the R-help mailing list