The algorithm gives the "bootstrap percentile interval",
i.e. percentiles of the bootstrap distribution.
The "basic bootstrap" confidence interval, in the terminology
of Davison & Hinkley 1997, is different, and corresponds to:
2*mean(x) - rev( quantile(z, c((1-conf.int)/2,(1+conf.int)/2)))
>Here x is a numeric vector of length n with no NAs present.
>conf.int is for example 0.95 and B is the number of boostrap
>reps (default B=1000)
>
>z <- unlist(lapply(1:B, function(i,x,N)
> sum(x[.Internal(sample.index(N, N, T),
> "S_sample",T,0)]), x=x, N=n)) / n
>quantile(z, c((1-conf.int)/2,(1+conf.int)/2))
The two methods handle asymmetry and bias in diametrically opposite
ways. Neither is particularly accurate. Alternatives include the
bootstrap-t and BCa intervals.
Tim Hesterberg
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news