limits.t(x, probs=c(25, 50, 950, 975)/1000,
df = "choose", adjust = T, z = F,
subset.statistic = 1:p,
frame.eval = x$parent.frame)
resamp.
"smaller",
"normal",
"pooled" and
"choose"
are described below.
TRUE then degrees of freedom is adjusted to reflect
extra variability caused by using a finite number of bootstrap samples.
TRUE then z-intervals (using Gaussian quantiles)
are computed instead of t-intervals; equivalent to
df = Inf and
adjust = F, and arguments
df and
adjust are ignored.
x can be found.
You need to specify this if objects can't be found by their
original names, or have changed; see
.
This does not produce what are commonly known as "bootstrap t" confidence limits--use for that. This produces t intervals using standard errors calculated using the bootstrap, jackknife, or another resampling method.
These intervals are not particularly accurate; under general conditions they are first-order accurate (coverage errors O(1/sqrt(n))), while BCa, tilting, and bootstrap t limits are second-order accurate (O(1/n)).
Among choices for
df,
"smaller" is the most conservative;
it sets the degrees of freedom to the smallest
sample or group size, minus 1. The least conservative is
"pooled";
it assumes the same within-group variance within each sample
(for two-sample problems created using
bootstrap2,
and within each group (across both samples) when sampling by group
(stratified sampling).
A compromise is
"normal"; it assumes the same within-group variance
within each group, but allows the variance to differ between two samples
for
bootstrap2. In the absence of strata, this uses the same basic
calculation for degrees of freedom and
t.test does when
var.equal==FALSE.
The default
"choose" selects
"smaller" if sampling by group
and
"normal" otherwise.
x <- rt( 100, df=6)
boot <- bootstrap(x, c(mean=mean(x),
trim=mean(x, trim=.2), median=median(x)), B=100)
limits.t(boot)
# Can also be used with jackknife, influence, and other "resamp" objects.