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.