Calculate t confidence intervals

DESCRIPTION:

Calculate t confidence intervals using standard errors from resampling

USAGE:

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) 

REQUIRED ARGUMENTS:

x
object of class resamp.

OPTIONAL ARGUMENTS:

probs
probabilities for one-sided confidence limits; e.g. c(.025, .975) gives a two-sided 95% confidence interval. Note that values are undefined at 0 and 1.
df
degrees of freedom to use for the test, may be numerical. Other choices "smaller", "normal", "pooled" and "choose" are described below.
adjust
logical flag, if TRUE then degrees of freedom is adjusted to reflect extra variability caused by using a finite number of bootstrap samples.
z
logical flag, if 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.
subset.statistic
subscript expression; if the statistic that was bootstrapped has length greater than 1, use this to request intervals for only some elements (parameters) of the statistic.
frame.eval
frame where the group argument (if any) used when creating x can be found. You need to specify this if objects can't be found by their original names, or have changed; see .

VALUE:

matrix of confidence limits with one row per parameter.

DETAILS:

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.

SEE ALSO:

, , , , .

EXAMPLES:

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.