bootstrap
objects.
This calculates percentiles and BCa confidence limits
for the bootstrap distributions of parameters, and correlations of
parameter replicates if applicable. The print method displays the same
results as
print.resamp
, plus these quantities.
summary.bootstrap(object, probs=c(0.025, 0.05, 0.95, 0.975), frame.eval = <<see below>>, narrow = F, ...)
object
can be found.
You need to specify this if objects can't be found by their
original names, or have changed; see
.
narrow=FALSE
(the default), then intervals are wider and coverage probabilities
are usually more accurate than with
narrow=TRUE
.
z0
,
acceleration
, and
group.size
), or
(for example,
t.adjustment
,
initial
).
object
(
call
,
B
,
observed
,
estimate
), plus percentiles
limits.emp
,
BCa confidence intervals
,
and (if
observed
has more than one element)
the correlation matrix
correlation
for the bootstrap replicates.
Tilting limits,
,
are also computed, if the bootstrap statistic is
"functional"
(the
observed value is the same as the value computed with equal weights)
and if either the bootstrap object contains component
L
(empirical
influence values) or if the size of the original bootstrap sample is
less than 100.
The print method
prints the results in a nicely formatted manner.
To get percentiles of the bootstrap distribution without computing
BCa confidence intervals, call
rather than
summary
.
If
acceleration
is estimated using
jackknife
,
the statistic must allow samples of reduced size; see
.
# statistic is not functional, tilting limits not computed summary(bootstrap(stack.loss, var)) # statistic is functional, tilting limits are computed summary(bootstrap(stack.loss, var(stack.loss, unbiased = F)))