CIPerm (version 0.2.3)

cint: Permutation-methods confidence interval for difference in means

Description

Calculate confidence interval for a simple difference in means from a two-sample permutation or randomization test. In other words, we set up a permutation or randomization test to evaluate \(H_0: \mu_A - \mu_B = 0\), then use those same permutations to construct a CI for the parameter \(\delta = (\mu_A - \mu_B)\).

Usage

cint(dset, conf.level = 0.95, tail = c("Two", "Left", "Right"))

Arguments

dset

The output of dset.

conf.level

Confidence level (default 0.95 corresponds to 95% confidence level).

tail

Which tail? Either "Two"- or "Left"- or "Right"-tailed interval.

Value

A list containing the following components:

conf.int

Numeric vector with the CI's two endpoints.

conf.level.achieved

Numeric value of the achieved confidence level.

Details

If the desired conf.level is not exactly feasible, the achieved confidence level will be slightly anti-conservative. We use the default numeric tolerance in all.equal to check if (1-conf.level) * nrow(dset) is an integer for one-tailed CIs, or if (1-conf.level)/2 * nrow(dset) is an integer for two-tailed CIs. If so, conf.level.achieved will be the desired conf.level. Otherwise, we will use the next feasible integer, thus slightly reducing the confidence level. For example, in the example below the randomization test has 35 combinations, and a two-sided CI must have at least one combination value in each tail, so the largest feasible confidence level for a two-sided CI is 1-(2/35) or around 94.3%. If we request a 95% or 99% CI, we will have to settle for a 94.3% CI instead.

Examples

# NOT RUN {
x <- c(19, 22, 25, 26)
y <- c(23, 33, 40)
demo <- dset(x, y)
cint(dset = demo, conf.level = .95, tail = "Two")
# }