Non-parametric confidence intervals using permutation tests
This package implements the methodology described in https://arxiv.org/pdf/2111.14966.pdf
The general univariate case (ie. a model and a test that satisifes the conditions of section 2.1 in the paper) can be handled by calling ciperm/ciperm0 directly.
Similarly, the general multivariate case in handled by calling ciperm/ciperm0.
ciperm0 and ciperm0.multi: The workhorses that perform the actual permutation scheme
ciperm: Computes the confidence interval
(ie. by using quantiles from ciperm0)
ciperm.multi: Computes the confidence interval.
Optionally calculates the joint confidence level and adjusted confidence intervals.
ciperm.twosample:
User-friendly function for computing the two-sample confidence interval.
ciperm.linreg:
User-friendly function for computing confidence interval for the slope in linear regression.
alpha.multi: Computes the joint confidence level (from output of ciperm0.multi)
adjusted_ci: Simple bisection algorithm that computes adjusted confidence intervals (from output of ciperm0.multi)
The recommended way is to use the devtools package, e.g. run devtools::install_github("naolsen/ciperm") from the R interface.
As the package is only based on R code, no special compilers are needed.
Not that the package has not been optimised for speed, but uses "crude" tools like uniroot.
Furthermore, following the remarks in the article, there is an O(2^K) cost of calculating the adjusted confidence level where K is the number of dimensions/coordinates.
Please fell free to contribute to the package.