[R-pkgs] {CIPerm}: Computationally-Efficient Confidence Intervals for Mean Shift from Permutation Methods
Jerzy Wieczorek
j@w|eczo @end|ng |rom co|by@edu
Wed Jun 22 03:16:43 CEST 2022
Dear R users,
Emily Tupaj and I recently released a new package on CRAN:
{CIPerm}: Computationally-Efficient Confidence Intervals
for Mean Shift from Permutation Methods
https://cran.r-project.org/package=CIPerm
Problem:
It is straightforward to carry out a permutation or randomization test of
H_0: mu_A - mu_B = 0
but the "naive" approach to getting a confidence interval for the mean
shift parameter
(mu_A - mu_B)
is more computationally-intensive. In this "naive" approach we would test
H_0: mu_A - mu_B = delta
for many values of delta, and let our CI be the delta values where we
did not reject H_0. This requires running a new permutation test for
each candidate value of delta.
Our contribution:
Instead of the naive approach, our R package implements an exact
method derived by Minh Nguyen (2009) to calculate such confidence
interval endpoints from a single set of permutations. Thus, it is no
more computationally expensive than a single standard permutation
test.
Please see our package's README and `nguyen` vignette for examples and
a brief overview of the method:
https://github.com/ColbyStatSvyRsch/CIPerm#readme
https://cran.r-project.org/web/packages/CIPerm/vignettes/nguyen.html
Our `naive` vignette contains timing tests to compare the
computational speeds of Nguyen's method against the naive approach.
For large datasets, Nguyen's method is substantially faster:
https://cran.r-project.org/web/packages/CIPerm/vignettes/naive.html
For details on the methodology, please see:
Nguyen, M.D. (2009). "Nonparametric Inference using Randomization and
Permutation Reference Distribution and their Monte-Carlo
Approximation" [unpublished MS thesis; Mara Tableman, advisor],
Portland State University. Dissertations and Theses. Paper 5927.
https://doi.org/10.15760/etd.7798
Feedback is welcome by email or at:
https://github.com/ColbyStatSvyRsch/CIPerm/issues
Best wishes,
Jerzy Wieczorek
--
Jerzy Wieczorek
Assistant Professor
Department of Statistics
Colby College
5841 Mayflower Hill
Waterville, ME 04901
207-859-5841
jawieczo using colby.edu
More information about the R-packages
mailing list