University of Minnesota, Twin Cities School of Statistics
There are three papers on this subject:
Fuzzy and Randomized Confidence Intervals and P-values
by Charles J. Geyer and Glen D. Meeden
To appear (with discussion) in Statistical Science
and
Fuzzy P-values in Latent Variable Problems
by Elizabeth A. Thompson and Charles J. Geyer
and
Fuzzy P-values and Ties in Nonparametric Tests
by Charles J. Geyer
The computing for Geyer and Meeden (2005)
was done with a somewhat primitive R package ump
.
Now (at version 0.5) it only does UMPU tests and confidence intervals,
fuzzy and randomized, for the binomial distribution.
The source tarball ump_0.5-1.tar.gz is available here, but the package is better installed from CRAN under contributed extension packages.
The computing for Geyer (submitted)
was done with an R package fuzzyRankTests
.
Now (at version 0.3) it only does fuzzy-exact tests and confidence intervals,
that are the competitors of the classical sign test, Wilcoxon signed rank
test, Mann-Whitney-Wilcoxon rank sum tests, and their associated confidence
intervals.
The source tarball fuzzyRankTests_0.3.tar.gz is available here, but the package is better installed from CRAN under contributed extension packages.
The R package can be tried out on line using Rweb.
The computing for Thompson and Geyer (2005) was done in the computer package MORGAN (in an experimental, not yet released version, that will eventually be released).
Overheads for talk given at University of Minnesota and at the JSM in San Francisco as PostScript or as PDF.
Overheads for talk given at the Fred Hutchinson Cancer Research Center in Seattle as PDF
Overheads for talk given June 23, 2005 at WNAR in Fairbanks as PDF
Fuzzy confidence intervals and P-values, computed using the
R package fuzzyRankTests
are being taught in
Stat 5601
(master's level service course on nonparametrics also taken by undergrads).
See web pages on
fuzzy
P-values and confidence intervals, the
sign
test and associated procedures, the Wilcoxon
signed
rank test and associated procedures, the Mann-Whitney-Wilcoxon
rank
sum test and associated procedures.
Fuzzy confidence intervals and P-values, computed using the
R package ump
are being taught in
Stat 5021
(master's level service course on basic statistics).
See the handout
A fuzzy binomal
confidence interval.
An example of fuzzy confidence intervals used as error bars that didn't fit in Geyer and Meeden (2005). This example was done in Sweave. The Rnw (R noweb) source is available.
The discussants for the paper by Geyer and Meeden (2005) proposed (or perhaps found in old literature) two fuzzy confidence intervals for the binomial distribution that are different from the UMPU intervals proposed by Geyer and Meeden (2005).
An example of these three fuzzy confidence intervals compared clarifies the agreement and disagreement between Geyer and Meeden (2005) and their discussants (who are cited therein).