NEWS | R Documentation |
Added unconditional difference option for the statistic for the sign test. This calls ncnemarExactDP, that is why the package now requires exact2x2 (>=1.6.4).
added utol argument to meldCD.
Changed wmwTest so that it gives a warning if the uniroot function may fail for method='exact.mc'. See Note section on the man page.
Added meldtTest and meldCD functions to do two-sample tests by melding.
Added tukeyWelsch function to do multiplicity adjusted pairwise tests from k-samples
Added anovaOneWay, to output an ICC or other estimate, also allow var.eqaul=FALSE using the Brown-Forsythe method.
Added ama1c1cpg data set.
Fixed wmwTest when exact and there are tied values and value is most extreme. For example, wmwTest(c(1,3),c(3,4)) previously had given an error for calculating the confidence interval.
use match.arg with method in wmwTest, so that method='asy' will work for method='asymptotic'
updated wmwTest reference
added output to bfTest so it outputs the R statistic.
Added wmwTest to calculate the Wilcoxon-Mann-Whitney test with confidence intervals for the Mann-Whitney parameter that are unified with the test.
Added the abcnonHtest function to give nonparametric ABC (approximate bootstrap confidence) intervals with p-values and estimates.
Fixed error in one-sided tests for wspoissonTest when midp=TRUE. Thanks to Michael Nelson for pointing out this error. Also fixed data description for wspoissonTest.
Created bfTest function to perform the Behrens-Fisher test using the Behrens-Fisher solution with p-values and CIs.
Made changes to the estimate and confidence interval for wsrTest
.
The default estimate previously used the estimator of the pseudo-median from the difference using
the estimator from wilcox.test
.
The new estimator is based on the shift value where the one-sided p-values are equal. This ensures that the
estimate will be within the confidence interval and will not change as the null hypothesis changes.
Updated citation for Fay and Kim (2017).
Corrected error in wsrTest
. The p-value was incorrectly calculated when mu was not 0, now it is correct. Also added the tieDigits option.
Updated citation for Fay and Kim (2017).
Modified wspoissonTest
to be calculated using uniroot by default, instead of by Monte Carlo simulation.
First CRAN release of asht, “Applied Statistical Hypothesis Tests” by Michael P. Fay (author, maintainer).
signTest
gives exact sign test with confidence intervals
cvTest
gives coefficient of variance tests based on normality or lognormal assumptions
quantileTest
and medianTest
give exact tests on a quantile (including median). No assumptions on the distribution are needed (ordered factors are allowed).
var1Test
gives one-sample test of normal variance.
wsrTest
gives exact Wilcoxon signed rank test, gives exact matching confidence intervals.
wspoissonTest
gives confidence intervals for weighted sums of Poisson, useful for directly standardized rates.