rankdifferencetest provides a modified Wilcoxon signed-rank test which produces consistent and meaningful results for ordinal or monotonically transformed data. The procedure was described by Diana Kornbrot in https://doi.org/10.1111/j.2044-8317.1990.tb00939.x
# Install from CRAN
install.packages("rankdifferencetest")
# Or the development version from bitbucket
remotes::install_bitbucket("bklamer/rankdifferencetest")# See ?rdt and ?kornbrot_table1
library(rankdifferencetest)
res <- rdt(
data = kornbrot_table1,
formula = placebo ~ drug,
conf_level = 0.95,
alternative = "two.sided",
distribution = "asymptotic",
zero_method = "wilcoxon",
correct = TRUE
)
res
#>
#> asymptotic Kornbrot-Wilcoxon rank difference test
#>
#> p = 0.262
#> Z = 1.12
#> Paired differences of ranks: placebo - drug
#> Alternative hypothesis: True location shift of paired
#> differences of ranks is not equal to 0
#>
#> Pseudomedian: 1.5
#> 95% CI: -2.5, 6Copyright: Brett Klamer - 2021 - GPL-3 (https://opensource.org/license/gpl-3-0)