2

I am not a regular R user so I'm struggling a bit on how to get an exact p-value and confidence limits for a single sample using the Wilcoxon Signed-Rank test. wilcox.test does not handle ties so it's automatically defaulting to normal approximation, but I have very small samples I am dealing with so need an exact estimate.

What is the equivalent of wilcox.test(y, alternative = c("greater"), correct = TRUE) for a single vector in wilcox_test from the coin package?

This is a paired sample problem...

I get all kinds of errors even when I follow the examples from the documentation

data("ToothGrowth")
df <- ToothGrowth

df %>% wilcox_test(len ~ 1, mu = 0)


Error in UseMethod("wilcox_test") : 
  no applicable method for 'wilcox_test' applied to an object of class "data.frame"
Ryan Boch
  • 123
  • Let's start with the simple stuff: have you loaded the coin package into your R environment? – whuber Aug 07 '20 at 17:19
  • 2
    The coin interface is made for the paired situation with two vectors. If you already end up with a vector of differences, you can invent a vector of 0 to play the role of the second vector. This is a bit a hacky solution, but it works: library(coin); x <- -3:10; zeros <- rep(0, length(x)); wilcoxsign_test(x ~ zeros, distribution = "exact") (sorry for the formatting, but answering a closed question is not possible for me). – Michael M Aug 07 '20 at 18:54
  • 1
    @whuber Yes :) Should've mentioned I'm a very proficient python programmer, I just don't use R that often. – Ryan Boch Aug 07 '20 at 19:24

0 Answers0