maybe the problems of Melissa could be arise from the asymptotic distribution of the statistic when n1+n2>20. n1 =size sample1; n2= size sample2; In this example the sum of the sizes of the samples is exactly 20.

I have a question about which rank sum has to be choice when samples are the same size. When n1==n2 which rank sum we have to choice the rank sum of the sample1 or the sample2?

Do you have any news on this issue?

Thanks

Chris

First of all, thank you for sharing your code.
I have got some questions/notices concerning it:
- It seems that the line:
“print(wilcox.test(data$x[data$y==1],data$x[data$y==y]))” provides an error while looking with data with decimals.

(I just remove it and no more errors when there are decimals)

- The adjustment of the medians does not seem to work.
Below is an example of the fact that it does not work:
### adjust.median=F
x<-c(177,200,227,230,232,268,272,297)
y<-c(47,105,126,142,158,172,197,220,225,230,262,270)
siegel.tukey(x,y)
## pval : 0.9385

### adjust.median=T
x<-c(177,200,227,230,232,268,272,297)
y<-c(47,105,126,142,158,172,197,220,225,230,262,270)
siegel.tukey(x,y,adjust.median=T)
## pval : 0.9079

### by adjusting before the medians

x<-c(177,200,227,230,232,268,272,297)
y<-c(47,105,126,142,158,172,197,220,225,230,262,270)
medx<-median(x)
medymedy){
x <-x -(medx-medy)
}
if (medx<medy){
y <- y -(medy-medx)
}
siegel.tukey(x,y)
### pval: 0.09716
We do not have the same pvalues (at all as you can notice).

Just for your information, these two vectors come from a book (i can give you the reference) and the pvalue they found is 0.0976, by doing the median adjustment.

For the moment, I do not know how to correct it (sorry for that, I will do this adjustment manually before), but I can in the future propose something if you are interested of course.
By the way I’m just wondering whether this test is meaningful if we don’t adjust the medians… for me no but I may be wrong.

Best regards,

Mélissa

I’ve updated the code with your corrections.

x1 <- c(85, 106)
x2 <- c(96, 105, 104, 108, 86)
iv <- rep(1:2, c(length(x1), length(x2)))
siegel.tukey(c(x1, x2), iv, id.col=TRUE)

The rank for the middle element (104) should be 7, but it's calculated as 8. If N is the length of the combined data, this works:

TF <- rep(c(TRUE, FALSE, FALSE, TRUE), ceiling(N/4))
up <- TF[1:min(N, length(TF))]
Rup <- rank(X)[up]
Rdown <- rev(rank(X)[!up])
Rx <- c(Rup, Rdown)