Wilcoxon ranksum on left and right data

[Becky Marks]

Hi all,

To begin, I am completely new to using Stata - this is my first use of a statistical program ever, and I did not get good instruction. So I know I am just using the program incorrectly, and need help.

I am trying to run a Wilcoxon rank sum test on nonparametric data. I have assessed my values - measurements of pressure - in left and right fingers. I am attempting to see if there is a significant difference.

Here is an example of the data:

ID	Left	Right
1	24	22
2	16	19
3	27	21
4	13	15
5	19	23
6	21	21
7	22	17
8	25	23
9	25	18

My questions are these:
How exactly do I enter this data into the data editor so that I can run a Wilcoxon ranksum? Right now it tells me I have more than 2 groups, and I can only pick left or right, or it makes me pick the person ID.
And then - what do I use for the main and by/if/in tabs?

I am horribly confused and at my wit's end - any help is greatly appreciated.

Thanks,

Becky

[Clyde Schechter]

You should not do a Wilcoxon rank-sum test on this data because it consists of matched pairs. The Wilcoxon rank-sum test is for independent samples. You should do the -signrank- test instead.

signrank left = right

[Marcos Almeida]

Hello Becky,

Welcome to the Stata Forum / Statalist.

Please read the FAQ. There, you will find great information on how to post data and start a query. Thanks.

That said, you seem to need the Wilcoxon signed-rank test, instead of the ranksum.

For this, you shall type:

. signrank Left = Right

Edited: crossed with Clyde's reply, on the same verge.

[Bruce Weaver]

I agree with Clyde (in #2): It looks like paired data, so you want Wilcoxon's signed ranks test. As far as I can tell, signrank gives only the large-sample (asymptotic) z-test. With your small sample size, you ought to use the exact test, which can be obtained via signrankex. Here is the command to install it:

net install st0297.pkg

HTH.

[Becky Marks]

That helps significantly, thank you. It is much easier when I am running the correct test. Thank you for your help today.

[Nick Cox]

Not disagreeing with any advice so far, but why assume that you have "nonparametric data" (procedures may be nonparametric, but not data)?. I'd expect pressures to be fairly well-behaved with at most some benefit from transformation. So, I would look at the distribution of the differences and an appropriate t test too.