Ranksum Test (Mann Whitney U /Wilcoxon Test)

[Anna Car]

Hi all,

I have problems conducting the Mann Whitney U test using the ranksum v, by(catvar1) exact command. In my case the two groups are already separated into two different columns "average_rating_top" and "average_rating_mid". How can I conduct the Mann Whitney U test by comparing the values in two distinct columns instead using "by"?

Any help is very much appreciated!
Thanks a lot!!

Anna

[Nick Cox]

If the alignment in observations (rows) means something then the test does not apply.

If it means nothing, and say just arises from a dopey spreadsheet, then you can only proceed with a different data layout.


There is no data example in #1, but this may help.

clear 
input average_rating_top average_rating_mid
1  5 
2  6
3  7
4  . 
end 

stack average_rating_top average_rating_mid, into(average_rating) clear 

rename _stack which 
label def which 1 top 2 mid 
label val which which 

list 

ranksum average_rating, by(which)

[Anna Car]

Thanks a lot! Very helpful

[Anna Car]

I use the exact option to report the p-value. There is an error saying "number of observations must be <= 1,000 for exact option". I have approx 1,300 observations in each group. How can I solve this?

Help is very much appreciated.
Thanks a lot.

[Nick Cox]

What is the reported P-value? Even a tiny difference will be highly significant with those sample sizes. In fact t tests might be defensible.

If you have the same number of observations in each group, it sounds as if they are really paired.

[Anna Car]

Thanks Nick. The p-value cannot be reported. The exact option used for calculating the p-value results in the above mentioned error message. The observations are not paired and the no of observations differs across groups.

[Nick Cox]

I think a comparative graph would yield more insight.