Ranksum - Wilcoxon rank sum test - Get more statistics using a for loop

Hello everyone!

I am running some statistics and I need to use the Wilcoxon rank sum test.

I have 2 matrices and I need to compare 25 vector pairs.

If I run the test in a for loop, I get only the p-values - I would love to get more statistics but I have not been able to find a way.

That is what I have now:

Matrix1 = n x 25
Matrix2 = m x 25

for i = 1:25
    testresults(i) = ranksum(Matrix1.(i),Matrix2(i),'method','exact')
end

If I run the test like this, I will get as a results just the pval, any clue how to get [p,h,stats] in a for loop?

I know how to run this out of the for loop.

Thank you all in advance for your help!

Answer 1

[Adam Danz]

[p,h,stats] = ranksum(x,y)

p = nan(1,n); 
h = false(1,n); 
stats = cell(1,n); 
for i = 1:n
    [p(i),h(i),stats{i}] = ranksum(x,y,__)
end

[Clemente Tagliafierro]

Thank you very much Adam, is there a way to name the result matrix?

Background is that I will need to perform the same test multiple times for different for loops and I guess the results are going to be overwritten everytime.

Can I just call p, h, and stats differently everytime, or it is important that the name stays p, h, and stats?

[Adam Danz]

Two idea.

1) you could a version of my answer into its own function so you can store the outputs in their own variables.

% very rough example. 

[p1,h1,stats1] = localRankSum(x1,y1);
[p2,h2,stats2] = localRankSum(x2,y2);
[p3,h3,stats3] = localRankSum(x3,y3);

function [p,h,stats] = localRankSum(x,y)
p = nan(1,n); 
h = false(1,n); 
stats = cell(1,n); 
for i = 1:n
    [p(i),h(i),stats{i}] = ranksum(x,y,__)
end

2) you could add a nested loop

% very rough example

p = nan(m,n); 
h = false(m,n); 
stats = cell(m,n); 
for i = 1:n
    for j = 1:m
        [p(i,j),h(i,j),stats{i,j}] = ranksum(x,y,__);
    end
end

[Clemente Tagliafierro]

Thank you, Adam - Really helpful

[Adam Danz]

Glad I could help!

The critical lesson to learn is to avoid creating a bunch of different variables that store the same type of data when the data can be stored in a vector / matrix / array. Indexing is the superpower of Matlab and nearly all of its functions are designed to operate on arrays in a fast and efficient manner.

[Yogesh Kumkar]

Thanks Adam,

I have similar code but nested for loop never finishes.

% CTL2008E1:(177,162)
% Xrel:(177,162,15)
% Yctl:(177,162,15)
% Zidl:(177,162,15)
% ranksum(Xrel(:,:,1:15),Yctl(:,:,1:15))
pRC = nan(size(CTL2008E1));
hRC = false(size(CTL2008E1));
statsRC = cell(size(CTL2008E1)); 
pIC = nan(size(CTL2008E1));
hIC = false(size(CTL2008E1));
statsIC = cell(size(CTL2008E1));
% do only for non NaN pixels.
for i = 1:size(CTL2008E1,1)
    for j = 1:size(CTL2008E1,2)
        while ~isnan(CTL2008E1(i,j))
            [pRC(i,j),hRC(i,j)] = ranksum(squeeze(Xrel(i,j,:)),squeeze(Yctl(i,j,:)));
            [pIC(i,j),hIC(i,j)] = ranksum(squeeze(Zidl(i,j,:)),squeeze(Yctl(i,j,:)));
        end        
    end
end

I am running on Macbook pro 6 cores. Is it possible to use multicore? (2,6 GHz 6-Core Intel Core i7, 16 GB 2400 MHz DDR4)

Thank you,Y

[Adam Danz]

The loop gets stuck in the while-loop at the first non-NaN value in CTL2008E1.

I think you want a condition rather than a while-loop.

        if ~isnan(CTL2008E1(i,j))
            [pRC(i,j),hRC(i,j)] = ranksum(squeeze(Xrel(i,j,:)),squeeze(Yctl(i,j,:)));
            [pIC(i,j),hIC(i,j)] = ranksum(squeeze(Zidl(i,j,:)),squeeze(Yctl(i,j,:)));
        end 

[Yogesh Kumkar]

Thanks! good catch!