# question with correlation

6 views (last 30 days)
Mate 2u on 11 May 2012
I have a array of X x Y size.
I need to do correlation of all the columns as combinations. So need to do correlation of each column with other columns and then once I have all the correlations of all combinations, I want to take a average.
Thanks

Richard on 11 May 2012
Not sure what you are asking, but this may help.
a = 1;
b = 20;
data = a + (b-a).*rand(20,4);
R = corrcoef(data);
RPairs = [nchoosek(1:size(data,2),2), nonzeros(tril(R,-1))];
Where data is your matrix, R is the correlation between each pair of columns, and RPairs represents the column numbers in RPairs(:,1:2) and the corresponding correlation coefficient in RPairs(:,3). Then if you want to calculate the average correlation you simply take the mean of the third column:
MeanR = mean(RPairs(:,3))
Hope this helps
Mate 2u on 11 May 2012
Error using horzcat
CAT arguments dimensions are not consistent.

Mate 2u on 11 May 2012
May I should of rephrased what I meant. When I have a correlation matrix....using corrcoef, how can I take average of the bottom left corner or top right corner?
Richard on 11 May 2012
This can be achieved with the script I provided above.
nonzeros(tril(R,-1))
This will find the lower triangular part of the correlation matrix, or as you say bottom left corner, 'nonzeros' then removes the zeros from the matrix which is created by 'tril' thus genberating a vector of these correlation values. Does this make sense?