significant correlation values

8 views (last 30 days)
Richard
Richard on 25 Apr 2012
How is it possible to test whether the correlation between 2 time series is significant?
For example:
clear all
a = 1;
b = 20;
Fieldnames = {'data1','data2','data3','data4'};
data1 = a + (b-a).*rand(100,1);
data2 = a + (b-a).*rand(100,1);
data3 = a + (b-a).*rand(100,1);
data4 = a + (b-a).*rand(100,1);
NewData = [data1,data2,data3,data4];
R = nonzeros(tril(corrcoef(NewData),-1));
R_values = [Fieldnames(nchoosek(1:length(Fieldnames),2)) num2cell(R)];
How would I find if the correlation values were significant?

Answers (1)

Wayne King
Wayne King on 25 Apr 2012
You can return the p-values from corrcoef
a = 0;
b = 1;
data1 = a + (b-a).*rand(100,1);
data2 = a + (b-a).*rand(100,1);
data3 = a + (b-a).*rand(100,1);
data4 = a + (b-a).*rand(100,1);
NewData = [data1,data2,data3,data4];
[r,p] = corrcoef(NewData);
% row and column indices of significant correlations
[I,J] = find(p<0.05);
Of course this is not giving the correlation for anything but at zero lag. Often in time series analysis you assume that there may be correlation between two time series if one is lagged with respect to the other.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by