How to produce a data set in MATLAB in which variables will correlate to a prespecified level?

Question

I'm looking to create a data set with three columns and an arbitrary number of rows.

I'd like column 1 to have a Pearson correlation .20 with column 2, column 1 to correlate .24 with column 3, and column 2 to correlate .3 with column 3.

How do I do produce this?

score 0 · Answer 1 · answered May 14 '13 at 04:03

0

you can generate multivariate guassian using mvnrnd and specify covariance to achieve the desired pearson correlation.

see the following documentation: http://www.mathworks.com/help/stats/mvnrnd.html

answered May 14 '13 at 04:03

Cici

1,407
3
13
31

I understand that in r = mvnrnd(MU,SIGMA,cases) I should set cases to 100. However, I do not know what to set MU and SIGMA to. – user1205901 - Слава Україні May 14 '13 at 04:25
When testing out SIGMA = [1 1.5 2; 1.5 1 3 ; 2 3 1 ]; and mu = [3 3 3]; I get the error message "SIGMA must be a symmetric positive semi-definite matrix." – user1205901 - Слава Україні May 14 '13 at 04:41
1

Try this -- `sigma=[1 .2 .24; .2 1 .3; .24 .3 1]; mu=[0,0,0]; X=mvnrnd(mu,sigma,1000); corr(X)`. Is that what you want? – Chris Taylor May 14 '13 at 07:31
That works well although the Pearson correlations came out to be .21, .24 and 34. Not that that was very important, but do you know why that happened? – user1205901 - Слава Україні May 14 '13 at 08:31
the sigma is variance covariance matrix. see this for an explaination http://www.randomservices.org/random/expect/Covariance.html – Cici May 15 '13 at 14:55

How to produce a data set in MATLAB in which variables will correlate to a prespecified level?

1 Answers1