This may seem to be a simple task, however I am new in the field of statistics. I am interested in creating an MxN value matrix X(where M rows can be taken as the attributes or values and N columns are considered the instances that you are testing) such that the correlation matrix of such MxN values C, (which will be of dim NxN) will be extremely sparse. I am just looking to do this for random data generation and testing. Therefore I while most of the data is to be random, I do want to have a few of the instances (col) in X have controled dependency. I have been guided to try to use conditional probability, however thus far have not been able to understand how this will help me. I am overall a little confused on what eventually creates a "sparse correlation matrix" because it seems to me that if I created streams of random numbers, and set a few col dependent on one another this would create a sparse matrix. However upon review of correlation functin, I can see that since the col are extremely different they are turning up with a rather dense correlation matrix. Any help would be greatly appreciated. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
