topic Re: Eigenvalues/vectors using correlation matrix? in SAS/IML Software and Matrix Computations

Eigenvalues/vectors using correlation matrix?

axelpuri — Wed, 03 Jun 2020 05:39:59 GMT

Hi all,

im using the following code for obtaining eigenvalues / vectors for deriving principal components .

My data-lines have a covariance matrix.

data have(type=cov); /* leave off a lot of the TYPE=COV data for simplicity */
_type_='COV';
input m1 m2 m3 m4;
datalines;
18.8767 26.8567 7.5550 -5.1533
26.8567 47.2433 5.1033 -15.2300
7.5550 5.1033 92.1900 58.8983
-5.1533 -15.2300 58.8983 48.5233
;

ods select eigenvalues eigenvectors;
proc princomp data=have(TYPE=COV) cov;
run;

proc iml;
use have; read all var _NUM_ into M; close;
eigm=eigval(m);
print eigm;

It works perfectly but what do i need to do while trying to use a correlation matrix ? when i change the cov to cor i get an error that the system is not expecting COR

So I tried the code below but this throws an error

ERROR: (execution) Matrix should be square.

ERROR: Matrix eigvals has not been set to a value.

proc iml;

sigma={1.00000 0.89933 0.18110 -0.17027
0.89933 1.00000 0.07733 -0.31809
0.18110 0.07733 1.00000 0.88061
-0.17027 -0.31809 0.88061 1.00000};

eigvals=eigval(sigma);
eigvec=eigvec(sigma);
print eigvals;
print eigvec;

Many thanks.

Re: Eigenvalues/vectors using correlation matrix?

IanWakeling — Wed, 03 Jun 2020 07:54:00 GMT

I think the issue with PRINCOMP is that the correct syntax is TYPE=CORR with two 'R's.

The error message from IML is informative since sigma is not square, actually it is a row vector. To declare a matrix you need to add commas at the end of the rows like this:

sigma={ 1.00000  0.89933 0.18110 -0.17027,
        0.89933  1.00000 0.07733 -0.31809,
        0.18110  0.07733 1.00000  0.88061,
       -0.17027 -0.31809 0.88061  1.00000};

Re: Eigenvalues/vectors using correlation matrix?

axelpuri — Wed, 03 Jun 2020 08:01:22 GMT

Thank you!!

Re: Eigenvalues/vectors using correlation matrix?

Rick_SAS — Wed, 03 Jun 2020 09:43:30 GMT

You do not need to type in the matrix. You can use the COV2CORR function in SAS/IML to rescale the covariance matrix into a correlation matrix.