Date: Thu, 4 Feb 1999 16:12:33 -0600
Reply-To: Gullion Christina <Christina.Gullion@COLUMBIA.NET>
Sender: "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From: Gullion Christina <Christina.Gullion@COLUMBIA.NET>
Subject: Re: IML
I've done monte carlo studies of number of factors criteria and
nonorthogonal rotation criteria in factor analysis using SAS matrix
language. Most of the criteria and rotation methods were and are not
available in SAS PROC FACTOR. I needed to do matrix multiplication,
inversions, rescaling (normalization) to column and or row sums of squares.
All are simple in IML because it operates on whole rows or columns with
equal ease. It can take a lot less programming to format printing large
matrices too, e.g., partitioning a 100 x 100 correlation matrix together
with its 100-item variable name vector to give row and column labels into
page-size chunks. It's also rather handy for inputting the lower triangular
of a symmetric matrix, then getting the full matrix with a simple transpose
and matrix addition. And finally, if you want to do anything statistical
that SAS hasn't implemented yet and that involves >1 variable, you might
find it's a lot easier to do in IML than a data step. I've seen IML
routines for doing survival analysis with time-dependent status, robust
regression, power analysis, etc. Or you could test a SAS PROC by writing
your own IML routine and verifying that it's handling certain feature of the
data the way you expect it to. It's an all-around fun product. Too bad
it's not in SAS/STAT.
Christina M. Gullion, Ph.D.
Biostatistician and Supervisor, Clinical Research
Department of Clinical Research
Medical City Dallas Hospital
> -----Original Message-----
> From: hongjie Wang [SMTP:hwang@ZSASSOCIATES.COM]
> Sent: Thursday, February 04, 1999 8:49 AM
> To: SAS-L@UGA.CC.UGA.EDU
> Subject: IML
> Dear Friends, Can any one offer me some insights on the usefulness of IML?
> Other than its matrix operations, does it offer anything else?
> In lots of textbooks and tutorials on SAS, I notice that IML is introduced
> in the context of regression.
> Authors show how one can implement the regression using IML directly,
> instead of proc reg.
> I sense it must have some other features that can not be accomplished in
> base sas.
> Your input is greatly apprecaited.