Date:         Wed, 17 Aug 2011 10:35:50 -0700
Reply-To:     Charles Axibund <caxilbund@IASINFO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Charles Axibund <caxilbund@IASINFO.COM>
Subject:      Re: Syntax for rotating canonical coefficients
In-Reply-To:  <4E4B6A45.2020608@gmail.com>
Content-Type: text/plain; charset="us-ascii"

Actually, I want to rotate the coefficients not the actual canonical
variates in order to get interpretable dimensions upon which to plot the
original categorical variables.

The outstat data set has a number of structures including correlations within
categories of the class variable and the canonical structure itself. It is the
coefficients in the canonical structure I want to rotate.


Charlie Axilbund
caxilbund@iasinfo.com
On Wednesday, August 17, 2011 at 12:14:13 AM Mark Miller wrote:
> Charles,
>
> My reading of the manual suggests that you may
> be attempting to work with the wrong output file
> or at least  not the easiest to process.
>
> The OUT=  file has original variables and the
> Canonical Variable Scores as Can1 Can2 ... up to NCAN variables.
> These are centered with mean zero & pooled within variance = 1
> These could be input to FACTOR as variables and rotated any which way.
>
> The OUTSTAT= file is really a heterogenous collection of  structures
> To get at a particular structure you need to
> select based on _TYPE_ & _NAME_
>
> I expect that a correlation matrix  produced  from
> the NCAN vars in the OUT file would  be the same as
> the rows of the OUTSTAT file selected where _TYPE_='PCORR'
>
> ... Mark Miller
>
> On 8/16/2011 6:05 PM, Charles Axibund wrote:
> > According to the SAS manual under the Candisc procedure you can use Proc
> > Factor to rotate the output canonical coefficients. To quote the manual,
> > "The canonical coefficients output data set can be rotated by the FACTOR
> > procedure." So, I run my candisc procedure and output an outstat data
> > set. I input this to the factor procedure and get an errr message
> > telling me that the corr matrix read from the input data set is not
> > symmetrical. I've tried method=score, method=prin, no method at all, and
> > get the same error.
> >
> > Clearly there is some secret syntax to get the rotation to happen. I
> > can't find it in the manual anywhere. It obviously works, since it is
> > referenced in the manual. I also did it successfully several years ago,
> > but have absolutely no memory of the syntax other than that it was not
> > obvious.
> >
> > Can any one enlighten me as to how to get SAS to rotate canonical
> > coefficients?
> >
> > Thanks.
> >
> >
> > Charlie Axilbund
> > caxilbund@iasinfo.com