Date:         Tue, 13 Feb 2007 22:42:03 -0800
Reply-To:     David L Cassell <davidlcassell@MSN.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David L Cassell <davidlcassell@MSN.COM>
Subject:      Re: MIXED: TYPE = VC/UN
In-Reply-To:  <Pine.GSO.4.58.0702121712500.3888@leukothea>
Content-Type: text/plain; format=flowed

kviel@EMORY.EDU wrote:
>
>Does the TYPE = VC option to the RANDOM statement constrain the covariance
>parameters (tij, where i ne j) to zero?

Yes.

>                                        If so, unless that is known a
>priori, wouldn't an analyst want to use TYPE = UN?

Sometimes yes, sometimes no.

Was that vague enough for you?  :-)

If you have *no* information about the structure of the covariance
matrix, TYPE=UN is worth considering.

But lots of times we have prior information about the covariance
matrix, so we have something more concrete to start out with.
(Some of the time, that information is wrong, of course.)

>Can anyone offer guidance?  I am aware that the processing time is more
>intensive for UN.

It's not bad.  Unless you have huge data sets.

>While I am at it, should one use the FAO(2) and CHOL (GLIMMIX) in the
>cases where G might not be positive definite?
>
>Thanks,
>
>Kevin

I would first recommend an effort to figure out *why* your G isn't
positive definite.

HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
Laugh, share and connect with Windows Live Messenger
http://clk.atdmt.com/MSN/go/msnnkwme0020000001msn/direct/01/?href=http://imagine-msn.com/messenger/launch80/default.aspx?locale=en-us&source=hmtagline