Date:         Thu, 1 Oct 2009 22:50:02 -0400
Reply-To:     Mark Miller <mdhmiller@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Mark Miller <mdhmiller@GMAIL.COM>
Subject:      Re: OLS estimates on clusterred data
Comments: To: Paige Miller <paige.miller@kodak.com>
In-Reply-To:  <98d5bf58-b6d8-44fa-b98a-af9a6797e888@h13g2000yqk.googlegroups.com>
Content-Type: text/plain; charset=ISO-8859-1

The estimation problem sounds to me like a model case
of Time-Series Cross-Section Regression for which
(SAS/ETS) Proc TSCSreg has been specifically designed.

http://support.sas.com/documentation/cdl/en/etsug/60372/HTML/default/etsug_intro_sect026.htm

TSCSReg provides various and flexible modeling structures which may overlap
with
Proc Mixed, but TSCSreg was originally aimed directly at economic
models which frequently deal with time-series of cross-section data.

... Mark Miller



On Thu, Oct 1, 2009 at 3:03 PM, Paige Miller <paige.miller@kodak.com> wrote:

> On Oct 1, 2:40 pm, Tony <tony.cross...@gmail.com> wrote:
> > Thanks!
> >
> > Can i just do the following?
> >
> > PROC MIXED DATA=imputdata NOitprint;
> > CLASS year state;
> > MODEL Y= X;
> > RUN;
>
> Unless you put Year and State in the model statement, you have a
> simple linear regression of Y versus X.
>
> Even if you put Year and State in the model, this doesn't account for
> the claimed serial correlation over the years, nor does it account for
> any clustering, which I assume is different than the serial
> correlation you are referring to, but which you don't explain further.
>
> There are many different types of covariance structures possible in
> PROC MIXED, including some that deal with autocorrelation. Check the
> docs to see if one of them meets your needs.
>
> --
> Paige Miller
> paige\dot\miller \at\ kodak\dot\com
>