Date:         Tue, 30 Oct 2001 09:40:46 -0500
Reply-To:     Jay Weedon <jweedon@EARTHLINK.NET>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Jay Weedon <jweedon@EARTHLINK.NET>
Organization: http://extra.newsguy.com
Subject:      Re: dscale or pscale: Proc Genmod
Content-Type: text/plain; charset=us-ascii

On 29 Oct 01 22:26:45 GMT, wschlech@ktc.com (Warren Schlechte) wrote:

>In Genmod, you can specify overdispersed model errors with the pscale and
>dscale options.  How do you decide which one is the more proper for any
>situation?  How does one report results if the results of the hypothesis
>tests change depending upon which scale term is used?  For example, in one
>model that I examined, the follow were the results:
>
>
>                Contrast        Estimate        SE        Alpha         Confidence Limits          Chi-Sq       Prob>Chi-Sq
>NOSCALE 10K-Control     -1.0151      0.4682
> 0.05     -1.9326     -0.0975      4.70        0.0301
>
>DSCALE   10K-Control     -1.0151      0.6818      0.05     -2.3514
>0.3212      2.22        0.1365
>
>PSCALE   10K-Control     -1.0151      0.5654      0.05     -2.1233
>0.0931      3.22        0.0726

Perhaps the proof of the pudding is in the residuals. If adding an
overdispersion parameter reduces the number of large residuals, then
you should consider keeping it.

FWIW,
JW