Date:         Thu, 15 Jun 2006 13:57:44 -0400
Reply-To:     mah-j@statworks.com
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "mah-j@statworks.com" <msoobader@VERIZON.NET>
Subject:      Re: significant estimates GLIMMIX? IGNORE
Comments: To: mah-j@statworks.com
In-Reply-To:  <CBEHKHLADEBPHMJJBGAOGEIMEDAA.msoobader@verizon.net>
Content-type: text/plain; charset=iso-8859-1

Ignore previous email- I have obviously lost it- the CI of the odds ratio
includes 1 so its not significant.

-----Original Message-----
From: mah-j@statworks.com [mailto:msoobader@verizon.net]
Sent: Thursday, June 15, 2006 1:26 PM
To: SAS-L@LISTSERV.UGA.EDU
Subject: RE: significant estimates GLIMMIX?


Hi there,

I'm unclear about how to choose whether a parameter in GLIMMIX is
significant or not. For a given parameter:

The fixed effect estimate is 0.6424 and the standard error is 0.8033 and the
Pr> |t|=0.4239.
If I convert the estimate to the odds ratio- I get Odds ratio 1.9010,
confidence interval 0.3937 - 9.1785.

Using the actual estimate and Pr> |t| ( t-test for the hypothesis that the
estimate is zero)- I would conclude that this parameter is not significant
and can be dropped from the model.

However, if I use the Odds ratios and confidence intervals- (even though the
CI is very wide)- it appears that the parameter is significant and should be
kept in the model.

Any advice on how to determine whether I should keep this parameter in the
model or not will be greatly appreciated.

Thank you,
Mah-J