Date:         Tue, 17 Jun 2003 21:47:20 -0400
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: proc glm vs proc mixed help
Content-Type: text/plain; charset=us-ascii

On 17 Jun 03 15:16:11 GMT, raypass@ATT.NET (Ray Pass) wrote:

>I'm sending this out for a colleague.  I am NOT a statistician, but I know
>where to go for help.

>>     Does anyone know how to calculate the confidence interval for the
>>estimate using proc glm?  I see how to calculate for the means by adding
>>to the model statement, but not for the estimate.
>>
>>     This is what I have so far:
>>
>>title2 "GLM p-values for &var";
>>proc glm data=gmean (where=(trt in (0,1))) ;
>>    class trt;
>>    model l&var = trt /solution  ss3 ;
>>    estimate "ratio" trt -1 1;
>>run;
>>
>>
>>It's easy in proc mixed as there is an option to do so.  See below.
>>However, biostat doesn't want me to use proc mixed.
>>
>>
>>title2 "Proc Mixed p-values for &var";
>>proc mixed data=gmean ;
>>   where trt in (0,1);
>>  class trt;
>>  model l&var = trt;
>>  estimate "Ratio" trt -1 1 / cl ;
>>  make "table" out=ratio estimate;
>>run;
>>
>>Thanks for any tips you can give me.

"Biostat" can verify the following by RTFM for PROC MIXED, but AFAIK
the method used by MIXED is the standard large-sample method:

Compute the t-statistic associated with the desired confidence level
and the error df, and multiply this by the provided SE for the
estimate. Then add/subtract this to/from the estimate value.

Example:

Estimate=4.8447     SE=1.7570     Error df=207.

2-tailed t-statistic for 207 df @ 95% CL = 1.9715.

1.9715*1.7570 = 3.4639.

CI = 4.8447 +/-  3.4639 = [ 1.3808,  8.3086 ]

JW