Date:         Wed, 20 Sep 2000 07:20:09 -0400
Reply-To:     John Jones <jonesj@PHARMARESEARCH.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         John Jones <jonesj@PHARMARESEARCH.COM>
Subject:      Re: F-Test vs T-Test
Comments: To: Zubin <binabina@MINDSPRING.COM>
In-Reply-To:  <8q96ce$1e8$2@slb3.atl.mindspring.net>
Content-Type: text/plain; charset="us-ascii"; format=flowed

I would not ignore the result of the t-test.  Sometimes meaningful effects
can be swamped by the non-meaningful effects in a model.  You gotta know
your data too though.  Consider how were the dummy indicators
defined?  Does that make sense?  Also, you may have by mere chance stumbled
onto something "statistically" significant but does it have "practical"
significance?  Can you interpret the result and is the result meaningful?

At 10:13 PM 9/19/00 -0400, Zubin wrote:
>I have run a regression where i have used dummy values to quantify a nominal
>variable.  Lets say 4 levels, therefore 3 dummies.  One of the t-tests on
>the dummies is highly significant, however the f-test testing if b1=b2=b3=0
>is indicating that as a group they are not significant.  What is the proper
>interpretation?  Does the F-Test take precedence over the T?