Date: Fri, 26 May 2006 09:51:49 -0500
Reply-To: Bryan <bryan.groups@HOTMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Bryan <bryan.groups@HOTMAIL.COM>
Subject: Re: Statistical question
Content-Type: text/plain; format=flowed
David's right on target. If you want more info there's more here on the
UCLA stat's page:
You might look especially at the example in section 2.3.
Now if you're supervisor is totally sold on the idea of transforming and
smearing that's a different matter : )
>From: David L Cassell <davidlcassell@MSN.COM>
>Reply-To: David L Cassell <davidlcassell@MSN.COM>
>Subject: Re: Statistical question
>Date: Thu, 25 May 2006 22:42:40 -0700
>>I got a question from my supervisor, But I do not know how to answer.
>>here is the question:
>> "Say you log transform a dependent variable that is not normal and then
>>perform regression. Maybe the dependent variable is $. Then to interpret
>>the results, you have to retransform the log back to $. Articles talk
>>using a smearing factor. What is a smearing factor and how is it
>>anyone can help me with the answer or point me to some reference.
>I see Robin has already given you a good answer.
>So let me just kvetch a bit.
>The distribution of Y has nothing to do with the problem. Y may be
>horribly distributed: a long tail, or huge outlying points, or anything.
>And it may be *fine* for the regression model. Because the values
>of your regressors may be making Y look like that.
>What matters is the distribution of the *residuals*. Not the distribution
>of Y. If the residuals are non-normal, then you need to fix things.
>But if you decide to fix things by transforming Y, or some of the X's,
>or both, then you should be using PROC TRANSREG and letting it
>do all the dirty work for you.
>David L. Cassell
>3115 NW Norwood Pl.
>Corvallis OR 97330
>Express yourself instantly with MSN Messenger! Download today - it's FREE!
Donít just search. Find. Check out the new MSN Search!