Dears,
many thanks for your attention.
as you said it needed to specify the constant value but different lambda is resulted with different constant values. that is the results is sensitive to c!!
how should I choose this c value?
Regards,
Rohullah
--- On Tue, 8/26/08, Lisa Yelland <[log in to unmask]> wrote:
From: Lisa Yelland <[log in to unmask]>
Subject: Re: box-cox trans
To: "Rohullah Abdullahpour" <[log in to unmask]>
Date: Tuesday, August 26, 2008, 5:31 AM
Dear Rohullah,
To apply the Box-Cox transformation, you need all the observations in your
dataset to be positive. You could add a small positive constant c to each
observation in a data step before using proc transreg. More simply, you can
specify parameter=c in the model statement within proc transreg to achieve the
same thing.
For example, if your outcome is y, you have 2 predictors (x1 and x2) and you
want to use c=0.1, then you would specify
model boxcox(y / lambda=-1 to 1 by 0.1 parameter=0.1) = identity(x1 x2);
Lisa
Quoting Rohullah Abdullahpour <[log in to unmask]>:
> Dears,
>
> I'm analysing a trait expressing in decimal between 0 to 1 without
normal
> distribution.
> I want use Box-Cox transformation,
> In SAS using PROC TRANSREG, I want to find the best value for LAMBDA,
>
> but with observation equal to zero in my variable it faces to problem
which I
> think due to the "log(0)".
>
> What can I do for this problem?
>
> many thanks.
> Rohullah
>
>
>
>
>
>
--
Lisa Yelland
PhD Candidate
Statistician, Data Management & Analysis Centre
Discipline of Public Health
University of Adelaide
SA 5005 Australia
Ph: 8303 3215
Email: [log in to unmask]
|