Date:         Mon, 27 Oct 2008 08:40:03 -0400
Reply-To:     Peter Flom <peterflomconsulting@mindspring.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject:      Re: To statisticians: Units,
              standardization and regression modelling
Comments: To: "cat.." <cat.b41@GMAIL.COM>
Content-Type: text/plain; charset=UTF-8

"cat.." <cat.b41@GMAIL.COM> wrote

>Let's say I want to fit a logistic model Logit P(diabetis) = Age +
>Glucose.
>
>I usually do not ask for standardizing the covariates before entering
>them into the model, only centering them, just to make the
>interpretation of exp(beta) easier (= OR when the covariate is
>incremented by 1 <unit used in the model>). Centering the covariate
>impacts the value of the intercept, reducing them will impact the
>value of their coefficient in the model.
>
>But now, I have a doubt: What about the units ? What senses does it
>make to add years and mg/dl ?
>
>Any thought about that ?
>
>What is your practice in modeling ? Do you always only work with
>standardized variables ?
>


I rarely standardize, because usually the units themselves make more sense than the standardized units.

Does it make sense to add years?  Well, I don't know much about diabetes, but it makes sense to me.
Does it make sense to add mg/dl?   I think here it depends on the unit and their range.  I have no idea what the range of this variable is.  But if the range is very large, it might make sense to divide it by 100 or 1000 or something.  This would be similar to using "income per year" as a covariate.  In most cases, one would not want to look at this in dollars, but in thousands of dollars.


Peter

Peter L. Flom, PhD
Statistical Consultant
www DOT peterflom DOT com