Date: Mon, 14 Jun 2010 10:49:29 -0400
Reply-To: "Viel, Kevin" <kviel@SJHA.ORG>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Viel, Kevin" <kviel@SJHA.ORG>
Subject: Re: when to do separate analysis for gender
Content-Type: text/plain; charset="us-ascii"
> -----Original Message-----
> From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Myung
> Sent: Monday, June 14, 2010 8:42 AM
> To: SAS-L@LISTSERV.UGA.EDU
> Subject: when to do separate analysis for gender
> Hello everybody,
> I have a question on when we need to do separate analysis and when to do
> combined analysis.
> To decide this, we have done descriptive analysis by gender and test many
> interaction by gender term as well.
> But the results were not straightforward for that decision.
> I prefer to do separate analysis complemented by gender interaction. This
> because although combined analysis is able to present estimates for sex
> variable, which separate analysis cannot, it doesn't allow comparison
> between gender with values. But others in my group think differently.
> So I would like to your opinion and possibly a reference on this issue,
> Any comments will be appreciated.
We need much more information. For instance, your model and a brief description of study design. The "interaction" might not sufficiently be handled by inclusion of the term "sex*exposure". How you code the terms could be a problem.
I think separate analyses are justifiable as the more conservative approach. Remember, that you can model a circle with a triangle, but the fit might not be the best. Sex as a dichotomous variable, to me, is always questionable, but how objectionably so depends on the study question.
Note that I used the term sex and you used gender. I made an assumption. I prefer gender as a behavioural construct and sex as a physical one. Despite being classically treatment as a dichotomous trait, sex is not male or female on a refined level. For instance, X0 (Turner's syndrome) and XXY (Klinefelter's syndrome) occur more frequently than some might expect. Also, the levels of "sex" hormones do not always accurately classify physical or karyotypic sex. However, as measurements of sex hormones or other traits are rarely done, we are usually stuck with female and male sex.
By the way, I have seen recommendations that p-values, if you will, on interaction terms as high as 0.25 should mean that the term should be retained in the absence of a (strong) a priori argument.
Kevin Viel, PhD
Senior Research Statistician
Patient Safety & Quality
International College of Robotic Surgery
Saint Joseph's Translational Research Institute
Saint Joseph's Hospital
5671 Peachtree Dunwoody Road, NE, Suite 330
Atlanta, GA 30342
(678) 843-6076: Direct Phone
(678) 843-6153: Facsimile
(404) 558-1364: Mobile
This e-mail, including any attachments is the
property of Catholic Health East and is intended
for the sole use of the intended recipient(s).
It may contain information that is privileged and
confidential. Any unauthorized review, use,
disclosure, or distribution is prohibited. If you are
not the intended recipient, please delete this message, and
reply to the sender regarding the error in a separate email.