Date: Wed, 11 May 2005 10:11:53 -0700
Reply-To: Dale McLerran <stringplayer_2@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dale McLerran <stringplayer_2@YAHOO.COM>
Subject: Re: Need help in statistical modelling
In-Reply-To: 6667
Content-Type: text/plain; charset=us-ascii
--- Peter Flom <flom@NDRI.ORG> wrote:
> kumsaa@hotmail.com wrote:
> <<<
> I have a project aimed to assess the determinants of allergen level
> in
> individuals. I have about 22 centres in 10 countries with about 200
> subjects in each. From every subject blood allergen level was taken
> once. Detailed questionnaire about household, pet keeping and
> personal
> history is available. I want to determine the influence of pet
> keeping
> on specific allergen to cat and house dust mite. My outcomes are
> continuous (not normally distributed) and my predictors are a mix of
> continuous, binary and categorical variables. There is a possibility
> of
> correlation within centres. One important issue as well is the fact
> that my outcome is highly skewed, in spite of transformation.
>
> Additionally, the study centres differ in many ways, such as
> climate,
> life style, income and so on. I would be grateful if one could give
> me
> some tips as to what statistical model to use in SAS 8.
>
> To summarize
>
> Centres 22 from 10 different countries
> Observations 4000 ~200 /centre
> Outcome continuous, right skewed
> Explanatory variables binary, continuous and categorical
>
> Individual variables: gender, presence of cat in household, smoking,
> season age of mattress, storey of building;
>
> Centre level variables: prevalence of cat ownership in community
>
> Peter L. Flom, PhD
> Assistant Director, Statistics and Data Analysis Core
> Center for Drug Use and HIV Research
> National Development and Research Institutes
> 71 W. 23rd St
> www.peterflom.com
> New York, NY 10010
> (212) 845-4485 (voice)
> (917) 438-0894 (fax)
>
>
>
> >>>
>
> First, for most purposes, what needs to be normal (or distributed in
> some other way) is not the DV, but the errors
>
> Second, if the assumptions about the distribution of the error are
> violated even after transformation, you need another
> transformation.....either than, or we need more information. e.g,
> even
> if the DV is continuous, it might be 'lumpy'.
>
> Third, it sounds like you need PROC MIXED, or possibly PROC NLMIXED,
> since you have variables at both the individual and center level. If
> you've not done hierarchical linear modeling before, I suggest either
> consulting with someone who has, or alloting a bunch of time to read
> through a text such as Raudenbush and Bryk Hierarchical Linear
> Models.
>
> Finally, the presence of IVs that are of different types is not
> really
> a big deal.....but, again, if you've not dealt with regression models
> that have all these types, it will involve some learning (not as much
> as
> HLM, though)
>
>
> HTH
>
> Peter
>
Peter,
I couldn't have said it any better myself! I might suggest that
SAS publishes a book which is excellent (in my opinion) for both
explaining the theory of mixed models as well as for showing how
mixed models can be estimated using SAS procedures. The book is
"SAS System for Mixed Models" and is available from the SAS
website as well as on Amazon. Probably other places as well.
Dale
---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph: (206) 667-2926
Fax: (206) 667-5977
---------------------------------------
Discover Yahoo!
Stay in touch with email, IM, photo sharing and more. Check it out!
http://discover.yahoo.com/stayintouch.html
| 
