|Date: ||Mon, 8 Jan 2007 08:29:16 -0800|
|Reply-To: ||Minya Pu <minya.pu@GMAIL.COM>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Minya Pu <minya.pu@GMAIL.COM>|
|Subject: ||Re: user-specified random effects in glimmix?|
|Content-Type: ||text/plain; charset=ISO-8859-1; format=flowed|
I am looking into some survey sample analysis tools at this very moment.
I am trying to analyze the survey data collected from respondent
driven sampling method (which is causing all the pain). I am trying to
adjust for both sampling weights and correlation of the data (cross
sectional, the dependence variable is binary).
On 1/7/07, David L Cassell <email@example.com> wrote:
> minya.pu@GMAIL.COM wrote back:
> >On 1/2/07, Dale McLerran <firstname.lastname@example.org> wrote:
> >>--- Minya Pu <minya.pu@GMAIL.COM> wrote:
> >> > Hi,
> >> > I want to do a logistic regression model with random effects but I
> >> > need to sepcify my own design matrix for the random effects. Can I do
> >> > it in proc glimmix or anything similar? If so, how?
> >> >
> >> > Thanks very much,
> >> > Minya
> >> >
> >>The GLIMMIX procedure employs two design matrices, one specified
> >>by effects (variables) listed on the right hand side of your MODEL
> >>statement (the fixed effect design matrix) and the other constructed
> >>from effects listed on RANDOM statements (the random effect design
> >>matrix, as if there was any doubt!). Either of these may be
> >>affected by naming an effect on a CLASS statement.
> >>Now, since the design matrices are simply the collection of all
> >>predictor variables which are identified as either fixed or random,
> >>then you can use any data manipulation methods you like to construct
> >>those design matrices. Specifically, you could construct a set of
> >>predictor variables in a data step and name those variables on a
> >>RANDOM statement.
> >>In order to offer advice which is more specific than that, you need
> >>to be more specific about what you want for your random effect design.
> >>Dale McLerran
> >>Fred Hutchinson Cancer Research Center
> >>mailto: dmclerra@NO_SPAMfhcrc.org
> >>Ph: (206) 667-2926
> >>Fax: (206) 667-5977
> > I wanted to use random effects to take care of the correlation
> >among the subjects caused by a complicated recruitment method. It is
> >easy for me to construct a design matrix than to derive the matrix
> >from some variables. I will think about the latter and ask for more
> >advice later.
> >Thanks for the response,
> That sounds like you have a survey sample, and you should be using
> survey sample analysis techniques instead of random effects to model
> your data structure. Perhaps, if oyu explain more about the data
> sources and the data scope and the meta-data, a better analysis
> plan will emerge.
> David L. Cassell
> mathematical statistician
> Design Pathways
> 3115 NW Norwood Pl.
> Corvallis OR 97330
> Find sales, coupons, and free shipping, all in one place! MSN Shopping
> Sales & Deals