|
Here is a short outline of the study:
The States and counties are fixed since our study is limited to
studying the social ecology of stimulant use in rural areas in general,
or how rural drug abusers are involved in high sexually risk behavior
specifically in Ohio, Kentucky, and Arkansas. Certain counties within
each state will be randomized to receive a community level intervention
or control.
The individuals were recruited using Respondent-Driven Sampling
(similar to Snow-ball sampling). The intervention uses the Social
Cognitive Theory and the Transtheoretical Model to develop tailored
individualized interventions. This will be a valuable framework for
understanding and reducing sexual risk behavior among our population of
interest.
Thanks,
Song
David L Cassell wrote:
> sounpra@YAHOO.COM wrote:
> >
> >Dear SAS Users,
> >
> >I'm trying to compute the intraclass correlation coefficients for a
> >multilevel model. I have subjects nested within counties which are
> >nested with states. I used the following PROCS to obtain the estimate
> >variance components.
> >
> >My code is setup to obtain the ICC for subjects within counties, but
> >how do I obtain subjects within counties within states?
> >
> >proc sort data=temp_data;
> > by state county;
> >run;
> >
> >proc nested data=temp_data;
> > class county id;
> > var response;
> > title'Using PROC NESTED to Estimate Variance Components';
> >run;
> >
> >proc varcomp data=temp_data;
> > class id county;
> > model response = county id(county);
> > title'Using PROC VARCOMP to Estimate Variance Components';
> >run;
> >
> >proc mixed data=temp_data covtest;
> > class id county;
> > model response= / solution;
> > random county id(county)/solution;
> > title'Using PROC MIXED TO ESTIMATE VARIANCE COMPONENTS';
> >run;quit;
> >
> >Thank you all!
> >
> >Song
>
> You have states, then counties, then people? That does not sound
> like a multi-level model. That sounds like a multi-stage sample design.
>
> If, in fact, you have a survey sample design, complete with design
> effects and sample weights, then you need to stop what you are
> doing, and focus instead on a survey analysis approach using PROC
> SURVEYREG instead of PROC MIXED.
>
> It sounds like you have a sample that will naturally fall into the
> structure of multi-stage sampling. In that case, you have no
> ICCs to muck around with, because you are not dealing with
> hierarchical lienar model estimation. You only have to focus on
> adjusting your sample weights at each stage of sampling, in order
> to accommodate things like non-response.
>
> HTH,
> David
> --
> David L. Cassell
> mathematical statistician
> Design Pathways
> 3115 NW Norwood Pl.
> Corvallis OR 97330
>
> _________________________________________________________________
> Try the next generation of search with Windows Live Search today!
> http://imagine-windowslive.com/minisites/searchlaunch/?locale=en-us&source=hmtagline
|
