Re: st: trend analysis for survey data

[Steven Samuels <sjhsamuels@earthlink.net>]

The mean of a binary variable is the proportion with the event. Thus - 
svy regression- will give a test for trend in proportions. For  
proportions between 20% and 80%, linear, logit, and probit  
transformations have similar fits (DR Cox, Analysis of Binary Data,  
Methuen 1970).  The 'Chi Square Test for Trend in Proportions' for  
independent, non-survey data (e.g. PA Armitage Statistical Methods in  
Medical Research,Wiley,  1971, p 363) is essentially a test for a non- 
zero linear regression coefficient for proportions under the null  
hypothesis. (Moreover, it is the maximum likelihood score test for  
trend in logistic regression (reference: I forget).)    All in all  
the linear version is much easier to interpret. Either the linear or  
logistic version have the proper type 1 error rate, but Audrey should  
decide which scale to use after plotting her data. Certainly if she  
wants to interpolate, extrapolate, or summarize the prediction with a  
regression coefficient, she should choose a well-fitting model.

-Steven

On Mar 10, 2008, at 2:35 PM, Mohammed El Faramawi wrote:


> Hi steven,
> It seems that I am missing something. Why do you
> recommend using svy:regress with a categorical
> variable?? why not svy:logit from the beginning?
> thank oyu
> --- Steven Samuels <samplerx@earthlink.net> wrote:
>
> > -svyset- your data. Then try -svy reg- for your 0-1
> > outcome on the
> > time dimension variable.  If the prevalences are
> > close to zero or
> > one, you could consider -svy logit -.  From your
> > question, I take it
> > that the surveys are independent, but taken over the
> > time dimension
> > and that the target populations were identical,
> > except for time
> > period.  You will need to define new strata, which
> > are combinations
> > of survey and the original strata.
> >
> > This assumes that primary sampling units were newly
> > sampled for each
> > survey. If not, you cannot treat the surveys as
> > independent and
> > should come back here for help.
> >
> > -Steven
> > On Mar 10, 2008, at 4:28 AM, audreyp@neo013.net
> > wrote:
> >
> > > hi,
> > > I have data on three/four independent
> > cross-sectional surveys (not
> > > panels). I need to test for a trend in the
> > prevalence of a health
> > > condition. How can I do a trend analysis in Stata
> > that will take
> > > into account sampling weights and stratification?
> > > thank you.
> > >
> > > Audrey P.
> > >
> > > *
> > > *   For searches and help try:
> > > *
> > http://www.stata.com/support/faqs/res/findit.html
> > > *   http://www.stata.com/support/statalist/faq
> > > *   http://www.ats.ucla.edu/stat/stata/
> > Steven Samuels
> > 845-246-0774
> > 18 Cantine's Island
> > Saugerties, NY 12477
> > EFax: 208-498-7441
> >
> >
> >
> >
> >
> > *
> > *   For searches and help try:
> > *
> > http://www.stata.com/support/faqs/res/findit.html
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
>
>
>        
> ______________________________________________________________________ 
> ______________
> Looking for last minute shopping deals?
> Find them fast with Yahoo! Search.  http://tools.search.yahoo.com/ 
> newsearch/category.php?category=shopping
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/