s-news
[Top] [All Lists]

Re: regTermTest & confint, survey package

To: jose Bartolomei <surfprjab@hotmail.com>
Subject: Re: regTermTest & confint, survey package
From: Thomas Lumley <tlumley@u.washington.edu>
Date: Wed, 4 Mar 2009 23:09:48 -0800 (PST)
Cc: s-plus news <s-news@lists.biostat.wustl.edu>
In-reply-to: <BAY105-W39F93FDB740F9A7D05475CB6A70@phx.gbl>
On Wed, 4 Mar 2009, jose Bartolomei wrote:


Dear R/S users,

I am running logistic regression on a complex sampling survey using the function 
“svyglm” of the survey package.

I have two questions.
1) I will like to confirm that regTermTest function is the adequate function
to test for interaction?

Yes, at the moment. I'm working on Rao-Scott type tests for generalized linear 
models, which might go in a method for anova(), but they aren't available yet.


2. Which is the formula/method use to calculate confident limit in the “confint” function?

It depends on your version. Version 3.10-1 of the package added a confint() for 
svyglm, which can do either Wald-type or profile-type confidence intervals.

With earlier versions under R you need to explicitly call confint.default() to 
avoid getting the confint.glm() method.
   confint.default(brGlm1)

Presumably something similar should be available in S-PLUS, but I don't know 
exactly what. Failing that, you may have to compute it yourself from 
coef(brGlm1) and SE(brGlm1).

      -thomas

Thomas Lumley                   Assoc. Professor, Biostatistics
tlumley@u.washington.edu        University of Washington, Seattle



<Prev in Thread] Current Thread [Next in Thread>