[R] Confidence intervals for logistic regression

[Prof Brian Ripley]

On Thu, 19 Feb 2004, Michael Levine wrote:

> I found myself trying to figure out the type of confidence interval used
> for the coefficients of the logistic regression fit by using
> glm(family=binomial)...

AFAIK, no confidence interval is given by that call: it does not even 
calculate standard errors for the coefs (the summary method does that).
I could guess what you meant, but the answer depends on the guess ....

> I suspect it is Wald confidence interval but am not sure...Does anybody
> know? Also, if so, how can I ask for likelihood ratio and/or score-based
> confidence intervals?

You can use confint() (whose glm method is in package MASS, which must be
attached) to give you profile-likelihood confidence intervals.  There
would be no point in trying to base confidence intervals on score tests,
as you would have to re-fit the model at all possible alternative values
and so the profile likelihood would be available.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595