Orthogonal designs and logistic regression

From: patrick f (pebf_at_large_at_yahoo.com.au)
Date: 02/03/05 

Date: 2 Feb 2005 21:33:44 -0800

Hi all,

I'm aware that a linear orthogonal design will not result in
orthogonality of logistic regression coefficients, in particular these
will usually (always?) be correlated.

I'm also aware of difficulties in estimating uncorrelated
coefficients, in nonlinear models in general, due to the parameter
information matrix depending on the unknown parameters themselves.

Does anyone know of work or publication(s) on the generation and/or
use of orthogonal designs for fractional factorial experiments in
logistic regression?

As an aside, would anyone care to comment on the benefits (from the
point of view of interpretation) of using such a design if it were
possible?

Many thanks,

Patrick FitzGerald