[R] warning in binomial analysis

[Thomas Lumley]

On Mon, 23 Sep 2002, John Maindonald wrote:

> Surely any problem is with the fitted probabilities, not
> with the coefficients.

Yes, but I think it's a symptom of a problem with the covariates.

If we were working in single precision there could well be other
explanations, but 1- exp(eta)/(1+exp(eta)) is only 3e-7 (and thus has
seven or eight accurate digits) for eta=15.  Even for eta=25 we are
nowhere near complete numerical breakdown.

You could  construct a situation where this warning would occur
even with finite MLEs, but I'd be surprised to see it happen in a case
where I would be happy with the fitted coefficients (the deviances are
another matter).  On the other hand, it happens quite often with sparse
design matrices when the MLEs are really infinite.

	-thomas

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._