R-beta: "glm" bug continued

Jim Lindsey (jlindsey@luc.ac.be)
Tue, 27 Jan 1998 09:25:56 +0100 (MET)

From: Jim Lindsey <jlindsey@luc.ac.be>
Message-Id: <9801270825.AA13777@alpha.luc.ac.be>
Subject: R-beta: "glm" bug continued
To: r-help@stat.math.ethz.ch (r-help)
Date: Tue, 27 Jan 1998 09:25:56 +0100 (MET)

Sorry for the false alarm on the core dump with glm. In fact, my
version calculates the AIC. For a saturated model with dispersion
parameter, such as the gamma, the dispersion parameter is estimated by
the deviance which should be zero. However, it is usually not exactly
zero and may be slightly negative. My AIC for the gamma distribution
uses lgamma(1/dev) which crashes R.
  Summary for negative arguments to lgamma:
0 to -176 yields a negative number
-177 to -1e5  NA
> -1e6  core dump
  This probably explains a number of my core dumps with nlm using the
gamma function in a likelihood function that I mentioned a few weeks
ago on this list. nlm can feed very strange values to functions while
searching and expects an NA if they are not valid.
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@stat.math.ethz.ch