Hi Justin,
I don't have Gelman et al. with me right
now, but I'll look it up when I get
home. I guess I can't complain when
"essentially impossible" inferences lead
to less than robust results. :-)
However,
Kenneth L. Lange, Roderick J. A. Little, and
Jeremy M. G. Taylor. 1989. Robust Statistical Modeling Using the t
Distribution. \emph{Journal of the American Statistical Association}
84 (408)
suggest estimating the degrees of
freedom, I think. I'll have to take a
closer look at how exactly they propose
doing that.
Thanks,
Holger
Justin Ryan Grimmer wrote:
Hey Holger and Gov 2001,
According to Gelman et al, including nu as a parameter to be estimated
represents an "essentially impossible" inference, because the parameter
governs the underlying shape of the distribution (pg 447). Perhaps you
can approximate the maximum nu by comparing the likelihood obtained from
several different values of nu (for example, 2, 4, 10, 15, 30). While
this won't give you the precise max, it will allow you to narrow the
region in which it occurs. I think this is prudent, does anyone else
have any thoughts?
Cheers,
Justin Grimmer
On Fri, 14 Apr 2006, Holger Lutz Kern wrote:
Hi all,
for ps 6 (f), I'm trying to estimate the optimal number of degrees of
freedom. The problem is that my estimates are quite dependent on the
starting value that I give optim. I've tried several methods but they
all produce unstable results even after setting maxit=10000 so that
convergence takes place. Does anyone have any suggestions what to do?
Thanks,
Holger
--
Holger Lutz Kern
Graduate Student
Department of Government
Cornell University
Institute for Quantitative Social Science
Harvard University
1737 Cambridge Street N350
Cambridge, MA 02138
www.people.cornell.edu/pages/hlk23
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
--
Holger Lutz Kern
Graduate Student
Department of Government
Cornell University
Institute for Quantitative Social Science
Harvard University
1737 Cambridge Street N350
Cambridge, MA 02138
www.people.cornell.edu/pages/hlk23