[R] estimating degrees of freedom iof student t

[Prof Brian D Ripley]

On Tue, 6 Jun 2000 gabrielle.kelly at ucd.ie wrote:

> I have come across the following situation when using the function 
> pt which calls the student t distribution function. I simulate data 
> from a normal distribution and fit the student t. The estimated 
> degrees of freedom gets  larger at each iteration and there is no 
> convergence. It seems there should be some mechanism where it 
> switched to a normal distribution when the degrees of freedom gets 
> very large.

(There is: pt switches at n = 4e5 to a normal approximation, although a
more accurate one than N(0, 1). It's in the source code
.../src/nmath/pt.c.)

I think you need to tell us a bit more: there is nothing in R that I know
of to estimate the parameters of a t distribution.  If you are using
maximum likelihood, there is no guarantee that I know of that the
likelihood has a global maximum, let alone a unique one.  It is entirely
possible that the normal is a better fit that any t.  I would
transform the space, and optimize over 1/nu constrained to [0, 0.5], I
think.  (For nu == 1 all sorts of thing go wrong.)

-- 
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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._