Peculiarity in non-central qchisq for ncp > 294.92 ...

Peter,

I'm sympathetic to the package author.  I'm running some pretty extreme
simulations, and am quite satisfied by artificially constraining the ncp
to be below 294.  This makes my conclusions slightly conservative, but
only very slightly, and the results are very unlikely to change much when
the proper value is used.

I respectfully submit that undocumented shortcomings are bugs.  This one
did me no damage because I was too impatient to wait for the result.
However, had my simulations been much longer term, I would have obtained
very inaccurate results!

I would be more than pleased to draft a note for the help file for this
and any other that you know of.  I'd view that as being higher priority
than fixing it (them).

Best wishes

Andrew

Martin Maechler <maechler at stat.math.ethz.ch> writes:

which is not really good either:

      plot(function(x) pchisq(x, df=1, ncp=295), from=1,to=1e4, log='x')

shows the pretty bad behavior (a clear bug).

Well... shortcoming, I'd say. The author of those routines didn't
expect people to use them with huge noncentralities. I've had this on
my TODO list for a couple of years by now, ever since fixing the
similar issue with dchisq--see Rnews 1/1. It looks fairly easy to do
something along the same lines, but there is now a double recurrence
relation and it requires a bit of concentration to get the tail bounds
for the termination criterion.

--
   O__  ---- Peter Dalgaard             Blegdamsvej 3
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907

Andrew Robinson			     Ph: 208 885 7115
Department of Forest Resources	     Fa: 208 885 6226
University of Idaho		     E : andrewr at uidaho.edu
PO Box 441133			     W : http://www.uidaho.edu/~andrewr
Moscow ID 83843			     Or: http://www.biometrics.uidaho.edu
No statement above necessarily represents my employer's opinion.