Re: strange f_cdf results

[Kurt Hornik]

>>>>> G Buerger writes:

> Hi,
> I have trouble with Fisher's cdf, f_cdf. For example, it gives the following
> incorrect values (the correct values, according to naglib, is 1):

> octave:165> f_cdf(30,4,150) 
> ans = 6.7157
> octave:164> f_cdf(30,4,200) 
> ans = 20638866

My code :-)

But seriously, the problem is that the m-file code for the incomplete
beta function, betai, simply does not work for large values.

The problem has been around for some time now.  JWE and I both think
that for reasons of efficiency etc the current code for both betai and
gammai should be replaced by a numerically stable etc version coded in
Fortran or C, such as those from the APSTAT archive.  But it simply has
not been done.

There is also an Octave interface of CDFLIB, but I am not sure whether
this will be merged into the distribution proper.

-k