Hi,
Does anyone know of a good algorithm for calculating the inverse cdf of the
beta distribution beta(p, q)? Or a good algorithm for calculating the cdf
of the beta, that is, the incomplete beta ratio, from which we can do a
linear search to get the inverse cdf? We need to achieve high levels of
accuracy (say 5 or 6 digits at least) for extreme beta distributions over a
wide range of parameter values, say 0.0001 < p,q < 50. We implemented
Algorithm AS 63 Appl. Statist. (1973), vol.22, no.3
which is based on Soper's (1926) algorithm. It does not seem to work for
extreme distributions (highly skewed or concentrated densities).
I am reading Johnson & Kotz but the accuracy of the algorithms mentioned in
the book seem low. Any advice or suggestions are appreciated.
Best wishes,
Dr Ziheng Yang
Department of Biology (Galton Lab) Phone: +44 (20) 7504 5083
University College London Fax: +44 (20) 7383 2048
4 Stephenson Way Email: [log in to unmask]
London NW1 2HE
England
http://abacus.gene.ucl.ac.uk/
|