F distribution on the 41C

[Dieter]

(12-22-2015 07:30 AM)Paul Dale Wrote:  This source is somewhat frowned upon in numerical analysis circles. There are a lot of simplifications in their algorithms and limited or no error handling.

I think the idea of the NR book simply is to show a first approach that works "good enough". A real world implementation often needs more care. And after all these are just "recipes". The real cooking is up to you. ;-)

(12-22-2015 07:30 AM)Paul Dale Wrote:  There definitely are better implementations of the incomplete beta and gamma functions

Could you say a bit more about this? I have used the suggested method (A&S 26.5.8) for the regularized beta function in the 35s distributions pack as well as in Excel's VBA – e.g. for testing the various quantile functions ;-) – and I did not find much to complain about. But I always like to hear about better methods.

Edit: among my PDF files I just found a paper by A. R. DiDonato and M. P. Jarnagin, titled "The Efficient Calculation of the Incomplete Beta-Function Ratio for Half-Integer Values of the Parameters a, b". I seem to remember that you mentioned it some time ago. However, the discussed methods are not trivial.

Dieter