F distribution on the 41C

[Namir]

(12-18-2015 10:25 PM)Dieter Wrote:  

(12-17-2015 12:52 AM)Namir Wrote:  Here is a Visual Basic implementation for the cumulative distribution function for the F statistics. It is based on the Handbook of Mathematical Function.

Namir, thank you very much for this extensive implementation.
But why do you do it this complicated instead of a simple regularized Beta function? Which could be used for other distributions as well.

Could you try x=3 with 1 and 5 degrees of freedom? I tried your code in Excel VBA (with the usual syntax modifications for my older version) and got a result of 0,535... while the correct result is 0,856... The problem occurs for df1=1 and df2 odd. Maybe I did something wrong here?

BTW I like the Gamma implementation which seems very exact. It seems to use a polynomial approximation for x=0...1. For comparison, here's my humble Gamma(k) function which I use for the distributions – after all only integers and half-integers have to be handled:

Code:

If k = Int(k) Then g = 1 Else g = sqrt(Pi)

Do While k > 1.25
  k = k - 1
  g = g * k
Loop

Gamma = g

;-)

Dieter

Thanks Dieter for the observation. I found the bug:

If Df2 > 1 And Df1 > 0 Then

And changed it (in the original thread) to:

If Df2 > 1 And Df1 > 1 Then

No I get the same result for Fcfd(1,5,3).

Namir