HP50G - Binomial Test Statistics

[cjmcc]

I'm working my way through some test statistics, trying to recreate everything on my HP for fun and to improve my understanding. Stat49 Pro has been a great help providing inverse statistical distributions.

My problem is determining the confidence bounds for some pass/fail tests. The example has 9 successes out of 10 trials. The point estimate for reliability is 0.9, but finding the 80% confidence bounds gets a little messier. The 'approved' method used the BETAINV function in Excel. Poking around on the web showed the Clopper-Pearson method can also provide the upper and lower reliability bounds using the cumulative inverse F distribution. So far, so good.

Stat49 Pro happily determines F(20.2.0.1)=9.44 which leads to an upper reliability bound of 0.99 (matching the example). But when I try to find the lower bound, Stat49 Pro claims F(18,4,0.9)=-2.19E70 instead of the expected 0.44 to provide the 0.66 lower bound.

Does anyone know why Stat49 Pro freaked out? Checking an on-line F distribution calculator gives the expected result and asking the HP50g : 18 4 0.44 UTPF gets 0.90 as expected.

-Chris-

Ref:
https://www.liquisearch.com/binomial_pro...n_interval