This is a query about what "should happen" when a function is given
arguments that are in some sense out of range.
The hypergeometric distribution function computes the probability that x
"successes" are observed in a sample of n trials from a finite
population of N elements where M of the N are "successes. We sample
WITHOUT replacement. Example: 10 people in a village of 100 have math
anxiety. We sample 5 of them and clearly don't check the same people
twice ('without replacement'). Hypergeometric distribution gives the
probabilities we get 0, 1, 2, 3, 4 or 5 with MA in our sample.
Now suppose we have nobody with the disease. Clearly P(0, 5, 0, 100) =1
and the rest should be 0. Gnumeric reports for "=hypgeomdist(1,5,0,100)"
the result
#NUM!
which has some merit, since we are doing something that is "impossible"
(getting 1 out of none).
I can see reporting an error for
=hypgeomdist(1,15,0,10)
i.e., sample bigger than population. But I know I'd rather get 0 for the
values that are "impossible" in a feasible sample.
There are several of these borderline cases throughout the functions. It
would be nice to document them and then come up with a good set of choices.
Comments welcome.
JN
_______________________________________________
gnumeric-list mailing list
[email protected]
http://mail.gnome.org/mailman/listinfo/gnumeric-list
