Version 1.1 of acs/ac-00083.txt

Unformatted version of acs/ac-00083.txt version 1.1
Other versions for file acs/ac-00083.txt

!standard A.05(00)          03-12-04 AC95-00083/01
!class amendment 03-12-04
!status received no action 03-12-04
!status received 03-09-27
!subject Statistics packages

From: Robert I. Eachus
Sent: Saturday, September 27, 2003  10:29 PM

I have been working for the past month or so off and on reorganizing the
statistics software I have.  The problem is that if you try to do anything
fancy, you need half a dozen distributions and functions which are hard to
calculate.  For example, a package that does Chi Square testing ends up
calculating F-distributions and incomplete beta functions.  So I decided when I
wanted to do another K-S test to reorganize things right, with all of the
distributions reachable, and a different view into the packages by type of data.
 (One sample, one variable, then two samples one variable, then multivariate

So that if you want to run a Mann-Whitney U test that requires the Two_Sample
package and so on.  (Rather than having all the two-sample tests provide a
different interface.)

Should I submit this as a candidate for Ada 0Y?  This has been on my "To-Do"
list for years, but I finally realized this summer how it should be organized.
And yes, most of the underlying code I wrote years ago.  A lot of it predates
Ada 95, but I finally realized how it should be organized.  There are really two
views, the user who doesn't care about the implementation, and the statistician
who wants to be able to access all the distributions to be able to do things
like K-S tests.

Oh, and if this does get included, we probably should allow vendors to supply
additional distributions.  For each distribution I provide PDF, CDF, Inverse
CDF, and a random variate generator.  (Two random variate generators for
Poisson, because both discrete counts and real interarrival times are useful.)
But I currently have only Normal, Poisson, Binomial, Student's t, Exponential,
Fisher's F, and Gamma.  These allow most useful parametric and non-parametric
tests to be made.  I could add the Chi-Square as a separate distribution, but I
haven't done that yet.


Questions? Ask the ACAA Technical Agent