Date:         Tue, 31 Mar 1998 10:43:53 -0600
Reply-To:     Jack Hamilton <jack_hamilton@HCCOMPARE.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Jack Hamilton <jack_hamilton@HCCOMPARE.COM>
Subject:      Re: Quality Assessment
Comments: To: msd@INEL.GOV
Content-Type: text/plain; charset=US-ASCII

Mark DeHaan <msd@INEL.GOV> wrote:
 
>I had an interesting question posed to me today and wanted to get your input:
>
>How does one verify the accuracy of a software PACKAGE? How would one verify
>that the PACKAGE is accurately doing what it is designed to do?
 
Yes, it is an interesting question, and there's no easy answer.  A few
possibilities:
 
- If there's another package that's supposed to accomplish the same
  function, run your numbers through both and compare the results.
 
- Double-code, and compare results.  For example, if you are performing
  a bunch of calculations with data steps and procedures, try coding the
  problem in SQL instead (or vice-versa).  You might also discover aspects
  of the problem that were hidden by your original method.
 
- Have code inspections, so that a different set of eyes and brains can
  check what you did.
 
- Create a test input dataset that contains lots of boundary conditions
  and pathological cases.  Does your program produce the right results?
 
- If possible, create a visual representation of the answers.  A graph
  will sometimes show problems that wouldn't be noticed in the sea of
  numbers produced by a tabular report.