Date:         Wed, 8 Dec 1999 20:20:50 +0100
Reply-To:     peter.crawford@DB.COM
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Crawford <peter.crawford@DB.COM>
Subject:      Re: When large number of significant digits may be required
Comments: To: jaclyn@bama.ua.edu
Content-type: text/plain; charset=us-ascii

: given this focus on input precision as relevant to results precision ( and
assuming the precision of processing is adequate within the other two, or, at
least measurable )

 are there commonly used algorithms in use in SAS for error analysis ?

or comparable algorithms from other (perl like) environments ?

hoping to hear...


Datum:         08.12.99 20:04
An:            SAS-L@listserv.uga.edu




Antwort an:    jaclyn@bama.ua.edu

Betreff:       Re: When large number of significant digits may be required
Nachrichtentext:



I am doing a very bad thing ... I am jumping into a thread whose
beginning I haven't read; please forgive me.

I just thought I would add my one cent and say that sometimes the issue
with significant digits isn't due to computational error at all, but
measurement error -- before it even gets to SAS.

For example, if you have a ruler that only measures to the nearest
millimeter, or if incomes are measured to the nearest dollar, you could
theoretically get into trouble if you wait until the end to round.  This
isn't my field, but I seem to remember from a beginning physics class
that you needed to keep track of sig. digits at each step of the process
or you might give an answer that is TOO exact and not backed up by the
measurements.

But I'll also point out that I never did well in that physics chapter.
:)

jaclyn