Date: Thu, 24 May 2001 23:33:38 -0700
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Karsten M. Self" <kmself@IX.NETCOM.COM>
Subject: Re: Months til Infinity?
In-Reply-To: <firstname.lastname@example.org>; from
sdziuban@QWEST.NET on Fri, May 25, 2001 at 12:18:08AM -0600
Content-Type: multipart/signed; micalg=pgp-sha1;
on Fri, May 25, 2001 at 12:18:08AM -0600, Stephen T. Dziuban (sdziuban@QWEST.NET) wrote:
> This is probably a data-design question rather than strictly a SAS issue,
> but y'all may have gone down this path:
> Suppose a (numeric) column should represent predicted-time-to-event
> (PTTE); say, months to inventory exhaustion = inventory remaining /
> monthly use rate.
> Some product rows have 'normal' (positive) PTTE values. PTTE=0 (months
> remaining) would mean that product's inventory is all used up. Some
> rows are blank (eg, remaining inventory is unknown). Some rows though,
> would have PTTE=Infinity (monthly use rate=0); not the same as the
> The user will want an ascending sort on the PTTE column (soonest
> exhaustion at top), and would expect to see the Infinities at the
> bottom. Other users will want an average PTTE by product group, and
> Infinities would confuse that.
> POSSIBLE SOLUTIONS:
> 1) Not-so-pretty: Store almost-infinity, use Median instead of Mean, ...
> 2) OK: Store an additional "case" column (Exhausting, NotExhausting,
> Unknown) and use it as well as PTTE_if_Exhausting.
> 3) ???
This is a business-rules problem, not strictly a data design problem.
The question is: what are the business needs, do they coincide, and of
what significance are they.
There is no single numeric solution which is going to suite all cases.
Even the built-in SAS features that might be used to address this won't
work smoothly on a single-field solution. You could code various SAS
missing values to indicate reasons for noncomputed values: unknown
inventory, zero exhaustion rate, etc.
Computation of means will have to toss uncomputed values. I'd add a
note to any such report stating the number of excluded values by
causal category. While reporting medians may be of interest, mean is
probably a more useful forecasting tool as an unbiased measure of
Ordering the data needs to be defined by report. Your clients have to
realize that there isn't a single definition that is appropriate to all
possible data interpretations, but that some fairly simple options
(which you've largely outlined) exist. I've seen fierce battles fought
by people who didn't realize that they were fundamentally talking about
different, though related, concepts.
Karsten M. Self <email@example.com> http://kmself.home.netcom.com/
What part of "Gestalt" don't you understand? There is no K5 cabal