Date:         Mon, 26 Apr 2004 08:39:43 -0400
Reply-To:     Susie.Li@US.SANOFI.COM
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Susie Li <Susie.Li@US.SANOFI.COM>
Subject:      Re: Percentile for a specific value against an empirical
              distribution
Content-Type: text/plain; charset=us-ascii

In pharma, we are used to pctldef=10 (pctltile 1-10=decile 1, pctltile
11-20=decile 2,..., pctltile 91-100=decile 10).  As the result, there are
many points tied on the decile boundaries.  We promote them to the higher
of the deciles in general.

What would be the easiest way to handle such 'DECILING', essentially a
collapsing of percentiles?

Susie Li
Sales Operations, Sanofi-Synthelabo, Inc.
90 Park Ave, New York, NY 10016
(212)551-4385
susie.li@us.sanofi.com

__________________


Susie.Li@US.SANOFI.COM wrote:
> How does PROC UNIVARIATE handle ties in calculating percentiles?

Actually, in the computations for just the percentiles (and other
quantiles),
the ties do not matter.  The values get treated just like every other
data
point.  What this means in practice is that if the Kth percentile would
be
the weighted average of values Xj and Xj+1 (PCTLDEF=1), and these are
equal
(they are tied), then the percentile value is still their average..
which is
the same as their (tied) value.  Ties are more problematic in the
computation
of a variety of nonparametric statistical methods, but they just don't
cause
a problem when we compute the percentiles.  In fact, if we're doing
percentiles
by hand, they just make things easier.

HTH,
David
--
David Cassell, CSC
Cassell.David@epa.gov
Senior computing specialist
mathematical statistician


Important:  The Information in this e-mail belongs to Sanofi-Synthelabo Inc.,
is intended for the use of the individual or entity to which it is addressed,
and may contain information that is privileged, confidential, or exempt from
disclosure under applicable law.  If you are not the intended recipient, you
are hereby notified that any disclosure, copying, distribution, or use of, or
reliance on, the contents of this e-mail is prohibited.  If you have received
this e-mail in error, please notify us immediately by replying back to the
sending e-mail address, and delete this e-mail message from your computer.