Date: Fri, 6 Aug 1999 14:30:21 -0400
Reply-To: "Hudson, Spencer" <shudson@VIROPHARMA.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Hudson, Spencer" <shudson@VIROPHARMA.COM>
Subject: Re: Real stats on real big data?
Content-Type: text/plain; charset="iso-8859-1"
One of the more absurd recent examples of the rejection of statistical
sampling is the recent U.S. Supreme Court decision to prohibit the U.S.A.
Census Bureau to use sampling to assess the magnitude of next year's census
undercount.
Cheers,
Spencer
P.S. The 'e' symbol only appears on my Branston pickle.
_________________________________________________
Spencer Hudson, Ph.D
Director, Biostatistics and Clinical Data Management
ViroPharma Incorporated
405 Eagleview Boulevard
Exton, PA 19341
Telephone: 610 458 7300 extension 154
FAX: 610 458 7380
Electronic Mail: shudson@viropharma.com
Web: www.viropharma.com
Ticker: VPHM (NASDAQ)
-----Original Message-----
From: John Whittington [mailto:medisci@POWERNET.COM]
Sent: Friday, August 06, 1999 10:57 AM
Subject: Re: Real stats on real big data?
--- snip---
I think you'll find that even the commercial, financial and 'bureaucratic'
worlds are gradually coming to accept statistical approaches in many
situations. It must be about half a century since 'statistical quality
control' (rather than 100% testing) came to be widely accepted in most
fields, and a few decades since (at least over here) statistical methods
became acceptable in terms of defining the weight/volume/whatever of
consumer goods (do you have the 'e' symbol over there?).
In some situations, there will, of course, always be a need to present
every single item of data - and there is no way around that. However, the
moment one moves oin from that, one has to remember is that any
'summarising' technique is essentially one of 'data-reduction' (i.e. detail
information is being lost). If one is deriving a summary statistic, it
therefore does not generally matter if it has some (very small) degree of
'uncertainty' associated with it - even for 'official' purposes. Indeed,
if the uncretainty gets small enough, it will often be lost in 'rounding'.
Say your government was producing a figure for the mean family income, for
a whole State or even the whole country, and presenting that mean figure
rounded to the nearest dollar. If one derived the figures from a sample,
if the confidence interval (at whatever desired degree of 'certainty' was
less than a dollar wide, it would essentially be 'absorbed' by the rounding.
| 
