- several questions, explicit and implicit -
On Tue, 31 Oct 2000 16:17:34 -0800, "Brian Vuong"
<[EMAIL PROTECTED]> wrote:
" I work for a bank. I have a data set with 2,600,000
observations[1]. I would like to determine an auditing sample[2] size
that is statistically significant[3] within 1% error." < snip, rest >
1] The 'observations' are probably not 'independent' unless it is a
single branch-bank, with 2.6 million customers. If you don't have
independent observations, then you face some complications in putting
good confidence limits on what you have measured.
2] I don't know what an 'auditing sample' is, but if it is something
to meet legal requirements, then I am pretty sure that you want to
read the pertinent law. It might spell out the fractions needed, and
even more detail about (say) stratifying -- which is probably needed,
if you are comparing dollar amounts. DO NOT spend too much time
trying to interpret fuzzy assertions of some administrator who knows
neither statistics, the law, nor accounting.
3] A sample size is not something that is 'statistically significant'
in the terminology of statisticians. This is an odd piece of
"almost-professional jargon" -- the mis-use is so prevalent
among non-statisticians, it might become sanctified some day.
However, for that to happen, it needs a definition:
Is a 95% Confidence Interval supposed to be assumed?
Is the 1% width supposed to be the S.E. of the mean,
1.96 SEs of the mean (for 95%), 0.675 SEs (for 50%),
or something else?
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
