Need a book? Engineering books recommendations...

Return to index: [Subject] [Thread] [Date] [Author]

Re: Invasive Testing

[Subject Prev][Subject Next][Thread Prev][Thread Next]
Ray;

Unfortunately all statistics is going to give you is numbers.  You then have
to argue about the meaning and significance of the numbers and how you
arrived at them (that's what the Lawyers are for, remember DNA and the OJ
trial).  Don't look to the math to make your decisions and value judgements
for you (or anyone else for that matter).  As you sound like you are, be
rather skeptical though about someone that claims to have found truth in
statistics.

The value of the statistic is too going to be greatly dependant on the
sample size and the confidence intervals you might therefore expect from
them.  In a really small sample the confidence interval may be so broad as
to make any firm conclusion questionable.  One very important issue is was
the sample as taken truly random (in space or time or whatever is relevant).

Its hard to know what you would do with your situation without knowing a lot
about what's involved and what sort of argument or view you could construct.
You might brush up on "sampling without replacement" and the "student
t-distribution" and all the things that lead up to and surround it. That's
just a guess though.

Depending on the nature of the problem, one sort of nifty method I've been
trying to come to grips with as I can is something called the bootstrap,
resampling, or non-parametric statistics.  It and its relatives can be quite
useful in dealing with small sample sizes.  A good text on it is
"Introduction to the Bootstrap" by Efron and Tibshirani.  There is also an
interesting web site www.statistics.com.

Like me, your probably not really interested in some academic exercise in
statistics usually.  Unfortunately, dealing with statistics in any given
case can get into just that real quickly if you want to keep you feet on the
ground and avoid fooling yourself and others.

cmd
----- Original Message -----
From: <RShreenan(--nospam--at)aol.com>
To: <seaint(--nospam--at)seaint.org>
Sent: Saturday, 14 August, 1999 1:10 AM
Subject: Invasive Testing


> To all forensic experts out there:
>
> What if any is the accepted and fair mathetical method for projecting the
> total number of residential units that are likely to have a specific
defect
> based on this defect, found by invasive testing in a small percentage of
less
> than half of the total number of residential units in a development?
>
> Is a direct proration valid even if the number sampled is small compared
to
> the total number of units involved?
>
> I would like to be able to refute some of the numbers that the plaintiff's
> expert witness came up with, with documentation of accepted methods if
indeed
> they do exist.
>
> Any advice or references regarding this process would be greatly
appreciated.
>
> THANKS,
> Ray Shreenan  SE
>
>