[MR] Pollings in general

[Richard Fitzgilbert]

I really didn't want to open a thread on survey methodology but...

The simplest forms (least complicated?) to calculate confidence estimates are
based on assumptions including normal distributions.  That's the stuff we
learned in undergraduate statistics and sampling classes.  In modern surveys we
get stuff that looks less like math and more like voodoo.

There are many other methods of sampling and their associated methods of
confidence estimation.  It is very common to use stratified multi-stage samples
in modern surveys.  The math is tedious and is just made for computers.
Nonetheless, reliable estimates of confidence can be obtained.  Or, one might
say that reasonable methods exist for making estimates from a wide variety of
samples.

None of this is particularly useful for the issue at hand.  I would not be
willing to make an estimate better than "high" for my confidence that the
previous polling presents a usable representation of the characteristics of the
whole.  I could not produce convincing math to support that supposition since
it's a combination of survey and SCA experience.

Jeff

-----Original Message-----
From: atlantia-bounces at atlantia.sca.org
[mailto:atlantia-bounces at atlantia.sca.org] On Behalf Of Peters, Rise J.
Sent: Monday, July 07, 2003 3:55 PM
To: atlantia at atlantia.sca.org
Subject: RE: [MR] Pollings in general


Richard, Jeff really, said:  "OK, on the other side of the coin, statistical
treatment of small sample surveys can determine the reliability of the results
drawn from the sample.  It's counter-intuitive, in fact, how little it takes to
reliably represent a large population accurately."

My recollection, from the long ago days when I knew something about this
subject, is that drawing conclusions from small samples leads to reliable
results if you can assume that your sample is random.  Here, it's pretty clear
that the sample isn't random.  Rather, there's been self-selection by those who
(a) care about the outcome, (b) believe that their response will matter, and (c)
have the wherewithal (time, form, whatever it took) to respond.

In the absence of a random sampling technique, generalizing from a small
response becomes much iffier (that's a technical term).

=Rise, who did this stuff in a former life...

========================================================================
                   The Merry Rose Tavern at Cheapside
    List Info: http://merryrose.atlantia.sca.org/
  Submissions: Atlantia at atlantia.sca.org
Subscriptions: http://seahorse.atlantia.sca.org/mailman/listinfo/atlantia