On 17 Nov 2003 08:04:15 -0800, [EMAIL PROTECTED] (Mike Kuehne) wrote:

> Dear list,
> 
> Maybe this is some off-topic, but I would like to enquire a question
> about the handling of correlations and the results of multivariate
> analysis of a complete survey (a poll of the entire population and not
> just a sample). 
> 
> Without the need of Statistical inference that draws conclusions about a
> population based on sample data every correlation is significant.

I believe that you have the wrong word; 
'significant' does not work very well in that context.

Every correlation  is *known*.
In the real world, very few people are interested.
What people are interested in, almost always, is
the use of inference to other circumstances, other
times, other 'populations'  in any infinite sense.

The main, public use of Finite Population Correction 
(likely keyword, FPC)  is on election eve, to predict the
outcome of a close race before the final 10% has 
been counted.  The big risk for those estimates arises
because the 'sampling'  (who has not been counted)
is not random; a good estimate must take into account
both (a) the number of votes, and (b) how they are
going to fall, if that won't be random.

FPC is used in animal management, I've been told, and
this is an application where the use of FPC  is called
'research'.   Mostly, the notion of 'population'  is basically
the far extreme from the sort of generalization that 
research is concerned with.

A version of FPC  is used in quality control: 
especially,  with stratified sampling.

> 
> Could somebody recommend me basic literature or personal experience of
> this topic?

I will be interested to learn if you actually have a 
multivariate analysis with a purpose that justifies FPC,
but my past experience says that you are probably
doing the wrong analyses.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization." 
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to