RM>The obvious difference in our points of view is that you see the
RM>uncontrolled factors as being fixed in NUMBER, while I'm
RM>inclined to see them as fixed in FREQUENCY...
CB>I don't see where you're headed with this...can you draw me a map?
I'll try.... Let's say you want to examine the effect that
a person's dress has upon the attitudes that people have
on a survey.... The survey is the control; everyone has 
the same survey, they simply are dressed differently. The 
study turns up differences in the results of the survey.  
It doesn't much matter what is being surveyed, there is 
bound to be some variation in the results.  Let's say, for 
the sake of argument, that respondants are less favorable 
on the surveys if dressed casually.
Does the result MEAN anything?  Maybe, maybe not.  There 
are other factors involved: age, sex, race, political
views, past experiences, family background, finances...
Ok... Now supposed we expand the sample to a larger group.
Your expectation is that all the other factors will even
out; I contend that that is highly unlikely.... Suppose
we extend the study to include all of Tennessee, for
example, and let us further assume that the survey has to
do with a topic such as gay rights... The results in
Tennessee are going to be much different than in New York,
and the dress of the surveyors is probably going to be a
minor factor in the outcome.  The study presumes that
there is only one factor involved, when there are in fact
MANY factors, and some more important than the one being
tested.  In such a situation, expanding the size of the
study does not impact on the validity of the outcome.  If
you cannot isolate a single variable, ANY conclusions you
draw are suspect.
You asked about polls.... We have a different situation
here, because we are looking at the results of a sample,
and then comparing that result to something we KNOW to
have resulted from past experience... Example: NYC is a 
democratic stronghold; a weak showing for a democratic
candidate there would spell real trouble - Because past 
results show that if a democratic candidate falls below 
a certain number, he or she has always lost.  There is
an anchor, a KNOWN result to work from.  That's not the
case with a study involving an UNKNOWN quantity....
___
 * MR/2 2.26 * OS/2 WARP...Opens up Windows, shuts up Gates.
--- Silver Xpress Mail System 5.4P1a
---------------