Attn: JMS, Re: No Hugo for *SiL*

[B5JMS Poster]

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
From: bt at templetons.com (Brad Templeton)
Date: 19 Sep 1999 00:06:03 -0600
Lines: 46

In article <19990918214453.23347.00001494 at ng-fo1.aol.com>,
Jms at B5 <jmsatb5 at aol.com> wrote:
>The more people you have in a poll, THE MORE ACCURATE IT IS.  The fewer people,
>the less accurate.  That utterly incontestable fact is at the core of every
>poll ever taken.  That's why they usually show you the raw numbers ("In a poll
>of two thousand people, ten percent thought Mars was not a planet but a candy
>bar").

If you are going to cite authority to correct somebody, take care.  The
truth is that self-selected polls can be incorrect in a wide variety of
ways, and it's enirely possible for a self selected poll to become
less accurate as the sample grows rather than more accurate.

A larger sample gives a more _precise_ measurement of whatever the
self selected poll is ending up measuring, but that thing may or may
not be an accurate representation of the truth.

A random sampling of 1000 people cab be -- and often is -- more accurate than
a phone-in 900 number vote of a million people.

The P5 polls measures the opinions of the sort of people who E-mail in votes
to such polls, nothing more, nothing less.

In addition, if you examine those results, you find that
there were 6100 votes cast for season 1 and only 3635 cast for
season 5.   So in spite of the show gaining more viewers, the actual
sample size is lowest in season 5 and largest in season 3.

(Clearly the longer amount of time to vote for earlier episodes contributes
to this, but not entirely, otherwise one would not have more votes for
season 3 than 2 and more for 2 than 4, and more for 4 than 1 -- and all
more than 5)   Indeed a whopping 113 expressed opinion on the highly
watched Sleeping in Light vs. 549 on Z'ha'dum.


Anybody with training in these areas will know the many flaws that
can occur in a self-selected poll like this.  Those flaws are even more
dramatic when the voters are able to see the intermediate results, and
thus may try to influence them.

You can get some data out of it, but it's simply not an "utterly
uncontestable fact" that a greater sample size confers more accuracy
on a self-selected survey.
-- 
	Brad Templeton			http://www.templetons.com/brad/


=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
From: jmsatb5 at aol.com (Jms at B5)
Date: 19 Sep 1999 01:07:48 -0600
Lines: 33

>You can get some data out of it, but it's simply not an "utterly
>uncontestable fact" that a greater sample size confers more accuracy
>on a self-selected survey.

What I was responding to was the statement that more responses made a survey
*less* accurate.  Surely you're not saying that that's the case?  And yes, a
larger sampling universe *does* make a survey more accurate within that
preselected universe.  As long as the sampling within that universe is random
-- and there's a wide range of diversity withih that preselected universe --
the survey is accurate.

For instance, phone surveys preselect anyone who has a phone.  At one point,
there were still not a lot of phones in common use in the hinterlands, but that
doesn't disqualify the poll.  A street corner poll preselects those who happen
to be walking by the street corner.

Now, if you're going to preselect for some particular specific variable -- all
asians, for instance, or only people under 25 -- then you start to skew the
data and thus the results, making it inapplicable to the greater, non-sampled
universe.

Point is, insisting that more numbers make a sample *less* accurate -- as VB
was doing -- is utterly and totally inaccurate.

 jms

(jmsatb5 at aol.com)
B5 Official Fan Club at:
http://www.thestation.com




-***
-*** B5JMS SUBSCRIBERS: Replies to messages go to the list maintainer,
-*** <b5jms-owner at cs.columbia.edu>.  If you want to reply elsewhere, adjust
-*** the "To" field.  See http://www.cs.columbia.edu/~ezk/b5jms/ for all
-*** other information about this list.