Mark writes:

> Bill wrote:
>  > "Self-selected polls are usually very inaccurate."
>  >
>  > Never stopped Fox News "news analysts" from having polls on their
websites
>  > and then reporting the results as if they actually MEANT something.
>
>  ALL "news analysts" do this.  USA Today/Gallup, WSJ/NBC, etc.  It's an
>  unscrupulous and incestuous practice that should be abandoned by all
>  self-respecting news organizations.
>
>  Even so, the difference between news organizations and an AOL poll is that
>  the former usually tie themselves to a polling company to make sure the
>  sample reflects the big picture.  There are exceptions of course, like last
>  month's poll taken by the LA Times, who used a different mix of Democrats
>  and Republicans that created some controversy.
>  http://washingtontimes.com/national/20040615-111040-3254r.htm
>
>  The problem with ANY online poll is controlling your sample.  You have no
>  idea who is voting or how often.  That is what Ken means by
>  "self-selecting".  In addition, you don't know what the context of the poll
>  was on its page.  I assume it is on the Home Page, next to an articles
about
>  Bush and Cheney and which are most likely not flattering.

Let me apologize for being so long getting back to this, but there is some
high nonsense in what Mark writes, material that should be addressed.

The notion that polls should be left to the "professionals" is more than a
bit misleading. Sampling design is an inherent part of every scientific
experiment, and although the underlying ideas are not complicated, they must be
understood by every experimenter.

Two kinds of errors plague sampling: systemic errors and errors of bias. They
aren't the same thing, but they both distort the end result. Systemic errors
tend to be mechanical errors, where certain groups are being oversampled or
undersampled simply because they're easier or harder to include for some
particular reason (or are the results of purposeful cheating; this latter error is
extremely rare in scientific enterprises, but it's not all that unheard of in
human affairs).

The second failing are errors of bias, either conscious or unconscious. If
the hypothesis under sampling precorrelates in some fashion with the
predispositions of the sampled group, you're going to get a skewed result. Orthogonality
(mathematical independence), where belonging to the defined group prespecifies
nothing about the result, is always the desired condition.

For both of the those reasons, the AOL poll would pass muster with me. People
joined AOL for reasons completely unrelated to the anticipation of their 2004
vote. Moreover AOL represents a deep and broad swath of Americans. AOL now
claims 35 million members worldwide, half of which at least must be Americans.
Given that there are now about 290 million US citizens, AOL membership thus
represents somewhere between 5 and 8% of the American citizenery, not an
insignificant number.

Even if there were an initial bias in the sampling the universal group, as
you increase the percentage of those samples, that bias is diminished. Indeed,
if you sample everyone, whatever sampling bias originally existed completely
disappears. But I can see no reason to believe that the AOL population is
anything other than a reasonably accurate representation of the voting population,
other than perhaps with the exception of class by wealth. I would reasonably
expect that the AOL population would oversample Americans above the poverty
level.

The second question is that of systemic errors. AOL knows very precisely
where you live, at least as well as National Security Agency. AOL has to send you
a bill, after all. The NSA is just looking for terrorists. Thus the votes cast
are going to line up with the states in which they are cast almost perfectly.

The second question is that of cheating. Can an individual vote more than
once? Yes. You can vote once on every PC on which you have AOL loaded and use the
same screen name. I've done that my accident just wandering around here. But
would that greatly influence the outcome? Do Democrats own more PCs than
Republicans and thus greatly distort the result? I doubt it.

Or alternatively, is there a national underground campaign by one party or
the other to greatly swing the votes by rallying their troops to vote one way or
the other?  I certainly haven't heard of any such activity. Because most
people do not even know of the straw poll, and because it is unknown and/or
considered unimportant, I would tend to believe that it is an accurate sample.

Nor is the poll on the front page of AOL. You find it only if you read the
News/Election section, thus I take that to indicate that those who are voting
are interested in the election and would therefore be more likely to vote than
the general population.

But what Mark says about the poll being next to unflattering articles about
Bush and Cheney is true. However the stories are not political propaganda. They
are the same AP, Reuters and other newswire stories that appear in your local
newspaper, just a day earlier. If the Bush-Cheney stories are unflattering,
they have no one to blame but themselves.

Moreover, the popular vote in the AOL poll is running 2:1 for Bush in places
such as Mississippi and Alabama, just as it running 2:1 against Bush in places
such as California, New York and Pennsylvania. What unfortunately distorts
the electoral college vote so much in favor of Kerry is that California, New
York and Pennsylvania have a heck of a lot more votes than do Mississippi and
Alabama. It is the nature of the electoral college to amplify small differences,
so that a popular vote that may be 55:45 becomes a landslide of 3:1 in favor
of one candidate or the other.

For all of these reasons, I would, at the moment, tend to trust the AOL poll
as much as I do any professional polling organization -- but for only the
moment. If it should ever come to be regarded as a "bellweather" predictor, I'm
sure that it would immediately come under attack as each group attempts to
distort its results.

Wirt Atmar

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *