[EM] advocacy by means of exit polls

[Michael Poole]

RLSuter at aol.com writes:

> It makes no sense in this case to define "wrong" as anything other
> than a wrong count of how people actually voted. Therefore, the
> official results of an election are just as capable of being wrong
> as exit poll results.

Sure, but in practice, the count that actually matters is the one that
is called "official" (perhaps after a court or political contest).
That is the count that third-party polls try to predict, and the one
that they are usually judged against.

> As for your claim that the margin of discrepancy is "closely
> comparable between exit polling and phone polling," that is
> such an improbable sounding claim that you are going to have
> to provide some pretty convincing documentation to convince
> me of it. Exit polling is widely understood as much more accurate
> than phone polling.

There are many widely understood things that are wrong or misleading.

http://www.mysterypollster.com/main/2004/12/what_about_thos.html
discusses why US presidential election national exit polls are weaker
than some other exit polls.  This entry discusses sources of
systematic error in exit polling:
http://www.mysterypollster.com/main/2004/12/what_is_the_sam.html

http://www.pbs.org/newshour/bb/politics/july-dec04/exitpolls_11-05.html
is an interview with the co-director of the NEP that mentions other
sources of error in the 2004 US exit polls (search for "margins").

Perhaps most tellingly, http://www.exit-poll.net/faq.html#a15 says
that the margin of error for the US national exit poll is +/- 3%, and
for individual states it is +/- 4%.

Most telephone polls are structured to +/- 3% error margins even
within a state.  For example,
http://www.quinnipiac.edu/x11379.xml?ReleaseID=492 claims just over 3%
error margin for each state, and considering the poll's undecided
voters, the official results from the three states are within those
margins.

http://www.harrisinteractive.com/harris_poll/index.asp?PID=516 cites
online and phone polls with error margins at or better than 3%.  The
phone polls in that and other Harris Interactive polls are close to
the official results, but the online polls had systematic bias in John
Kerry's favor.

It is clear that some polls (like the German national exit polls cited
by Mark Blumenthal) are superior to both the exit and phone polls done
in the US, but there are also a lot of inaccurate exit polls out there
-- my original post mentioned a few non-US cases.  Quality differences
seem much more a result of sample sizes or other processing strengths
than of the poll location.

Michael Pooles