Re: [MLUG - DISCUSSION] [POLITICS] Was the 2004 Election Stolen?

["Jonathan King" <EMAIL:PROTECTED>]

On 6/2/06, Vern Green <EMAIL:PROTECTED> wrote:
> Admittedly I did not read the entire thing, it was a little winded.
>
> As for exit polls though, they suck as a measure as who will win in a
> national in a country as large as the US. They seem to work alright in
> Germany, England, and maybe Haiti or whereever, face it Germany is
> what, the size of California?

I'm not sure how your argument makes any sense.  The essence of an
exit poll is to take a stratified random sample of a population of
interest (e.g., people who have just voted) and see whether you can
predict the outcome from tallying the whole population.  I admit it's
disquieting, but it's pretty much true that if I RANDOMLY picked
10,000 ballots out of a million actually cast and counted just them,
it would have to be a really close election for me not to figure out
who the winnner would be.  The surprising fact is that it is the size
of the sample that really matters, not the size of the population.

Now, an exit poll differs from my ballot box sample in two key
respects.  The first is that I am not randomly sampling from
individual voters.  Usually I'm sampling precincts first (my guys only
show up at some polling places, and not all of them) and then within
precincts, picking random people who come out of the polling place.
If we sample enough precincts, we can become surer and surer that our
precinct sample is representative of the whole population of
precincts.  Similarly, as we sample more and more people from each
precinct, we get a better idea how that precinct will vote, and by
combining our totals across larger units, how states, regions, and the
whole country will go.

The second difference between exit polling and ballot box sampling is
that you are asking people a question, and they can refuse to answer
you, or they could even lie.  Either of these things could present a
real problem if the probability of them giving you no data or bad data
is not independent of how they voted.  Lying is presumably the worst
problem, since you cannot detect it or adjust for it.  Declining to
participate is also a concern, but because you can at least identify
the gender of everybody leaving a polling place, you can make sure
that you're aware of potential problems like having women
disproportionately refuse to participate.  Similarly, you'll become
aware of the fact that nonparticipation is higher in some precincts
than others, and attempt to adjust for that as well.

> Look at who performs exit polls in this country? And consider where
> most of the polling is done? I am sure there are all kinds of
> convoluted formuli out there to take numbers generated in one
> geographic area and figure out what they might be in another
> geographic area, but it is still a "best guess".

It's not the actual vote count, but it's more than just a guess in
most situations.  It's a really carefully formulated case if the
polling people are doing their job right.

So let's look at the case of the Gray Davis Recall/Schwarzenegger
election.  The exit polls had the recall vote at 55% for and 45%
against.  The actual result was 54% for and 46% against.  Again, if
the election had been really close, the exit poll wouldn't have been
conclusive, but we would have known this.  The Kerry/Bush race was
obviously closer, but when the exit polls have one guy up by 2.5% and
he loses by 2% and the MOE is 1%...something weird happened.  What
that weird thing was, we still don't really know for sure.

jking

_______________________________________________
discussion mailing list
EMAIL:PROTECTED
http://mlug.missouri.edu/mailman/listinfo/discussion