[SfN] Gallup Poll

[Paul Riismandel]

Note that in these stats:

At 10:52 AM 10/23/00 -0500, John Paul Schmit wrote:
>Here is today's results for each poll
>Gallup - Bush 50%, Gore 41%, Nader 3% error +/- 4%
>Battleground - Bush 44%, Gore 39%, Nader 5% error +/- 3.1%
>Zogby - Bush 44%, Gore 42%, Nader 5%, error ?
>Portriat - Bush 46.7%, Gore 41.3%, Nader 3.6% error rate +/- 1.8%

the spreads between Bush and Gore barely cover the margin of 
error.  Further, I doubt that these spreads would be considered 
statistically significant by any responsible statistician--though I would 
have to know more about the methodology to say for sure.  Although it's an 
important concept in social science research, you never hear the words 
"statistically significant" with regard to political polls, because they 
don't seem to take account of it, nor care about it.  But taking it into 
account, I'd bet, would render Bush and Gore about equal in the polls.

One other thing  to remember about polls that also gets overlooked is 
selection criteria--who are they asking?  If this is a telephone poll, what 
about people without telephones?  With unlisted numbers?  With cell phones 
instead of wired phones (cell phone numbers are always unlisted)?

One particular problem-- inner city and poorer neighborhoods tend to have 
fewer wired phones than suburban and affluent neighborhoods.  In many 
instances households don't have wired phones, relying on sharing with 
neighbors, payphones or maybe cell phones.  How is this accounted for in 
the poll?  The non-ownership of phones and the replacement of wired phones 
with cell phones are NOT random occurrences and cannot be corrected for in 
an utterly random poll.  Simply, these folks are systematically excluded 
from the poll.  How do you think this might affect results?

But of course the Gore folks would rather you got scared that Nader's 4-15% 
is what's making Gore trail Bush in these polls, and will lose the race to 
Bush.  Don't buy it.

--Paul