All the polls share a common bias strong enough to make their numbers meaningless: the demand for their product depends on the perception of a "horse race." They can therefore be counted on to adjust their (unpublished) raw numbers by such fudge factors as "demographics" and "voting likelihood" to make their published percentages as close as possible.
An accurate poll would seek to determine the swing from the 2000 election, narrowly won by Gore. It would have a sufficiently large sample (roughly the number of people polled each week by all the polls taken together) and would ask these questions:
1) In 2000 did you --Vote for the candidate of the Republican Party? (Bush) --Vote for the candidate of the Democratic Party? (Gore) --Vote for the candidate of the Green Party (Nader) --Vote for some other candidate? --Not vote in the presidential election?
2) In 2004 do you expect to --Vote for the candidate of the Republican Party? (Bush) --Vote for the candidate of the Democratic Party? (Kerry) --Vote for the candidate of the Green Party? (Cobb) --Vote for the independent candidate? (Nader) --Vote for some other candidate? --Not vote in the presidential election? --Vote for some candidate about whom you have yet to make up your
mind, if you vote at all?
The answers to these questions would give a meaningful and very accurate picture of the actual voting intentions of the electorate. I would be very, very surprised were such a poll to be conducted and published by anyone. Until that unlikely event, all the horse-race polls should be written off as so many tout sheets.
Shane Mage
"When we read on a printed page the doctrine of Pythagoras that all things are made of numbers, it seems mystical, mystifying, even downright silly.
When we read on a computer screen the doctrine of Pythagoras that all things are made of numbers, it seems self-evidently true." (N. Weiner)