...the person or group paying for the poll.
These polls aren't done in a vacuum, and--because there is zero oversight or accountability by the government regarding fair practices--they are wide-open to corruption.
Groups/companies/campaigns/individuals contract to have polls done, and how much influence the entity contracting the polling service on the way the questions are asked, how the questions are worded, etc. is unknown.
The polls have a lot of influence on the outcome of an election because humans are, in many things, herd animals, and especially the mushy middle of the political spectrum puts a lot of value in how the majority of other voters are voting (the belief being that if that many people have that opinion, the opinion must be valid).
Political candidates and organizations spend a good deal of money funding polls, which they use in their campaigns to influence voters.
I compiled some analyses of polls during the 2002 and 2004 elections cycles, and came to the conclusion that the vast majority of polls to which we are exposed cannot be validated based on the way they were conducted and who controls the questions asked and what material is published, and when. Some polls, just by the information given in the spin pieces written about them, can be seen to be flawed without even having to trying to get information from the polling organization.
(
http://www.ncpp.org/?q=node/4) “20 Questions a Journalist Should Ask About Polls” is a good explanation of some of the issue about polls.
There are other issues I didn't see mentioned in this article, and they are important to know when reading about polling results.
*Polling phone calls made in the evening and weekends are more likely to reach conservatives.
*Polling questions asked to those "likely to vote" (voted in the last election) are more likely (by up to 7%) to return a conservative bias. So there is a difference in outcome if you ask questions of "those likely to vote" and "registered voters," and if the former group ("likely to vote") is asked it favors the conservative candidate or opinion.
Here's an example of how the second issue becomes important, especially in a highly partisan political election, where Republican voters woud be highly likely to vote for Republican candidates and Democratic Party voters would be highly likely to vote for Democratic Party candidates.
In the 2004 election, before the primary was held, when Dean was at his strongest, CNN published an article about a poll that they funded (from what I could find out--and polling groups do NOT want to give you enough information so that you cann assess the poll for validity--if they say "CNN/Zogby Poll" it usually means CNN paid for it, and it may mean that they had control over what questions were asked, how they were asked, when the calls were made, how the data was manipulated for results, and which of the polling results were allowed to be published).
The premise of the CNN poll was "If you were to vote now, and your choice was between GWBush against one of the Democratic candidates, who would you vote for?" Then they would ask the question of Bush v. each of the Democratic candidates.
For all of the candidates except Dean--and they listed the non-Dean candidates first in the article--the polling was done of "registered voters." In all of the non-Dean candidates, GWBush won by a good margin.
The last Dem candidate they listed was Dean, and for his hypothetical run against GWBush, the pollers asked the question of "those likely to vote." Right away, there is up to an 8% more conservative outcome. Even with that, Dean was less than 1 or 2 points behind GWBush, and the poll had a margin of error of 3 to 4 points.
Yet the article was entitled, "If Election Held Now, Bush Would Win."
It gets better. The number of people polled was between 1300 and 1400, which is a good sampling, PROVIDED THE POLL IS CONDUCTED LEGITIMATELY. Out of those polled, around 380 were U.S. veterans. Out those veterans, a majority said they would vote for Bush. But this is the problem: 380 is far to small of a sample to predict the outcome of a national election. Also, people who identify themselves as veterans are more likely to be conservative.
Also, out of the 1300 to 1400 people, less than one-third were Democrats. I am sure the poll was weighted (but there are problems with weighting), but the fact that there were so many more Republicans makes me think that the polling calls were made at a time when it was more likely that conservatives would be reached.
So the questions asked of veterans were from a small group of respondents, and out of those (since only a third of the entire poll was Democrat, and since self-identified veterans are more likely to be conervative), the number of Democrats in this small sample of veterans was probably very, very small.
Yet the subtitle, right after, "If Election Held Now, Bush Would Win," was this: "Veterans Vote for Bush by Large Margin."
I spent a great deal of time trying to get information from these polling groups. Zogby got downright pissy when I would not give up in trying to find out specifics, such as when the calls were made, and, more importantly, questions about how much influence the person or group funding the poll had on the questions asked, how they were worded, when they were asked, and what results were published, as well as how the data was manipulated.
Anyway, be wary of the polls, and if your gut feeling tells you that the polling results are flawed, I would suggest you go with your gut feelings because, IMO, there is a very good chance that many polling results ARE flawed.