Both politicians and the populace may need to exercise a little more caution before trumpeting the latest poll numbers.
Polls have become a major focus of media attention in this election, though the accuracy of some is questionable. Experienced pollsters said that they were concerned with the sheer number of polls that are appearing.
"The problem is that there are too many being done," said Jefrey Pollock, a pollster for democratic polling institution Global Strategy Group. "They don't follow a rigorous methodology."
All polls have a great potential for inaccuracy if researchers do not properly consider the many complexities of sampling or accounting for a low response rate. Previously unknown polls are more susceptible to these errors, or they may have their own specific agendas and are therefore biased.
"We have a lot of private polling companies that have decided to get into the yacht for their own marketing purposes," said Terry Madonna, director of the Keystone Poll that analyzes Pennsylvania. "I don't know to what end we need more than eight or nine national polls. There are enough there that we can use the average of the polls."
Pollsters say a good sample group is the key to a poll, so it must be representative of the population as a whole. Recently, the well-established Gallup Poll has come under criticism from several Web logs on the grounds of over-sampling Republican voters and therefore reporting results that were unrealistically in favor of President George W. Bush.
Some pollsters, however, disagree with this judgment. Pollock said he did not believe there was anyone systematically sampling Republicans to swing the poll their way, and Madonna dismissed all such claims.
"My own belief is that, over the years, there's probably no one who's been more open about their research than Gallup," Madonna said. "I'm not convinced that Gallup's wording is flawed particularly. ... I haven't seen anything that makes me doubt it."
Polls can often be misleading or confusing to readers because generally only 1,000 people are interviewed to establish the opinion of 200 million.
This is not necessarily a problem, however, if the sample is representative of the larger group. Statistics professor Bob Stine explained that two large samples will have common characteristics, no matter the size of the overall population.
"The bigger the sample, the less the average bounces around from sample to sample," he said. "If the samples are similar, they must be similar to the [composition of the] population."
Likely voter screens -- which remove, among other criteria, individuals who have not voted in the recent past -- are another potential pitfall for polling reliability. Communication professor Kathleen Hall Jamieson, the director of the Annenberg Public Policy Center, noted that many voters may not be included in polling, even if they were previously under voting age.
"I don't trust the use of likely voter screens this year," she said. "A lot of new voters are being registered, and the likely voter screen will screen them out. ... It assumes that someone who hasn't voted before will be screened out."
Yet in this election, even the major national polls are not necessarily the best method to track the likely outcome of the election, considering that there are only 17 swing states.
"There's so much focus on national numbers, when strategy is really driven at the state level," said Josh Starr, a former pollster for Bill Clinton. "You can do 600 [people] nationally, or look at 600 in each state. [The state is] where you keep your attention."
