http://www.upi.com/view.cfm?StoryID=20040928-050716-3788r
The Web: Problems with online polling
By Gene Koprowski
United Press International
Published 9/29/2004 7:38 AM
CHICAGO, Sept. 29 (UPI) -- An online service offers a poll to
its subscribers, asking them whom they prefer for president of
the United States in the Nov. 2 election -- George W. Bush or
John F. Kerry. A sample of 250,000 "votes" is taken
and a landslide is declared for one of the candidates. Is the
poll more accurate than a small sample of public opinion,
interviews with 1,000 people, taken by a behavioral scientist,
over the telephone?
Probably not, experts told United Press International.Some
forms of online polling are not scientifically sound, and
calling an important presidential election based on the
results of these online surveys -- at least now, in 2004 -- is
risky. The science of statistics and probability is an
important factor in discerning between the soundness and
unsoundness of polls. Moreover, polls are most effective when
they are done randomly, eliminating the chance participants
will game the system, experts told UPI."Randomness is
needed to reduce bias in surveys," J. Michael Dennis,
vice president and managing director of Knowledge Networks
Inc., an online polling firm in Menlo Park, Calif., told UPI.
"There's a big risk with a volunteer panel --
self-selection biases. Those who are on the Internet are
generally more informed about politics than the average
American. That changes the results of the survey."
Frequent online visitors also are more likely to have a higher
income than the general population -- enabling them to afford
the monthly fees to access America Online, Earthlink, and
other Internet Service Providers.That kind of wealth bias in
polling was seen in the 1930s, when telephone polls first were
conducted for a presidential election. "Even with large
sample sizes, like the 250,000 cited in your example, you fall
into the trap that pollsters fell into 70 years ago, with a
sample size more than 10 times that," J. Patrick McGrail,
assistant professor of communications at Susquehanna
University in Selinsgrove, Pa., told UPI. "They failed to
predict the victory of Roosevelt. That's because they used
people who owned a telephone -- an expensive and relatively
rare device back then."
An analysis by RealClearPolitics.com of scientific polls,
taken by leading news organizations, showed President Bush,
for the week Sept. 20-27, holds an average 5.6 percentage
point lead over Kerry in the race for the White House.
That average is generated by examining polls, such as the CBS
News poll showing Bush beating Kerry 51 percent to 42 percent,
as well as other, closer polls, such as the Fox News poll that
shows the president ahead of the Democratic nominee 46 percent
to 41 percent. Another poll in the mix, conducted by
Investor's Business Daily, shows the race in a statistical
dead heat, at 45 percent for each main candidate, with Ralph
Nader generating 2 percent support."These polls represent
households representative of people across the United
States," Dennis said. "What the pollster has to look
at are factors like age, gender and education. It's also
important to look beyond that -- for attitudes, knowledge and
behaviors -- and to determine if it is representative along
those lines as well. This is a way to bridge the digital
divide online."
Some pollsters have tried to eliminate the online bias by
screening samples of online voters and then randomly selecting
them. Such polls have been used successfully on the Internet
for test marketing and test branding of new products, helping
major companies determine what the best brand strategy might
be for a particular product, or what features to add, or
eliminate, in a new technology.
Knowledge Networks will be using a hybrid online polling
technique Thursday night to poll 200 undecided voters for CBS
News during the course of the first presidential debate.
"These people will take a survey of attitudinal
questions," Dennis said. "We will see where they
stand before the debate. During the debate, we will register
their warmth or coldness toward a particular candidate. That
will be relayed to CBS in real-time for their
analysis."Even with provisions for making the online
surveys random and representative, some experts remain
skeptical."Our feeling is that these polls are just about
worthless, have no accuracy whatsoever," Michael Hamill
Remaley, a director of Public Agenda, a non-profit research
organization in New York City, told UPI.
A number of experts told UPI they were concerned account
holders at AOL and other online services could create multiple
account names and vote several times with different nom de
plumes."Allowing users to vote only once hardly qualifies
as a proper control," said Dana Harsell, assistant
professor of political science at Hartwick College in Oneonta,
N.Y. "AOL lets its subscribers create many sub-accounts,
so it is possible to vote multiple times on accounts that a
person created for their personal correspondence, their home
office, or even their children."
Though some online polling techniques might not be scientific
enough for many of these analysts, the technology has proven
quite effective at harnessing other types of public opinion.
On the Democratic National Committee's Web site,
Democrats.org, the DNC notes it has "been conducting a
major petition drive in partnership with MoveOn.org." The
drive, conducted last fall, generated more than 300,000
signatures, and more than 172,000 of them in one, 36-hour
period, the DNC said.In addition to petition drives, there are
other effective techniques to monitor public sentiment online.
Accenture Technology Labs, a unit of Accenture, a consulting
company, has developed a tool that "intercepts and
instantly analyzes" Internet chatter about political
races, new products or proposed public policies, said Ed
Trapasso, a spokesman for Accenture in New York
City."Some tests have shown the tool to be more accurate
than Internet polling," he added.
Pundits are not completely writing off online polling, however
-- just yet. They reckon the technology will be improved with
time. "I wouldn't take the position that online polling
is awful and never going to be any good," John McIntyre,
founder of RealClearPolitics.com, an online political news
site in Chicago, told UPI. "It might be something that
they can figure out in 2006, 2008, or 2012."
--The Web is a weekly series examining the global
telecommunications phenomenon known as the World Wide Web.
E-mail sciencemail@upi.com