October 27, 2004

Purdue survey expert: Numbers in political polls may be deceptive

WEST LAFAYETTE, Ind. – Most political polls can't predict the outcome of presidential elections because such polls are based on individual preferences and don't equate to Electoral College votes, says a Purdue expert on survey methodology.

"The point estimate has nothing to do with the results in a national election," says John Stahura, a professor of sociology and director of Purdue's Social Research Institute. "State polls mean more, because they're based on popular vote and can actually be a pretty good predictor."

Individual polls could be extrapolated by state to get an Electoral College estimate, but a sample of 600 to 1,000 people for each state would be required. Most big polling companies use at most 100 respondents from any one state in national polls and usually use fewer than 100 to none from most states, Stahura says.

But the biggest problem Stahura sees in the methodology of national polling companies is that they don't report response rates, which could miss polling up to 70 percent of the people who are called. With response rate that low, it's difficult to project the results to the population, he says.

"Many of the larger commercial polls report confidence intervals or 'margins of error' but they don't publish response rates," Stahura says. "They predict the outcome of an election with a 20 to 30 percent response rate."

In sociology, many journals require a 50 percent response rate before they will publish a survey's findings, Stahura says.

Many of the larger polls often discriminate against young voters when they define "the likely voter," a term used to define survey respondents. New voters, especially first-time voters, are less likely to be identified as likely voters by the various screening questions used by different organizations, he says.

Polls relying on phones also miss those who only use cell phones. That group usually is made up of young people with modest incomes who are disproportionately minorities, making up about 7 percent of the population. Also, in general, young people and minorities tend not to participate in phone surveys, Stahura says.

And forget about online polls.

"Online polls are garbage. There's no way to do scientific sampling with online polls," Stahura says. "Many online polls are done in such a way that one person can respond multiple times. In addition, online polls have poor response rates, usually less than 5 percent. Online polls may be the future of survey research, but are not today an effective tool for national polling."

CONTACT: Stahura, (765) 494-4694, stahuraj@soc.purdue.edu

Writer: Maggie Morris, (765) 494-2432, news.writer@purdue.edu

Purdue News Service: (765) 494-2096; purduenews@purdue.edu

To the News Service home page