The quality and accuracy of survey work over the past decade has declined as the Internet has become an increasingly seductive tool for quick, cheap data collection, said a methodology expert speaking at Cornell, March 1.

Jon A. Krosnick, director of the Stanford University Methods of Analysis Program and professor of communication, political science and psychology, gave the Cornell Survey Research Institute's first annual lecture, "The Brave New World of Survey Research: One Surprise After Another," in Cornell's School of Industrial and Labor Relations Conference Center.

Survey data can have a vast influence on informing policy and other important decisions -- which is why people should take the time to participate, he said. However, the accuracy of information surveys can vary greatly, depending on how data are collected, he said.

When it comes to Internet surveys, for example, "It seems we have unlearned a lesson that was good to have learned a hundred years ago," Krosnick said. And that is, the lesson of using a representative sample of the population of interest.

Why, asked Krosnick, should the Internet make this any different? Because, he said, few Internet survey companies strive to attain a representative sample, even though studies show that doing so reaps significantly more accurate results.

Another issue with Internet survey work, he said, is the potential for low response rates. For example, big media outlets often provide attitude poll results almost instantly after national events such as the State of the Union Address, but they never report their response rates because it would devalue the data, Krosnick said.

Struggling to attain the highest response rate, however, might not be as critical as researchers previously assumed, he said. A University of Michigan study, for example, found no relationship between response rates and accuracy; in fact, the study found that lower response rates actually increased the balance of respondents of different gender, age, income and education level.

Nevertheless, to mitigate low response rates and representative sample problems, some companies pay respondents for each survey they complete and for their monthly Internet access and provide a Web/TV setup if a household does not own a computer. In a pilot study aimed at obtaining high participation rates, Krosnick combined this method with personal requests from surveyors to participate, and 67 percent of households that were approached signed up. In the future, Krosnick said, researchers will offer free laptops in an effort to increase participation by a further 10 percent.

The most accurate survey data collection method, Krosnick said, is the face-to-face interview, though it can be very expensive. For example, a one-hour survey in the home of a representative sample member costs $1,000. For important federal surveys where stakes are high, the government will pay this, he said, because this method is significantly more accurate than any other.

The next best method, Krosnick said, is computer and phone surveying, which is better than phone alone; the least accurate method is pencil-and-paper mailed surveys, which cost about the same as phone surveys but are less accurate. In other words, Krosnick said, when it comes to data collection, you tend to get what you pay for.

Graduate student Sara Ball is a writer intern at the Cornell Chronicle.