30 Percent May Decide Senate Race

That'?s lower than either of the two previous elections where Senate candidates did not share the ballot with a presidential race. In 1990, when Sen. Jesse Helms was re-elected over Harvey Gantt, turnout was 41 percent of adults. In 1986, when Terry Sanford beat James Broyhill, 35 percent voted.

Like other voter surveys in North Carolina this year, the Carolina Poll is showing a toss-up between Republican Sen. Lauch Faircloth and challenger John Edwards.

The poll was conducted by the School of Journalism and Mass Communication and the Institute for Research in Social Science at the University of North Carolina at Chapel Hill.

"This poll is showing North Carolina voters a rare opportunity," said professor Phil Meyer, who assisted in its analysis. "There are very few elections where the probability of one person'?s vote making a difference is so high."

The 50-50 Edwards-Faircloth split is based on 411 likely voters who have made up their minds and are not planning to vote for a third-party candidate.

Minor parties drew 2 percent and 14 percent were undecided. The poll did not ask which way people are leaning to reduce the undecided as other polls have done this fall.

Among those not expected to vote but who still gave a preference, Edwards had a four point lead, still within the margin of sampling error. A large turnout would help the Democrat, but voters are signaling him not to expect one.

The Carolina Poll turnout estimate is based on a question that asks, How certain are that you will vote this November 3? Very certain, fairly certain, or not at all certain??

In past elections, the actual turnout has been about half the proportion claiming to be very certain.? This year, it was 63 percent, compared to 79 percent in 1990 and 76 percent in 1986.

Like other pre-election polls around the country, the Carolina Poll has experienced difficulty in getting voters to share their views. The cooperation rate, among all eligible citizens contacted was 39 percent. That compares to 64 percent in the Carolina Poll of fall 1986.

Professor Don A. Dillman of Washington State University, author of the leading textbook on mail and telephone surveys, says that changing attitudes toward unwanted phone calls, and technical barriers like answering machines and caller ID, have seriously reduced the usefulness of the telephone in public opinion measurement.

"What was a mandatory instrument (the telephone), is now a voluntary one,"? he says.

Lower cooperation rates have not, so far, visibly affected the ability of telephone polls to predict election outcomes. Past studies have shown that people who refuse to be interviewed tend to be poorer, less educated, and members of minority groups -- the same categories that are less likely to vote.

But those studies were done when cooperation rates were much higher, and the true effect of lagging cooperation today is not known.

This Carolina Poll was a close match to the age and racial characteristics of the state. It underrepresented males by 11 points, and the data were weighted to account for that.

The margin for sampling error in a poll of this size is less than five percentage points. Other sources of error, including the low cooperation rate, are potentially more serious. The total number interviewed, including likely and unlikely voters, was 589.

Interviews were conducted by students in journalism and political science under the direction of professor Donald Shaw, director of the Center for Research in Journalism and Mass Communication, between Oct. 19 and Oct. 30.

Cooperation rates were calculated by the formula for the minimum rate established by the American Association for Public Opinion Research in May 1998.