Research: Wisconsin primary results could provide strong clue about Texas and Ohio primaries

February 18, 2008

The Wisconsin primary on Tuesday could be a bellwether for primaries in Ohio and Texas, according to analysis by two University of Washington researchers. The findings indicate that in American politics, race still matters.

Two polls in the last few days give Obama a 4 percent edge in Wisconsin, which has 92 delegates. However, Wisconsin has an open primary and a low black population, 6 percent. In recent primaries, those conditions have been associated with the Bradley effect -- and an over prediction of the Obama vote by about 8 percent, according to psychology professor Anthony Greenwald and political science professor Bethany Albertson.

The Bradley effect was identified 25 years ago when Los Angeles Mayor Tom Bradley had a solid lead in pre-election gubernatorial polls but lost a close election to his Republican opponent, George Deukmejian. Results from that and other races involving black candidates indicated that polling tended to overstate support for black candidates compared to their actual vote percentages.

If the Bradley effect occurs in Wisconsin, the researchers say, Clinton can win even though the recent pre-primary polls give Obama the edge. Clinton may then do somewhat better than polls predict in the March 4 primaries in Ohio and Texas.

If there's no Bradley effect in Wisconsin, Albertson and Greenwald said, it's good news for Obama, who might then do better than expected in Ohio and Texas, and achieve victory over Clinton by May.

"The Democratic Party's primary contests have been remarkable for inaccuracy of polls predicting their outcomes," Greenwald said.

Errors of 8 percent or more, quite rare in pre-election polls, have occurred in about half of Democratic primaries this year -- and the errors are systematic, he and Albertson found. The inaccuracies have occurred especially in open primaries, which allow voting regardless of party or status. The average error in open primaries has been 9 percent, compared to 3 percent in closed ones. Also, the researchers found, polls in states with few black voters have overpredicted support for Obama -- which appears to be a modern Bradley effect.

Albertson and Greenwald have also seen the opposite, a reverse Bradley effect: polls in states with relatively many black voters have noticeably underpredicted support for Obama.

"The data show a clear relationship with racial composition of states' populations, but reasons for this are not yet clear," Greenwald said.

This latest research grew out of the Implicit Association Test, which Greenwald invented about a decade ago to measure unconscious roots of people's thinking and feeling. More than 6 million people have taken versions of the test, providing insight on such topics as race, gender, sexuality and various ethnic groups.

Source: University of Washington