Counting Votes Is Hard

Counting Votes Is Hard By KEN DAUTRICH Hartford Courant September 28, 2008 You see the results of the latest poll, and so you think you know who's ahead and who's behind in the race for president? Think again. There have been plenty of polls taken in the 2008 campaign since the New Hampshire primary. More than 40 organizations in all have conducted hundreds of polls nationally and in the 50 states. Historically, pollsters have pretty much gotten it right in presidential races — notwithstanding the Literary Digest Poll's prediction in 1936 that Alf Landon would defeat FDR, and the Gallup Poll's 1948 prediction that made its way to the Chicago Tribune headline "Dewey Defeats Truman." Flaws in the methods of Literary Digest and Gallup accounted for their bad predictions in those years. Is 2008 shaping up to possibly be one of those elections where polls are giving us a distorted view of what voters will do on Election Day? If the polling problems in many Democratic primaries earlier this year are any indication, pollsters may have a lot of explaining to do on election night. Don't forget this year's New Hampshire surprise, when all the major polls predicted that Obama would win — some by a much as 10 points. Hillary won by 3 percent. Then there was South Carolina, which the polls said was close. In fact, Obama cleaned up there. There was Connecticut, where some polls showed Hillary in command. Obama won. In Rhode Island, Texas, Ohio, Pennsylvania and some other states, Hillary out performed and Obama under performed many of the polls. The inaccuracy of many polls in the primaries is one thing — pollsters will tell you that trying to figure out who will turnout in a primary election is difficult, and can lead to faulty predictions. But the wide variation in national and state polls also suggests something is going on. Just in the past week, a number of national polls have shown anything from an 8 percentage point Obama lead to a 2 percentage point McCain lead. So why might the general election polls in 2008 be so volatile, inconsistent and maybe downright inaccurate? Why are we in for a possible big surprise on election night, with pollsters scrambling to explain what went wrong, as they were in 1936 and 1948? One possible explanation is the so-called "Bradley effect." In 1982, polls in California predicted that Los Angeles Mayor Tom Bradley would win the governor's race. Bradley, an African American, lost to a white candidate. Since then a number of races with African American candidates had a similar result. A portion of white voters (mostly white male Democrats) tell pollsters they'll vote for their party's candidate, but won't actually cast a ballot that way on

Election Day. Some argue that a few of Hillary Clinton's surprise wins may be accounted for by the "Bradley effect." Another problem for pollsters is the proliferation of cellphones. By Election Day, about one-third of all households will either primarily, or only, use cellphones. But cellphones are not typically in pollsters' call lists. List of cellphone numbers are expensive, hard to construct and cell users aren't very happy about spending their minutes with a pollster. Since younger people are more likely to use cellphones, they are often misrepresented in the polls. So, if the youth vote is high as some expect and they are under-represented in the polls, that may lead to a Nov. 4 surprise. Yet another problem is declining voter participation in polls. The problem is bad and getting worse. Busy lifestyles, annoying telemarketers and increased suspicion about the motives of an unfamiliar caller asking for information have led people to be less cooperative with pollsters. In addition, caller ID and answering machines have made it easy for voters to screen calls. The average participation rate is now about 20 percent, and often much lower. As participation declines, polls become more subject to serious accuracy problems. Another problem: Most polls are funded by media organizations, which all have experienced budget crunches. The media still love to report the polls, but can't pay the price of conducting a quality poll. So in many cases, quality standards have dropped (earlier this year, for example, only one of the 40 organizations that were conducting polls in the primaries was even keeping track of participation rates). Anyone can ask questions, get answers and tally the results. Unfortunately, the cost of a quality pre-election polling is upward of $20,000 to $30,000. Most organizations don't have the budget for that, but they can readily find polling organizations to do a poll on the cheap. After the Literary Digest debacle in 1936, George Gallup and Elmo Roper helped pollsters understand that accurate polls require scientific sampling, which the Literary Digest didn't do. And after Truman defeated Dewey, pollsters learned that polling must continue right up to Election Day to catch last minute trends. (Gallup stopped polling two weeks before Election Day in 1948.) Will election night 2008 be a repeat of 1936 or 1948 for the pollsters? If so, we already know the reasons why. >> Ken Dautrich is an associate professor in the Department of Public Policy at the University of Connecticut, where he teaches survey research.