Don't Trust the National Polls (posted October 21, 2000; reprinted from October 19, 2000 Newsday)
The 2000 elections may discredit the polling industry in as dramatic a fashion as the 1948 presidential contest did. That year, pollsters incorrectly forecast a victory by Republican Thomas Dewey over Democrat Harry Truman and were embarrassed by the ensuing result.
Part of the problem was that survey researchers were so confident of their methodology that they quit polling a week before the election and thereby missed a last-minute vote swing to Truman.
No one expects pollsters to quit polling before Election Day, 2000, but there already are troubling signs that this year's polls may be way off. For starters, there has been little consistency in poll results over the last month. Surveys conducted at virtually the same time have shown results that differ by as much as 20 percentage points. The bipartisan Battleground survey conducted by Internet website Voter.com consistently has shown better results for Bush than those undertaken by other media outlets. For example, in mid-September when a Newsweek poll showed Gore up by 14 points, Voter.com put Bush in the lead by 5 points.
Similar discrepancies have occurred in regard to the debates. Following the first presidential debate, network polls varied enormously in summarizing who won the vote. While all four network surveys showed Gore had done better, the actual margin varied from +3 percentage points on ABC and +7 on CNN to +10 on NBC and +14 on CBS.
Performance was no better following the second debate. While ABC showed Bush winning the debate by 16 percentage points and CNN showed him besting Gore by 13 points, CBS placed Bush's margin at only 3 points.
Part of the problem with pre-election polls is the difficulty of predicting the actual turnout. There seems little doubt that turnout will be low this year, even with a very competitive presidential campaign. Debate viewership has dropped nearly 50 percent below 1992. The number of people watching the most recent presidential debate was the second smallest television audience in debate history. Primary turnout this year also tended to be low with barely one in 10 voters going out to the polls.
In a situation where turnout is small, likely voters will be a little more Republican than typically is the case in a presidential campaign. One of the reasons why the Battleground survey has produced results more favorable to Bush is that in anticipation of a low turnout election, its organizers have been oversampling areas with traditionally high turnout. This produces a sample that leans a little more Republican in its preferences.
Response rates also have fallen this year, according to national pollsters. Pollster John Zogby, for example, has noted that 16 years ago, 65 percent of those called agreed to participate in the survey. Now, response rates are down to 35 percent. With response rates this low, there is greater sampling variability from survey to survey, depending on who happens to be home during the time pollsters call.
It is no accident that in 1996, where we had the lowest turnout in a presidential election in nearly 50 years, pre-election polls underestimated GOP nominee Robert Dole's vote by 7 percentage points. The last Gallup survey that year, for example, put the race at 52 percent for Clinton, 34 percent for Dole, and 10 percent for independent candidate Ross Perot. Clinton's actual margin over Dole was 49 to 41 percent, which means the survey was within the margin of error for the Democratic, but not the Republican vote.
Low turnout leads to pre-election polls that inflate the size of the Democratic vote. With Vice President Al Gore running neck and neck with Republican nominee George W. Bush, despite a strong economy, public opinion polls at the presidential level may be lulling Democrats into a false sense of complacency about their campaign competitiveness.
