PENNSYLVANIA POLLS AND THE 1996 PRESIDENTIAL ELECTION
G. Terry Madonna and Berwood Yost, Center for Politics & Public Affairs, December, 1996.
Presidential campaign surveys conducted for news organizations in Pennsylvania provided reliable assessments of the recent presidential campaign. Most of the news organizations which poll usually do so twice during a campaign cycle, once about four to five weeks before election day and again one to two weeks prior to the election. Polling results can vary sharply from one polling period to another. Not so in 1996. President Clinton held a consistent lead in polls conducted during both periods. Slightly less than half of the state's voters consistently indicated they would vote for him. If the eight surveys conducted during the entire cycle are averaged, they show Clinton's support at 49%, which was the actual vote total he received on election day in the state. While Clinton's support remained stable, Dole's support improved in the waning days of the campaign, and even Ross Perot picked up a few percentage points in the final voter tally. Polls conducted later in the campaign caught some of the Dole and Perot movement.
As momentary snapshots of the electorate, these polls were accurate. What these polls could not account for was the decision-making process made by many voters after the polling cycle ended. According to the Voter News Service Exit Poll for Pennsylvania, about 16% of the state's voters made their choice in the final week of the campaign. In fact, most of these voters actually decided during the three days preceding the election. For the eight polls we studied, the average undecided vote was ten percent. These last-minute deciders were more supportive of Bob Dole and Ross Perot than those who had made a vote choice earlier in the campaign: 38% voted for Clinton, 41% voted for Dole, and 20% voted for Perot. This should remind us that a survey is only a picture of the electorate at a point in time. Only when the pollster uses some model to predict the election outcome, which is unrelated to how well a survey is conducted, can survey data be used as a predictor.
This leads to an important and unresolved question for those who try to predict election outcomes from surveys. How does one count undecided voters or weak supporters for predictive purposes? There are several techniques for assigning undecided voters to a particular candidate, but none are error proof and all involve assumptions about how undecided voters make up their minds. Unfortunately, what compels undecided voters to choose a candidate varies from election to election. For some issues, public a ttitudes are stable and opinions change infrequently or only at the margins. But election campagins by their nature strive to change voters' perceptions about candidates, especially the perceptions of undecided voters. A majority of undecided voters normally cast their votes for the challenger(s) when an incumbent's popular support is below 50 percent, which is what happened this year. The Pittsburgh Post Gazette, the Pittsburgh Tribune, and the Keystone Polls had Clinton receiving 48%, 49%, and 48%, respectively. The problem in 1996 was trying to figure out what undecided voters would do with their votes. That is, would they act like the voters who already had made up their minds, or would they act differently. In 1996 they acted differently by voting more heavily for Dole and Perot. Clinton remained where he started, with support from slightly less than half of Pennsylvania's voting population. In our view, concerns about the accuracy of the polls are unfounded and predicated on a misunderstanding of what they can accomplish. The real story in campaign year 1996 is whether the 600 milllion dollar presidential campaign made any differenct to the nation's electorate.
