Polls’ accuracy turns on turnout
by Phil Burgess, Unabridged from the Rocky Mountain News, October 31, 1996
Public opinion polls have dominated the 1996 election cycle. One reason: There are simply more polls. Example: According to the Roper Center for Public Opinion Research, there were 10 presidential “trial heats” between September 1 and election day in 1968. This year that number will exceed 125.
Even worse, the horse race aspect of the presidential contest has occupied a lot more air time than the issues. The same pattern holds for print media, where accounts have given lopsided attention to reporting and analyzing polling results from trial heats (“Who would you vote for if the election were held today?”) and what demographic groups are supporting which candidate — rather than polling results about issues which, at least, might lead to some discussion of issues and civic education.
If we believe all these polls, where incumbent Bill Clinton has held a steady double-digit lead over challengers Bob Dole and Ross Perot, then we must conclude that Bill Clinton is on the verge of an historic landslide victory on November 5. However, there may be a flaw or two in forecasts of a run-away election victory for Clinton next week. One of those flaws may be voter turnout.
Estimating voter turnout is a real problem in a year when voter apathy and disengagement appear to be at an all-time high. For example, Curtis Gans, director of the non-partisan Committee for the Study of the American Electorate, says “interest in this election is a full 15 percentage points below what it was in 1992.” All this despite the fact that the new motor voter law has generated five million new voters and there have been unprecedented voter education and mobilization efforts by everyone from the Christian fundamentalists to the Democracy Project of PBS and the First Vote project of People for the American Way. According to Gans, all this should be “a recipe for a high turnout election.”
In fact, Gans and most other political observers see a low turnout. Indicators: Audiences for the nominating conventions of the Democrat and Republican parties were at an all-time low. Network coverage of politics and the presidential campaign, measured in minutes, is down 40% from 1992.
These expections for 1996 follow a long-term trend where voter turnout in American presidential elections has declined steadily from an all-time high of 62.8% in 1960, which pitted two young contenders — John Kennedy and Richard Nixon — in an historic election where excitement was increased by the first-ever televised debate. Since then there has been a steady decline to 50.1% in 1988 (George Bush vs. Michael Dukakis) followed by a slight increase to 55.2% in 1992. There is a general consensus among analysts and practitioners that turnout in 1996 will come below the 1992 percentage — and some believe it may fall below 50%.
Increasing cynicism about government and other factors leading to lower turnout are well-known. But whatever the causes, estimating turnout creates a special problem for pollsters and a challenge in interpreting the polls. Reason: While the left-brained science of polling is well-established — sampling, formulation and placement of questions (called “instrumentation”), interviewing techniques and the like — estimating voter turnout is a mixture of art and science. Yet these estimates are very important because they determine which of those 400 or 800 or 1,500 people interviewed will be counted in the final tally. So, many of the results we are seeing now, which assume a “moderate” turnout, can be substantially affected by how many actually turn out on Election Day — and right now it looks like about half the eligible voters will stay home. That could narrow the gap considerably.