Reading Polls Too Much Too Soon

Presidential election polls have become a staple of political commentary and analysis. How much stock should we place in them? At this time in the election season, very little. But tell that to scores of columnists and cable news show hosts who read great significance into weekly shifts in the public standing of Governor George W. Bush and Vice President Al Gore.

Washington punditry is not known for the accuracy of its forecasts, as any casual observer of the stark misreading of public opinion during the Clinton impeachment saga can attest. Performance measurement may be the rage in public management, but it is a concept entirely foreign to Beltway political analysts. We routinely draw on the latest polls (and sometimes our own preferences) to speak with great confidence about the political future, with little fear of anyone remembering months from now what we said, much less holding ourselves accountable. A presidential election seems to bring out the worst of this practice. In hopes of restraining myself and others from falling into this pattern once again, I offer some historical lessons about public opinion polls, economic performance, and presidential elections, drawing from the work of my scholarly colleagues.

First some basics: Most individual votes are determined by underlying attachments to a political party. Swing votes are strongly influenced by broad forces operating on the electorate, including the state of the economy and the level of satisfaction with the incumbent administration. Since most citizens ordinarily pay little attention to politics and public affairs, campaigns are important mainly in mobilizing fellow partisans and in capitalizing on these broad forces in the election. This takes time, and highly visible events like national party conventions and presidential debates. Persuading citizens to vote against their party, or against the current of those national forces, is decidedly an uphill struggle. The quality of the candidates and their ideological positioning relative to the average voter can make a difference, but the impact is usually modest at best. More often, campaigns of the opposing candidates largely neutralize one another. Finally, a two-term administration seeking to extend its hold on the White House typically runs into some “time for a change” sentiment, but the salience of that sentiment is shaped by conditions in the country and retrospective assessments of the current administration.

Now on to the lessons. Trial-heat polls taken before the late summer of an election year are notoriously unreliable in anticipating the November victor—in other words, not much better than flipping a coin. Recall that in June 1980 President Jimmy Carter enjoyed an 11 point lead over Ronald Reagan. In June 1988 Michael Dukakis was up 14 points over Vice President George Bush. Of course, both Reagan and Bush took full advantage of the broad forces working in their favor and won decisive victories.

By late summer, however, the polls do begin to gel. In every election between 1948 and 1996 in which the president had a July approval rating exceeding 50 percent, his party won in November. And every nominee since 1952 who led by more than 53 to 47 percent in the post-convention polls won the general election.

The performance of the economy is the single most important determinant of the reelection prospects of the incumbent party, but its impact is complicated. Election forecasts based on economic growth during the first half of an election year are accurate only if they are combined with subjective measures from public opinion data. Public perceptions of economic performance tend to lag objective indicators—it takes a while for citizens to believe the economy has recovered or that budget deficits have turned into surpluses. So economic conditions throughout the incumbent president’s term are important. And public evaluations of which party will better maintain prosperity in the future are also related to success in November.

What, then, does this tell us about the shape of the 2000 presidential election? Most importantly, conventional wisdom in Washington has relied much too heavily on early polls and on the imagined personal strengths and weaknesses of the candidates, and on their supposedly fateful campaign decisions, during the winter and spring. The historical record suggests there was no credible basis for proclaiming Governor George W. Bush the frontrunner last summer and fall, when he enjoyed a 15-20 point lead in the polls, nor is there today, when his lead has shrunk to some fraction of that.

The long economic expansion, exceptionally robust growth in the first half of 2000, public confidence in the economy, and President Clinton’s job approval ratings consistently above 55 percent over the last several years suggest a favorable referendum by the public will outweigh any “time for a change” sentiment. The prediction of one of the most accurate forecasting models taking these factors into account gives Gore the presidency with 53.3 percent of the two-party vote. Not a landslide by any means, but a comfortable popular vote victory and a decisive majority in the electoral college. Other models generally anticipate a similar result.

My point is not to declare Gore the inevitable winner, but to set a realistic baseline for subsequent reporting and analysis, and to urge caution in relying too much on early polls. Based on historical patterns and current conditions, the election should be viewed neither as a likely Bush win nor a dead heat but instead as Gore’s to lose. Whether and how he might manage to snatch defeat from the jaws of victory is a fascinating subject for another time. But if Gore is defeated, it will be a first in American electoral history.