THE “TIME FOR A CHANGE” FACTOR

In an article published in this journal in 1988, Alan Abramowitz called attention to an important regularity governing presidential elections: The longer a party has been in the White House, the more difficult it is to win the next presidential election. In particular, there seems to be a big difference between the situation that confronts a party that has only been in the White House for the last four years (i.e., the situation of George W. Bush in 2004 or Barack Obama in 2012), and the situation a party faces when it has held the presidency for eight or more years consecutively.

As shown in table 1, since 1900 there have been 12 presidential elections in which the party of the incumbent president had only been in possession of the White House for four years. In 11 of those 12 cases (92%), the incumbent party won the election. By comparison, there have been 17 elections in which the incumbent party had held the presidency for at least the last eight years. In this type of election, the incumbent party was only victorious seven times (41%). As shown in the bottom half of table 1, the predictive power of what Abramowitz called the “time for a change” factor looks even more impressive if we limit our attention to presidential elections held since 1952. This one variable correctly predicts the winner in 14 of the last 16 presidential contests.

Table 1

How Long a Party Has Been in the White House Affects How Likely It Is to Win the Next Election

Why do the odds against the incumbent president’s party increase so dramatically after it has been in power for eight or more years? This question has received less attention than it deserves, but an important part of the answer concerns the public’s readiness to “cut some slack” for the incumbent president and his party and to accept their excuses for poor performance. Confronted with various problems that have not been solved, a party that has only been in the White House for one term may reasonably say, “Our policies have only had a couple years to work. We need more time.” After eight years in office, such excuses sound increasingly hollow. In 2012, for example, the American economy was, by almost all objective measures, performing below the level normally needed to guarantee a president’s reelection (see, in particular, the data presented in Campbell Reference Campbell2013). Fortunately for the Democrats, Barack Obama was able to assuage the voters’ dissatisfaction by the stratagem of blaming the problems on his predecessor. According to the 2012 National Election Pool exit poll, 53% of voters said that Bush was “more to blame for current economic problems”; just 38% held Obama at fault.^{Footnote 2} It is unlikely that voters will be quite so forgiving in 2016 if the economy continues to perform poorly and the Democratic presidential nominee still tries to blame the problem on George W. Bush.

Abramowitz’s theory has come to be widely accepted—sufficiently so that a “time for a change” or “number of terms in the White House” variable is now included in many other forecasting models. The more important point for now is that, as straightforward and commonsensical as this finding may now appear, it was entirely unnoticed until Abramowitz wrote his seminal article in 1988. Score one for election forecasting.

THE IMPORTANCE OF RETROSPECTIVE VOTING

How do voters make reasonably intelligent voting decisions, given that most of them are not terribly interested in politics, follow it only intermittently, and therefore do not have a great deal of information about the candidates and their issue positions? Political scientists have developed a number of major answers to this question, but one of the most popular is the idea of retrospective voting. According to this theory, voters are less concerned with a candidate or party’s promises about future policy than with their past performance in office. Where information about campaign promises is costly to acquire and difficult to evaluate, most citizens develop relatively solid perceptions about the performance of an incumbent president and his administration simply by going about their normal lives and paying a minimal amount of attention to the news.

The most admired and theoretically rigorous treatment of this topic is undoubtedly Morris Fiorina’s Retrospective Voting in American National Elections (1981), but the strongest empirical evidence in support of the theory, at least with respect to presidential elections, comes from election forecasting models. While forecasting models have a checkered record in predicting upcoming elections, they clearly do a good job in accounting for past results. Most of the regression equations that generate forecasts have an R² value well in excess of .80, meaning that they explain at least 80% of the variance in past election outcomes. With only a few exceptions, the variables in the forecasting equations are entirely retrospective. They measure various characteristics of the incumbent administration—in particular, its economic performance, the president’s early summer approval rating, and the number of consecutive terms his party has occupied the White House—but say nothing about the quality of the opposition party candidate or, if the incumbent president is not running for reelection, about the incumbent party’s candidate either.^{Footnote 3} Most forecasting models, in short, are based on the assumption (often made explicit) that presidential elections serve as a referendum on the performance of the incumbent administration. That they are even roughly successful in their forecasts is strong testimony to the accuracy of this assumption.

THE RHYTHM OF THE FALL CAMPAIGN

Shortly after every midterm election, and sometimes well before that, Americans are treated to a mind-numbing succession of polls that pit the likely presidential contenders against one another: Nixon vs. Muskie, Humphrey, McGovern, and Wallace; Carter and Kennedy vs. Reagan, Ford, and Connolly; Reagan vs. Mondale, Glenn, et al. Taken a year or more before the actual election, we sense that such polls probably do not tell us anything very valuable. But when do the polls become meaningful? When do they finally start to give us a reliable indication about whether the upcoming election will be close or lopsided and which candidate is the favorite?

Political scientists had no solid answers to these questions—or even any good guesses—until Campbell and Wink’s article “Trial-Heat Forecasts of the Presidential Vote,” published in 1990. What Campbell and Wink showed is that polls taken even in the summer before the election tell us virtually nothing about the final outcome. But the polls become dramatically more accurate right after Labor Day, and do not get significantly more accurate after that. The national conventions, even in their current, greatly attenuated state, seem to play a major role in “setting the table” for the fall campaign.

THE NATURE OF ECONOMIC VOTING

That the state of the economy has a large impact on elections is hardly a well-kept secret. But how exactly does the economy affect voting? What specific aspects of economic performance have the strongest effect? Political scientists have developed answers to these questions from studies of presidential approval ratings, but even more from the presidential forecasting literature. Two conclusions from this work are particularly worth noting:

First, in evaluating the condition of the economy, voters are much less concerned with the absolute level of economic performance than with the trend: is the economy getting better or getting worse? is unemployment going up or down? In the fall of 1984, the United States had what was, in historical terms, a fairly high rate of unemployment—about 7.5%. Yet Ronald Reagan won a huge reelection victory because the trend was in the right direction: In the two previous years, the unemployment rate had come down by more than 3 percentage points. In contrast, during George H.W. Bush’s single term in office, the unemployment rate never rose above 7.8%; but because the preelection trend was either stagnant or in the wrong direction, Bush was decisively voted out of office.

Second, in assessing the trends in economic performance, the public seems to have a fairly short time horizon. Most models use a measure of economic conditions that looks back, not at a president’s full term (or two terms) in office, but at the last six to 12 months. Abramowitz’s model, for example, includes a variable that measures the real growth of the country’s gross domestic product during the first half of the election year; Lewis-Beck and Tien use the growth of gross national product from the fourth quarter of the year before the election to the second quarter of the election year. Erikson and Wlezien look at the cumulative growth in leading economic indicators during the first 13 quarters of a president’s term, but each quarter’s results are weighted so as to lose 20% of their impact for every subsequent quarter. The index thus gives little weight to anything that happens during the first two years of a president’s term.^{Footnote 4}

DO CAMPAIGNS MATTER?

Although many election forecasters would probably cite this as the most important implication of their research, I list it last—and with a question mark attached—for two reasons. First, well before the advent of presidential election forecasting, most political scientists were already arguing that the general election campaign was less important in determining the final result than it was generally portrayed to be in the popular media. Indeed, this was one of the principal conclusions of what is usually described as the first great empirical study of presidential elections: Lazarsfeld, Berelson, and Gaudet’s The People’s Choice, first published in 1944.

Second, how unimportant campaigns are depends (in part) on how accurately one can predict an election’s results before the campaign takes place—and the record here, to say the least, is mixed.^{Footnote 5} In 2000, as is well-known, all seven of the models presented at APSA’s quadrennial forecasting panel predicted a comfortable victory for Al Gore. While several of the forecasters could claim that they were “in the ballpark,” four of seven said that Gore would win at least 55% of the two-party popular vote (he actually received 50.3%). In 2012, several forecasts came quite close to the actual result, but three of the ten forecasts listed in the October 2012 issue of PS were off by more than 4 percentage points.

What are sometimes called the “fundamentals” of an election—the popularity of the president, the state of the economy—clearly matter in determining the final outcome, but so does the quality of the major-party campaigns. In 1984, when the economy was growing rapidly and 54% of American adults approved of Reagan’s job performance, there was probably no realistic scenario through which Democratic candidate Walter Mondale could have won the election. But how much better might he have done if he had run a more skillful campaign—if he had, for example, not yielded to the entreaties of feminists and selected Geraldine Ferraro as his running mate, or if he had not openly promised to raise everybody’s taxes? At the current state of the literature, I am not confident that we have a plausible answer to this question.