THE GROWING EFFECT OF THE ECONOMY OVER THE CAMPAIGN

We start with the early stages of the campaign, 200 days before the election (April of election year). At that time, economic perceptions barely register as a predictor of the trial-heat polls.^{Footnote 2} See the left panel of figure 1, which plots the incumbent party candidate’s poll share and economic perceptions in April of the election year. The correlation between economic perceptions and the polls at that time is negligible (Pearson’s r = 0.05). By Election Day, however, voter preferences are clearly responsive to perceptions of economic progress. This trend can be seen in the right panel of figure 1, which displays the incumbent party candidate’s vote share and economic perceptions at the end of the cycle (Pearson’s r=0.76).^{Footnote 3}

Figure 1 Predicting the Polls in April and the Vote in November from Economic Perceptions at Each Point in Time

Economic perceptions increasingly predict the vote over the campaign timeline. (Color online)

Next, we divide the daily trial-heat polls into the portion predicted from current economic perceptions and the portion that is not. For each date we regress the poll results on economic perceptions and then compute both predicted values and the residuals. We want to observe how the spread of the daily predictions changes over the timeline, which tells us how economic forces and other issues come to matter to voters. For this exercise, figure 2 illustrates the variance of the daily readings of the predicted values (economy) and the residuals (other) by day, for the final 200 days of the election cycle.^{Footnote 4}

Figure 2 Variances of from the Two Components of Daily Polls: The Parts Predicted by Economic Perceptions and Other Sources

The variance of the economic component of polls grows while the variance of the larger noneconomic component shrinks over the campaign. (Color online.)

In figure 2 we see that in terms of cross-election variance of aggregate preferences, noneconomic factors (the residuals) dominate at the outset of the campaign, but then decline over time. Meanwhile, the economic variance—the portion of trial-heat polls explained by economic perceptions—is small at the beginning of the campaign but grows over time. It almost catches up with the noneconomic variance by the end of the timeline. In terms of the external fundamentals, at least as measured by economic perceptions, the 15 election results become more dispersed.

How do we interpret this pattern? Early in the election year, voters are not paying much attention to the economy when they answer pollsters’ questions about how they will vote. But they must be taking into account other things, such as their views of the current administration and early perceptions of the eventual presidential candidates. As the campaign proceeds, these considerations become less important in voters’ minds than their internal fundamentals, especially partisanship (see Erikson and Wlezien Reference Erikson and Wlezien2012a). Meanwhile, the economic cue increases in salience and continues to matter more in the aggregate, despite the growing force of voters’ internal fundamentals. The campaign effectively brings the economy to the forefront bearing in mind that the economy evolves over the campaign—it is not a constant during the election year. The same is true of noneconomic fundamentals, including issue proximity (see Erikson and Wlezien Reference Erikson and Wlezien2012a).

PREDICTING THE VOTE FROM THE ECONOMY OVER THE CAMPAIGN TIMELINE

Let us say, for some future election, we take daily poll readings and we account for the portion determined due by economic perceptions and the remainder due to other factors. Which is more important in terms of predicting the Election Day vote? To see, we regress the actual presidential party vote on the daily economy-based poll prediction and the residual portion of the polls. The results are shown in figure 3, where we plot the coefficients for each. We are interested in the size of the coefficients. A coefficient of 1.0 means that the effects on the polls project completely(on average) to Election Day, and a coefficient of less than 1.0 means the effects do not fully translate into the final vote, that is, they matter more to the polls than the actual vote. A coefficient greater than 1.0 means just the opposite—that the effects matter more to the actual vote than they do to the polls at that time.

Figure 3 Coefficients Predicting the Vote from the Economic and Non-Economic Components of Daily Polls

The larger coefficients for the economic component indicates that poll results based on economic perceptions have the most lasting power until Election Day. (Color online.)

Figure 3 shows that the coefficient for the economic portion of the polls starts out much greater than 1.0 and then declines. This decline means that the economy-induced poll numbers from early in the timeline are more important in the final tally than the meager economic effects in the early polls would indicate. Meanwhile, the noneconomy-induced portion has a coefficient of less than one. Why the asymmetry? Look at it from the perspective of early dates in the campaign timeline. At that time, the influence of the economy on the vote will only grow beyond what current polls show, while the influence of other, noneconomic causes remains relatively stable. On Election Eve, when polls more fully reflect economic perceptions, the coefficient for the economy-driven portion has fallen to just about 1.0 (0.92), slightly larger than the 0.65 coefficient for the noneconomy portion.^{Footnote 5}

Next, we look at the net predictive power of the two components of the polls—the economy- induced part and the noneconomy-induced part. Figure 4 traces the portion of the Election Day variance in the vote that can be explained by each component over the campaign timeline. Because the two components are uncorrelated by construction, we can add the explained variances from each of the two independent variables.^{Footnote 6}

Figure 4 Proportion of the Variance in the Election Day Vote Explained by Daily Readings of the Two Components of Trial-Heat Polls—Economic and Noneconomic

The economic component offers greater electoral predictability throughout. (Color online.)

We see from figure 4 that, over time, both components of the trial-heat polls explain an increasing portion of the variance in the vote. The later in the campaign, the more you can predict from current preferences. This, of course, is as it should be, although we see that the trend for the residual portion is more ragged. Note that the portion of the polls that reflects the economy explains the greater portion of the variance in the Election Day vote. This is true from the beginning of the campaign until the end. In other words, if one could determine how much the trial-heat polls were influenced by economic perceptions, that portion would be far more helpful for an Election Day prediction than the portion predicted by other factors.

ACCOUNTING FOR OBAMA’S REELECTION

So, why did Obama win in 2012? After all, the economy was in bad shape. All that Romney needed to do, it seemed, was wait for the campaign to do its work and remind voters that the economy suffered, and therefore they should throw the rascals out. The problem for this expectation was that in terms of perceptions of the current economy, people were seeing positive signs (Erikson and Wlezien Reference Erikson and Wlezien2012b). As seen in figure 1, economic perceptions on Election Day actually gave the president a slight advantage, which he realized. As in past elections, in 2012 it was recent economic growth that mattered most to voters.

The economy actually was an advantage for Obama for much of the election year. Figure 5 shows the estimates of the economic and noneconomic components of voter preferences over the 2012 campaign, as derived from the polls. But for a short period in the middle of the summer—and well before the party conventions—economic perceptions favored Obama. The reverse was true for noneconomic factors.

Figure 5 The Economic and Non-Economic Components of Trial Heat Polls in 2012

For most of the campaign, Obama was actually favored by economic perceptions but not by other sources. (Color online.)

SUMMARY AND CONCLUSION

Let us summarize. Early in presidential campaigns, voters are not incorporating economic evaluations into their electoral preferences; the economy has little impact on presidential polls of the moment. But early-campaign economic perceptions are predictive of the national verdict on Election Day. For this reason, the slight influence of economic perceptions on early trial-heat polls is magnified in its overall forecasting ability. At the same time, we see that the residual portion of voter preferences that is not due to the economy has its impact, too. The influence of these other, noneconomic considerations on the polls decreases over time just as its impact on the vote increases. As the polls absorb the economic and noneconomic fundamentals, the polls offer increasing forecasting power. To predict US presidential elections in advance, however, both the economy and the polls should be used. This is especially true well before Election Day, when the polls only marginally capture economic conditions, but it also is true on Election Eve. Even the final preelection polls do not fully reflect the state of the economy or other noneconomic fundamentals.