WATCHING THE RACE

Existing knowledge about variation in the candidates' trial-heat support relative to standard vote choice is limited. Prior to 2008 trial-heat polling was not as abundant, limiting previous scholarly work. Not only were there gaps in the faux election series in earlier cycles, but the data also were not available earlier than the last few months of the race. A review of the literature yields three previous works that examine support for the candidates over the course of the campaign: Holbrook (Reference Holbrook1994) explains the 1984, 1988, and 1992 faux elections; Shaw and Roberts (Reference Shaw and Roberts2000) explore 1992 and 1996; and Stimson (Reference Stimson2004) examines every race from 1956 through 2000.

Holbrook (Reference Holbrook1994) argues that a combination of both campaign events and national conditions explain candidate support over the course of the election. He suggests campaign events will have a more significant effect on candidate support within each individual election, but that across elections the impact of national conditions will be more consistent and substantial. To gauge these effects he examines data from the 1984, 1988, and 1992 presidential elections individually and then as a pooled data set. He draws his trial-heat poll data at the daily level from June through November of each race and transforms candidate support into a measure of the Republican candidate's relative standing. He then models the relative standing of the candidates as a function of campaign events—namely the conventions, debates, other staged campaign events, and momentum—and national conditions, measured by consumer sentiment and presidential popularity. His initial results offer support for his campaign-events hypothesis, finding that the conventions and even some of the debates are significant predictors of candidate support within each election. The pooled model results offer further support for Holbrook's hypotheses, illustrating that economic considerations and presidential popularity are more significant when modeling across multiple elections, as opposed to within any individual election. Holbrook's results also indicate campaign-event effects should be present from election to election, while national conditions have a less predictable, but potentially more significant effect. Based on Holbrook's findings, we argue that the 2008 race should be a function of campaign events.

Shaw and Roberts (Reference Shaw and Roberts2000) argue that candidate support is a function of media coverage, campaign events, and the state of the economy. They use daily data from the Iowa Political Stock Market to examine support relative to expectations of the final election outcome. This differs from the more traditional trial-heat polls that measure candidate support as if “the election were held today.” Despite this difference in measurement, their findings appear consistent with the studies mentioned above. Specifically, they offer evidence that campaign events and economic conditions play an important role in determining the candidates' relative standing. More interestingly, they suggest that both newspaper and television media also play an important role in determining the public's expectations for the candidates. Based on these findings regarding media effects, we supplement our argument by suggesting that media framing of the economy in 2008 could have a significant role in determining the support for the candidates.

Stimson (Reference Stimson2004) engages in a comprehensive discussion of the dynamics behind candidate support. He argues that the conventions exert a significant, permanent influence on the candidates' levels of support. He tests this hypothesis by examining candidate support across every presidential election between 1960 and 2000. The results indicate that in almost every case the conventions exert a significant influence. Stimson suggests that the debates should not be as influential as the conventions because they take place so late in the race when a large segment of the voting public has already formed their candidate preferences, limiting any effect on candidate support. This is only further evidence that campaign events play a significant role in determining candidate support.

While the literature predicting candidate support is relatively limited, another way to think of candidate support in the tracking polls is as a series of separate elections. This suggests that we should consider the literature on election forecasting. Across numerous forecasting models there is some consensus that a candidate's final vote share should primarily be a function of economic conditions (Campbell Reference Campbell2008; Erikson and Wlezien Reference Erikson and Wlezien2008) and presidential approval (Abramowitz Reference Abramowitz2008; Lewis-Beck and Tien Reference Lewis-Beck and Tien2008).

Based upon this literature, we believe candidate support should operate as a function of economic conditions, presidential approval, and a series of campaign events. In addition, given the importance of the Iraq war in recent elections, we argue that it should have had an effect on candidate support in the 2008 election as well. We present our model below:

where S is candidate support, E is economic conditions, A is presidential approval, W is the Iraq war, and CE is a series of campaign events.

GOING RACING

Before leaping headlong into statistical estimation we examine the 2008 race. Starting in January 2008, James Stimson compiled every trial-heat poll reported in the media. This resulted in a collection of almost 900 trial-heat polls across more than 20 polling firms.Footnote ¹ These polls were then combined into a single time series using Stimson's algorithm for public mood. The result is a series spanning 304 days reporting public opinion in response to the question “Suppose the election were held today, which candidate would you support?” While many scholars argue that movement in the polls is often within sampling error, the combination of so many polls across so many firms means that even small movements in the candidates' positions may represent real movements in opinion.

The values for each candidate represent the percent of respondents who indicated that they would vote for the candidate in the trial heat. Approximately 6 or 7% of respondents per day indicated a preference for other candidates, neither candidate, or expressed no opinion.

Looking at Figure 1 we see Senator Obama was the clear favorite throughout January and February. At the start of March, Senator McCain appeared to reach near parity, perhaps in response to securing the presumptive nomination of the Republican Party. As Obama continued to fight it out with Senator Clinton, McCain was able to turn his sights toward the general election. Once the Democratic primaries began to shift in Obama's favor, the two candidates once again entered a period of parity, with both candidates trading the lead until the beginning of June. Had the election been held during this period, it would have been anybody's race.

Figure 1.

The 2008 Presidential Horse Race

On June 3, following the Democratic primary in Montana and a shift of several superdelegates, Obama secured enough delegates to become the party's presumptive nominee. Hillary Clinton conceded shortly thereafter. These events can be seen by both a surge in support for Obama and a serious collapse in support for McCain. The result is a lead of 7 percentage points for Obama, a lead maintained for the next three months.

The conventions were accompanied with expected increases in support for each candidate. The Democratic National Convention clearly boosted Obama's standing with the public, as indicated by his support in the polls reaching 50% for the first time since early February. Similarly, the announcement of Alaska governor Sarah Palin as the vice-presidential nominee in conjunction with the Republican National Convention produced a surge in McCain's support in early September.

The RNC/Palin bump, however, was short lived, as the issues caught up with the horse race. September brought a flurry of economic downturns. The federal government's seizure of Fannie Mae and Freddie Mac marked the beginning of the economic crisis that would become the most important issue of the campaign. These events appear to have come at the expense of McCain, causing his support to deteriorate to around 44% for the remainder of the race. Obama reclaimed the lead and his support continued to grow, topping 51% by the start of October and reaching as high as 54% immediately before the election. Obama's lead would not be relinquished this time. In fact, his support stabilized in the low 50s for the final weeks of the campaign.

EXPLAINING THE 304 ELECTIONS OF 2008

We estimated three models of candidate support using an autoregressive distributed lag model in which support is a function of economic conditions, economic crises, economic news, presidential approval, the cumulative number of soldiers killed in Iraq, campaign events, and lagged support. Independent variables were hypothesized to affect support with a one-day lag and then to have an effect through lagged support that decays over time, thus capturing a dynamic causal effect. The inclusion of a lagged dependent variable means that regression coefficients represent current, short-term effects, and that long-term effects cumulate at a rate determined by the coefficient on the lagged dependent variable. Thus the effect of, for example, presidential approval will be felt one day after it occurs and also feed forward into the future through the lagged dependent variable with an exponential rate of decay.

The three models we estimated include support for Obama, support for McCain, and their relative standing (Obama support minus McCain support). We estimated these models for three reasons. First, when surveyed about their vote, about 7% of respondents said that they had no opinion or were undecided or would vote for someone else. This means we considered support out of all response options. Second, this allowed us to ascertain whether events have a symmetric or asymmetric effect on support for the two candidates or influence only one candidate. Finally, we looked at relative standing because this was, after all, an election pitting the two candidates against each other.

We begin by discussing the details of the operationalization of the variables and then present and discuss the results in Table 1.

Table 1.

Candidate Support

Std. errors in parentheses. +p < 0.100; *p < 0.050; **p < 0.010 (two tailed)

THE ECONOMY AND THE MEDIA

Due to its salience, we began with the economy, explaining the daily faux elections with variables designed to pick up its importance. First, we identified an indicator of objective conditions at the daily level. Standard bearers from the aggregate elections and forecasting literatures (personal income, inflation, unemployment, and subjective indicators) were not available to us daily. Many indicators were available daily, from federal funds rates to treasury bonds to crude oil prices. We used the change in the Dow Jones Industrial Average daily closing price.Footnote ² We selected the Dow because it is widely broadcast, increasing numbers of Americans have retirement accounts tied to the market, and increases and decreases in the Dow have clear implications for the health of the overall economy. Second, we included a measure of media coverage of the economy by looking at the New York Times.Footnote ³ This measure is more general than our objective indicator, capturing the sort of information about the economy that might have been reaching voters. Third, we used a timeline of economic crisis events, beginning on September 7 with the seizure of Fannie Mae and Freddie Mac and ending with the passage of the Emergency Economic Stabilization Act of 2008. The series was scored so that each event added to the weight of the economic crisis in general, compounding the effect that came with each additional piece of bad news, so the series recorded the cumulative number of economic events to date.Footnote ⁴

PRESIDENTIAL APPROVAL

Based upon the forecasting literature, we included daily presidential approval in our models of candidate support.Footnote ⁵ We know that the standing of the incumbent president affects the success of his party's nominee and here we have a case of record low approval ratings for Bush. We suggest that as Bush's approval sank, so did McCain's support while Obama's support grew.

THE IRAQ WAR

In addition to the economy, the war in Iraq played prominently in discussions during the campaign. Whether this issue played to either candidate's advantage is not entirely clear as those who wanted a quick end to the war were likely to be swayed in Obama's direction. However those who did not support an immediate end to the war were reminded of Obama's foreign policy inexperience, meaning we may have seen a rise in McCain's support as the war dragged on and American lives were lost.Footnote ⁶ Ideally we would have liked to include a daily measure of public support for the war, however, as such data were not available we included a count of the cumulative number of Americans killed in Iraq as an indicator of dissatisfaction with the war.Footnote ⁷

CAMPAIGN EVENTS

The daily campaign events themselves had the potential to alter the candidates' support. Of these a handful stand out: the conventions, debates, and securing party nominations. We had reason to expect each party would benefit from its convention as the other party took a campaign holiday, the media is unusually generous to the party, and there is a general celebration of the party designed to bring home the faithful by reminding them of the reasons for their loyalty. We followed Stimson (Reference Stimson2004) and Holbrook's (Reference Holbrook1994) example and modeled the horse race as a function of the party conventions, estimating, as they did, a permanent effect for each party's convention by creating a dummy variable that is 0 before the convention and 1 from the convention until the election. We followed similar suit with the debates but there was much less in the way of expectations as to their effect. They came late in the campaign after much of the action in the series and followed each other in quick succession, making it difficult to assess their separate effects. Lastly, securing the nomination was likely to help the candidate in the tracking polls.Footnote ⁸

MODELING THE RACE

What determined support for each candidate and for their relative standing? We see strong inertia in support, as the previous day's support is strongly related to current support. Other variables also break the significance boundary. We started with the economic crisis events. As expected, each successive event added to Obama's support and subtracted from McCain. For Obama each event added 0.16 points in the short run the day after it occurred and a long run effect of 0.75 points.Footnote ⁹ Across the seven events, the total long-run effect equates to a 5.27 point gain for Obama. However, the effect of the crisis is not symmetric, as McCain lost even more support than Obama gained, dropping 0.21 points in the short run and 0.97 points in the long run for a single event. Across all seven events the long-run total effect is a loss of almost seven points for McCain. Relative support greatly advantaged Obama with a total 12.64 point shift in his favor. This is a large effect indeed, one not approached by any other variable in the model. Yet, in spite of the importance of the economy as measured by crisis events, neither the Dow Jones Industrial Average daily closing price nor the net number of positive news stories achieve significance in any of the models.

In addition to the economic crisis and in spite of the strong collinearity in the post-convention data, presidential approval managed to exert a significant effect. The model indicates that when Bush's approval ratings rose, we saw the expected effect: Obama's support fell and McCain's support grew. Over the 13 point downswing in presidential approval that characterized the 2008 campaign, the long-run effect on Obama's support results was an almost 3 point increase. Conversely, McCain's support fell by 2.30 points in the long run for a total effect on their relative standing of 5.65 points. The estimated effect is not as large in magnitude as for the crisis events, but is important nonetheless.

The campaign variables have a mixed showing. When Obama clinched the Democratic nomination, his support was not significantly altered. The effect of Obama's nomination operated through McCain's support, which dropped by 3.43 points in the long run. Although Obama's gain is not significant, the relative support model estimated a 4.59 point swing for Obama. McCain securing the Republican nomination, while visible in Figure 1, had no significant effect across the models. Turning our attention to the national conventions, we find that the Democratic convention significantly benefited Obama to the tune of half a point in the short run and 2.49 points in the long run. However, the DNC did not seem to affect McCain's support, and is not significant in our model of Obama's relative standing. In contrast, the Republican convention appears to have played a much larger role in determining candidate support. The results indicate that it produced an immediate 0.50 loss for Obama, further compounding with an immediate 1.00 point gain for McCain. In fact, we suggest the Republican convention not only wiped out the gains from the Democratic convention for Obama, but McCain benefited even more. While the initial effect of the RNC improved McCain's relative standing by roughly 1.50 points, the long-run effects are much more pronounced. McCain eventually gained 4.68 points in his support and his relative standing improved by a sizable 7.45 points. We can only speculate whether this jump in McCain's support was in response to his campaign efforts or a bump from naming Sarah Palin as his running mate. Regardless, McCain's significant gain from the RNC was counteracted as both presidential approval and the economic crisis events appeared to turn the tide against him, causing his support to fall in the aftermath of the conventions.

Lastly, looking at the debates, we can see that only one of the debates showed any sign of significantly affecting the candidates' standings. We find that the second debate reached marginal significance in our model of McCain's support, indicating that the debate helped his standing. This finding is consistent with Stimson's (Reference Stimson2004) observation that the debates come so late in the campaign that they tend to have a negligible effect on the race. However, the boost for McCain supports the belief that McCain was best suited for the second debate as it followed a town-hall format. The short-run effect implies that McCain gained approximately one-third of a point, with the long-run effect totaling out to roughly 1.50 points. Yet, we are hesitant to overstate the effect of the debate, as our model of Obama's relative position does not show any significant change.

SIMULATING THE “WHAT IFS”

Although we have reported the short- and long-run effects of the economy and presidential approval on candidate support, we provide a clearer illustration of how the 2008 presidential horse race might have played out under different scenarios. In order to do this we ran a pair of simulations using the model results to predict candidate support under different circumstances. Specifically, we examined two alternatives: if the economic crisis had not occurred, and if Bush's approval was constant at 34%. The results of the simulations are presented below.

Looking at Figure 2 we see the simulation results if the economic crisis had not occurred. In this scenario we allowed the Dow Jones Industrial Average and media coverage to follow their observed values but recoded the economic crisis events all to zero. The simulation indicates that in September McCain's support surges, presumably in response to the RNC/Palin bump, which now persists and actually grows. The period of brief parity between the two candidates is followed by a slight Obama decline (from 48% to 46%) and a more notable McCain upswing (from 48% to 51%). The timing of this change could not come at a worse time for Obama, as the McCain lead persists through Election Day, resulting in a 2 point McCain victory.

Figure 2.

Predicted Candidate Support—No Economic Crisis

Figure 3 illustrates how the race would have differed if Bush's approval ratings did not reach record lows during the campaign. In this scenario we allowed the economic crisis and other events to occur normally but presidential approval was fixed at 35%. Admittedly one could argue that, given the high collinearity between the economy and presidential approval, holding Bush's approval rating constant at its maximum value for the series was an unlikely scenario. However, our results suggest that even when modeled together, the economy and approval have significant, independently distinguishable effects, meaning that we can estimate an independent effect of presidential approval.

Figure 3.

Predicted Candidate Support—Presidential Approval Held Constant at 35%

Looking at the figure, the race clearly follows a significantly different path. From the onset of the race we can see changes in McCain's support, as the race no longer exhibits any periods of parity between the candidates. Rather, Obama maintains the lead in the early goings and again following Clinton's concession, but McCain now holds a 2 point lead during the height of the Democratic primaries. We also see that there is movement in candidate support in response to the different campaign events, with McCain's support reaching its peak in response to the RNC/Palin bump at roughly 48% in mid-September. Despite the boost to McCain's support from increased presidential approval, the economic crisis still manages to counter McCain's RNC/Palin bump momentum. The end result of this scenario is a 4 point reduction in the margin between McCain and Obama, with a final 6 point Obama victory. Presidential approval appears to play a significant role, but cannot completely alter the outcome, as Bush's approval ratings would have had to reach 50% to secure a McCain victory.