The Polls Defy Retrospective Based Expectations

The foundation of these forecasts is the trial-heat polls, in particular the polls taken after public opinion about the election begins to gel, generally those from August on. This year, in the 24 days from August 1 to the beginning of the Democratic Party Convention, the polls indicated that the presidential race was about even. McCain's mean percentage of two-party support in Gallup's poll of registered voters was 48.7%. This was certainly a far closer race than the retrospective models would have anticipated based on President George W. Bush's 31% July approval rating, a July rating tied for the lowest in election years since 1948. Moreover, McCain's poll standing in Gallup's registered voter poll understated his support among those who would actually vote, since registered Republicans routinely turn out at higher rates than do registered Democrats. Whether McCain was slightly ahead of Obama or slightly behind, however, is not what is important. What is important is that McCain was in the hunt, even though by retrospective voting standards he should have been running far behind Obama.

While he entered the convention period in a virtual tie with Obama, McCain emerged from the conventions with the poll lead.Footnote ¹ In part, his lead can be attributed to the short-lived effects of his convention bump, but in the past a portion of these bumps have survived and McCain held his lead for over a week (longer if you consider the higher turnout of registered Republicans). This was important. Of the 13 previous early September frontrunners since 1948 with more than 51% in the two-party poll split, only one lost the popular vote—Tom Dewey in President Harry Truman's legendary come-from-behind 1948 victory.

Despite the weight of retrospective evaluations against President Bush, Senator McCain seemed well positioned entering the last two months of the campaign and this was reflected in the Trial-heat Forecast in his favor. Two months later, voters gave McCain 46.5% of their votes to 53.5% for Obama.Footnote ² The Trial-heat Forecast missed the actual vote by a whopping 6.2 percentage points, more than twice the model's average error. What happened?

The Wall Street Meltdown “Game Changer”

The Trial-heat Model, like all forecasting models, assumes that the election being predicted is essentially normal, affected by the same set of considerations and in similar ways as previous elections. Campaigns matter and many of their effects are common and themselves predictable. Competition, incumbency, and the pre-existing economy structure many predictable campaign effects, while some additional effects are unpredicted, perhaps unpredictable, but fairly minor (a point or so). This later category of effects encompass all the unforeseen gaffes and maneuvering in the campaign. The assumption of a normal campaign is implicit in using past elections to estimate the forecasting equations. The September events of this election season violated the normal election assumption—big time.

The Wall Street meltdown was an unforeseen game changer. It was not even mentioned in either party's nominating conventions, yet it dominated the campaigns within days of the conventions. Commonly acknowledged as the greatest economic crisis that the nation had faced since the Great Depression, the meltdown began on September 7 with the federal government bailout of Fannie Mae and Freddie Mac. A week later, financial giants like Lehman Brothers, Merrill Lynch, AIG, and Washington Mutual were on the brink of or in bankruptcy. A few days later the president asked Congress to pass an emergency economic stabilization bill, what became known as the $700 billion bailout bill. After initially defeating the Senate-passed bill, the House then passed and the president signed a second version. Meanwhile, the stock markets went into a nose dive. In the space of one month, they lost about a quarter of their value (as measured by either the Dow Jones or the Standard and Poor's indexes). In the steady drumbeat of this economic calamity, McCain's poll standing dropped about seven points. President Bush's approval rating, already in the cellar, dropped by a like amount. Unlike any election in modern history, the 2008 presidential campaign had been derailed by an entirely unanticipated and unpredicted external event. As its absence from the wide-ranging agendas of the conventions attests, no one saw it coming.

An Unpredicted and Unpredictable Election

Some of my colleagues may claim that their retrospective-based models were quite accurate in forecasting this election. Under the circumstances, I think that whether a forecast was close to or distant from the vote this year is largely a matter of good or bad luck. Nothing in any of the models would have either directly or indirectly anticipated the effects or the timing of the Wall Street meltdown. Nothing. Retrospective models based largely on late summer approval ratings do not explain why the election was so closely fought in the month before the conventions, nor do they account for why McCain led Obama in the post-convention polls (through September 15 in Gallup and September 17 in the RealClearPolitics' average of polls, and this does not take into consideration the undercount of Republican support because of turnout differences between registered partisans), and they certainly do not explain why the biggest financial crisis to hit the country since the Great Depression amazingly left their forecasts intact.

Presidential approval ratings in midsummer have historically been related to the early September poll standings (explaining why the retrospective models have worked in the past). Since 1948, all nine in-party candidates whose presidents held 46% or higher approval ratings in July were also early September poll frontrunners. Conversely, of the seven in-party candidates whose presidents were sub-46% in July approval ratings, each trailed in the early September polls—with one exception, John McCain this year. Despite President Bush being tied for the lowest July approval rating in an election year since 1948, McCain led Barack Obama in early September (52.7% in Gallup and 51.5% in RealClearPolitics). If the retrospective models were working, Obama should have held large leads both before as well as after the conventions, but he did not.

How was it possible for McCain to have overcome the considerable burden of President Bush's low approval rating and to defy the expectations of the purely retrospective forecasting models and hold the lead in early September? First and most obviously, open-seat elections are less retrospectively oriented than incumbent elections. As McCain continually insisted, he was not George W. Bush and voters understood the difference. Second, most of the loss of President Bush's positive job approval ratings between his reelection in 2004 and July of 2008 was among Republicans. In his second term, Bush's approval dropped only four points among Democrats (from 11% to 7%). Among Republicans, it dropped 26 points (from 93% to 67%). Along with some independents, McCain could draw many of these Republicans disgruntled with President Bush back into the fold. What McCain could not do was to hold these votes with the additional weight imposed by September's Wall Street collapse.