UNITED STATES: Forecasting models show a Bush win

Analysis

Seven of the eight forecasting models of the presidential election constructed by political scientists (and one economist) anticipate that President George Bush will win a plurality of the popular vote on November 2. The eighth model essentially predicts a repeat of the 2000 'dead-heat' election between Bush and then-Vice President Al Gore. The median forecast is that Bush will win 53.8% of the popular vote compared to 46.2% for Democratic presidential nominee John Kerry.

Model components. The models examine patterns in previous elections in order to determine the conditions that facilitate or impede the election of the presidential candidate of the party that controls the White House (the so-called 'in-party'). The focus on predicting the in-party candidate's share of the national two-party popular vote stems from the fact that, historically, the popular vote has been closely associated with the electoral college votes needed to win the election.

That said, the 'winner-take-all' provisions for awarding electoral college votes (all but two states award all of their electoral college votes to the plurality vote winner in the state) allows for the possibility that the popular vote winner may not win an electoral college majority in very close elections. This was the case in 2000 (see UNITED STATES: Tight election tests constitution - November 6, 2000). Thus, the models correctly identified that Gore would win a popular vote plurality in 2000, even though Bush ultimately won a narrow electoral college majority (the first time since 1888 that the victor did not win the popular vote).

Model forecasts.The forecasts are presented below in order of their accuracy in the 2000 election, from most to least accurate. The span of elections used to estimate each model is in parentheses. The Campbell and Abramowitz models have the strongest overall records, with the average error of these two models being about 2 percentage points.

Three elements are common to many of the models -- a pre-campaign indicator of public opinion; a measure of the election-year economy; and an indicator of presidential incumbency (often implicit in structuring the analysis in terms of the in-party):

• The Fair Model (1916-2000) predicts that Bush will secure 57.5% of the popular vote. Fair's model employs seven indicators of the economy, partisanship, and incumbency. Although often portrayed as an economic-driven model, the forecast depends more heavily on incumbency (both personal and how long a party has held the presidency). The most serious potential hazard is that it includes no public opinion indicator. It also includes a 'war' variable, but does not count 2004 as a 'war election'.
• The Campbell Model (1948-2000) predicts Bush will win 53.8% of the popular vote. This is based on the Gallup poll in the first week of September and the GDP growth rate in the second-quarter. Bush's forecast advantage in November thus stems largely from his lead among 'likely' voters in the post-Republican National Convention polls (see UNITED STATES: Can the 'Bush bounce' last to November? - September 17, 2004) and on the historical advantage for the in-party.
• The Abramowitz Model (1948-2000) forecasts that the president will secure 53.7% of the vote. This is based on Bush's job approval rating in the last Gallup poll conducted in June, the GDP growth rate in the first half of this year (4%), and a variable indicating whether the in-party is seeking more than a second consecutive presidential term -- an in-party candidate seeking a second term is at a relative advantage.
• The Norpoth Model (1912-2000) predicts that Bush will receive 54.7% of the popular vote. This is based on the candidates' performances in the presidential primaries (an index of early party cohesion), the presidential vote percentages in the 1996 and 2000 elections, and a normal party vote baseline to take into account that this model incorporates data from elections starting in 1912. This model incorporates no information about election year public opinion or economic growth.
• The Wlezien and Erikson Models (1952-2000) forecast that Bush will win 51.7 to 52.9% of the popular vote. They use a cumulative index of the leading economic indicators and either the average presidential approval or the presidential preference poll during a quarter of the year (in this case the second quarter).
• The Lewis-Beck and Tien Model (1952-2000) predicts Bush will secure 49.9% of the popular vote. This model, which has been considerably revised since 2000, is based on July presidential approval; GNP growth in the first half of this year (with different effects depending on whether an incumbent is running); an in-party advantage measure; and a new variable measuring job growth during the presidential term.
• The Holbrook Model (1956-2000) forecasts that Bush will win 54.5% of the popular vote. This is based on the average of the president's approval rating over the summer, perceptions of personal finances weighted by an index of economic news, and the party-term variable used in the Abramowitz model.
• The Lockerbie Model (1956-2000) predicts Bush will win 57.6% of the popular vote. This is based on an index of prospective consumer sentiment drawn from the University of Michigan's Survey of Consumer Attitudes and Behavior and the party-term variable used in the Abramowitz model. This model is substantially revised from that used in 2000. Like the Fair and Norpoth models, it includes no indicator of the public's political opinions going into the campaign.

Potential error. There are three main ways in which the models can be in error:

• They may misestimate the importance of the factors potentially affecting the election. This may account for some errors in individual forecasts; but is unlikely to be much of a factor overall this year, since nearly all of the forecasts favour Bush.
• The fundamentals that serve as the basis for the forecast may be 'misplayed' by the candidates. That is, it is presumed that campaigns will make use of these fundamentals to convert them into votes on election day. However, some candidates fail to do this effectively. For instance, Gore was widely perceived to have run a poor campaign in 2000 that did not claim enough credit for the strong economy or utilise then-President Bill Clinton effectively.
• Unanticipated events can reshape the contours of the campaign. For instance, further developments in Iraq and in the anti-terrorism campaign may cause voters to 'stray' from where the fundamentals alone would have steered them. Additionally, any significant change in turnout from previous elections can generate error. Both major parties are conducting aggressive voter registration campaigns and newly registered electors are usually less predictable in their voting behaviour.