Election Forecasters Preparing For Historic Election

BUFFALO, N.Y. -- Anticipating what is likely to be one of the most interesting elections in modern history, University at Buffalo professor of political science James E. Campbell and Michael S. Lewis-Beck, professor of political science at the University of Iowa, have assembled the insights of prominent election forecasters in a special issue of the International Journal of Forecasting published this month.

Each of the articles demonstrates the challenges of election forecasting, according to Campbell, chair of UB's Department of Political Science, who since 1992 has produced a trial-heat-and-economy forecast of the U.S. presidential election. His forecast uses the second-quarter growth rate in the gross domestic product and results of the trial-heat (preference) poll released by Gallup near Labor Day to predict what percentage of the popular vote will be received by the major party candidates.

The articles range from descriptions of diverse election forecasting models, such as those that use political futures markets and historical analysis, to articles that evaluate the success of election forecasting in past elections.

Two of the articles address a topic particularly pertinent to the 2008 presidential election: whether open seat and incumbent elections should be treated differently by election forecasters.

"One of the biggest misunderstandings about election forecasting is the idea that accurate forecasts must assume that the campaign does not matter," Campbell explains. "This is not true.

"First, one of the reasons that forecasts can be accurate is that they are based on measures of the conditions that influence campaigns. So campaign effects are, to a significant degree, predictable.

"Second, forecasters know that their forecasts are not perfect. Forecasts are based on imperfect measures and may not capture all of the factors affecting a campaign. Some portion of campaign effects is always unpredictable."

Though some campaign effects are unpredictable "the extent of these effects is usually limited," Campbell points out.

In the historic contest between presumptive presidential nominees Barack Obama and John McCain one thing is certain: "Forecasting this election will be more difficult than usual," Campbell says.

"First, there isn't an incumbent. Approval ratings and the economy are likely to provide weaker clues to an election's outcome when the incumbent is not running. Second, Democrats had a very divided nomination contest and it is unclear how lasting the divisions will be.

"Third, many Republicans are not very enthusiastic about McCain and it is unclear how strong Republican turnout will be for him."

Of the six different forecast models described in the journal articles, only two have a forecast at this point. The other four will have forecasts between late July and Labor Day. The journal articles can be downloaded at

http://www.sciencedirect.com/science/journal/01692070 . Below are brief descriptions:

• In "U.S. Presidential Election Forecasting: An Introduction" journal co-editors Campbell and Lewis-Beck provide a brief history of the development of the election forecasting field and an overview of the articles in this special issue.

• In "Forecasting the Presidential Primary Vote: Viability, Ideology and Momentum," Wayne P. Steger of DePaul University takes on the difficult task of improving on forecasting models of presidential nominations. He focuses on the forecast of the primary vote in contests where the incumbent president is not a candidate, comparing models using information from before the Iowa Caucus and New Hampshire primary to those taking these momentum-inducing events into account.

• In "It's About Time: Forecasting the 2008 Presidential Election with the Time-for-Change Model," Alan I. Abramowitz of Emory University updates his referenda theory-based "time for a change" election forecasting model first published in 1988. Specifically, his model forecasts the two-party division of the national popular vote for the in-party candidate based on presidential approval in June, economic growth in the first half of the election year, and whether the president's party is seeking more than a second consecutive term in office.

• In "The Economy and the Presidential Vote: What the Leading Indicators Reveal Well in Advance," Robert S. Erikson of Columbia University and Christopher Wlezien of Temple University ask what is the preferred economic measure in election forecasting and what is the optimal time before the election to issue a forecast.

• In "Forecasting Presidential Elections: When to Change the Model?" Michael S. Lewis-Beck of the University of Iowa and Charles Tien of Hunter College, CUNY ask whether the addition of variables can genuinely reduce forecasting error, as opposed to merely boosting statistical fit by chance. They explore the evolution of their core model – presidential vote as a function GNP growth and presidential popularity. They compare it to a more complex, "jobs" model they have developed over the years.

• In "Forecasting Non-Incumbent Presidential Elections: Lessons Learned from the 2000 Election," Andrew H. Sidman, Maxwell Mak, and Matthew J. Lebo of Stony Brook University use a Bayesian Model Averaging approach to the question of whether economic influences have a muted impact on elections without an incumbent as a candidate. The Sidman team concludes that a discount of economic influences actually weakens general forecasting performance.

• In "Evaluating U.S. Presidential Election Forecasts and Forecasting Equations," UB's Campbell responds to critics of election forecasting by identifying the theoretical foundations of forecasting models and offering a reasonable set of benchmarks for assessing forecast accuracy. Campbell's analyses of his trial-heat and economy forecasting model and of Abramowitz's "time for a change" model indicates that it is still at least an open question as to whether models should be revised to reflect more muted referendum effects in open seat or non-incumbent elections.

• In "Campaign Trial Heats as Election Forecasts: Measurement Error and Bias in 2004 Presidential Campaign Polls," Mark Pickup of Oxford University and Richard Johnston of the University of Pennsylvania provide an assessment of polls as forecasts. Comparing various sophisticated methods for assessing overall systematic bias in polling on the 2004 U.S. presidential election, Johnston and Pickup show that three polling houses had large and significant biases in their preference polls.

• In "Prediction Market Accuracy in the Long Run," Joyce E. Berg, Forrest D. Nelson, and Thomas A. Reitz from the University of Iowa's Tippie College of Business, compare the presidential election forecasts produced from the Iowa Electronic Market (IEM) to forecasts from an exhaustive body of opinion polls. Their finding is that the IEM is usually more accurate than the polls.

• In "The Keys to the White House: An Index Forecast for 2008," Allan J. Lichtman of American University provides an historian's checklist of 13 conditions that together forecast the presidential contest. These "keys" are a set of "yes or no" questions about how the president's party has been doing and the circumstances surrounding the election. If fewer than six keys are turned against the in-party, it is predicted to win the election. If six or more keys are turned, the in-party is predicted to lose. Lichtman notes that this rule correctly predicted the winner in every race since 1984.

• In "The State of Presidential Election Forecasting: The 2004 Experience," Randall J. Jones, Jr. reviews the accuracy of all of the major approaches used in forecasting the 2004 presidential election. In addition to examining campaign polls, trading markets, and regression models, he examines the records of Delphi expert surveys, bellwether states, and probability models.