Presidential Election Forecasting Research Articles

A Fractured Electorate? French Presidential Election Forecasting for 2022

This paper forecasts the first round of the French presidential election five months in advance. It uses a key industry-standard structural variable in its forecasting model, approval of the French president. Its parsimony allows for an early forecast, but its potential for error is higher. In the end, the forecast was somewhat off the mark, though it accurately showed the strength of a somewhat attenuated left. However, the structure of French political parties has changed dramatically over the past several years, leaving some doubt as to the reliability of this forecasting technique.

The Campbell Collection of Presidential Election Forecasts, 1984–2016: A Review

An abstract is not available for this content so a preview has been provided. Please use the Get access link above for information on how to access this content.

An Inquiry of Bellwether Counties in US Presidential Elections

Bellwethers have been studied for decades in the realm of election forecasting. Presidential election forecasting has historically focused on the study of bellwethers at the level of the state, likely due to the fact that presidential candidates win electoral college votes based upon the winner of the popular vote in each state. However, states can be difficult to study due to their large sizes and diverse political environments, leading these authors to examine whether bellwethers can also be studied at the finer resolution of individual counties. To do so, the popular vote winner of each county for every presidential election from 1980 through 2016 was tabulated and mapped, yielding 19 true bellwether counties for the period. These counties were geographically clustered in the Midwest and in the Northeast. Demographically, the bellwether counties tended to be whiter, older, less educated, have lower median incomes, have a lower percentage of workers in the labor force, and have higher rates of vacant housing than the country as a whole. Finally, numerical bellwether scores for all 3142 counties in the US were computed based upon the number of elections in which each county voted for the winner of the election. Counties with the highest bellwether scores were also disproportionately found in the Midwest and in the Northeast.

Bayesian Trivariate Analysis of an Opinion Poll: With Application to the Kenyan Pollss

There has been a growing interest by political pundits and scholars alike to predict the winner of the presidential elections. Although forecasting has now quite a history, we argue that the closeness of recent Kenyan presidential opinion polls and the wide accessibility of data should change how presidential election forecasting is conducted. We present a Bayesian forecasting model that concentrates on the national wide pre-election polls prior to 2013 general elections and considers finer details such as third-party candidates and self-proclaimed undecided voters. We incorporate our estimators into WinBUGS to determine the probability that a candidate will win an election. The model predicted the outright winner for the 2013 Kenyan election.

Candidates and campaigns: How they alter election forecasts

We know that candidates and campaigns matter in democratic elections, but that knowledge may not be readily observed in most structural models of national election forecasting. For one, these models virtually never include direct, explicit candidate-related campaign variables as predictors. At most, these candidate/campaign variables are picked up indirectly, usually in polling measures, such as vote intention. For another, the models often manage accurate, ex ante forecasts of US presidential election results, even without the obvious pr​esence of such variables. In this effort, we aim to overcome this paradox by including more direct candidate and campaign measures in a long-standing structural equation model of presidential election forecasting, namely the Political Economy model. We find that inclusions of these candidate and campaign variables do improve the theoretical specification and the statistical performance of the model, and do yield generally more accurate forecasts. However, at least for the test case of the 2016 contest, that increased precision failed to substantively alter the Clinton popular vote forecast.

Limitations of Ensemble Bayesian Model Averaging for forecasting social science problems

We compare the accuracies of simple unweighted averages and Ensemble Bayesian Model Averaging (EBMA) for combining forecasts in the social sciences. A review of prior studies from the domain of economic forecasting finds that the simple average was more accurate than EBMA in four studies out of five. On average, the error of EBMA was 5% higher than that of the simple average. A reanalysis and extension of a published study provides further evidence for US presidential election forecasting. The error of EBMA was 33% higher than the corresponding error of the simple average. Simple averages are easy both to describe and to understand, and thus are easy to use. In addition, simple averages provide accurate forecasts in many settings. Researchers who are developing new approaches to combining forecasts need to compare the accuracy of their method to this widely established benchmark. Forecasting practitioners should favor simple averages over more complex methods unless there is strong evidence in support of differential weights.

Under-performing, over-performing, or just performing? The limitations of fundamentals-based presidential election forecasting

U.S. presidential election forecasts are of widespread interest to political commentators, campaign strategists, research scientists, and the public. We argue that most fundamentals-based political science forecasts overstate what historical political and economic factors can tell us about the probable outcome of a forthcoming presidential election. Existing approaches generally overlook the uncertainty in coefficient estimates, decisions about model specifications, and the translation from popular vote shares to Electoral College outcomes. We introduce a Bayesian forecasting model for state-level presidential elections that accounts for each of these sources of error, and allows for the inclusion of structural predictors at both the national and state levels. Applying the model to presidential election data from 1952 to 2012, we demonstrate that, for covariates with typical levels of predictive power, the 95% prediction intervals for presidential vote shares should span approximately ±10% at the state level and ±7% at the national level.

Proxy Models and Nowcasting: U.S. Presidential Elections in the Future

U.S. presidential election forecasting has lately received considerable attention. A leading approach, statistical modeling, has undergone considerable change. We have contributed to that change in two ways, by stressing prediction over explanation and dynamics over statics. For prediction, we offer a proxy model of U.S. presidential election forecasting, based on an empirical (political economy) indicator measured six months in advance. For dynamics, we offer nowcasting, which permits the model to issue updated, current forecasts over time. To illustrate these offerings, we test them against the 2012 contest. Then, we issue our first, ex ante forecast of the 2016 presidential election.

US Presidential Election Forecasting

An abstract is not available for this content so a preview has been provided. Please use the Get access link above for information on how to access this content.

What, If Anything, Have We Learned from Presidential Election Forecasting?

What, If Anything, Have We Learned from Presidential Election Forecasting?

Issues in Presidential Election Forecasting: Election Margins, Incumbency, and Model Credibility

Three issues of presidential election forecasting research are raised. The first concerns the consequences for forecasting models of presidential elections becoming more narrowly decided. A realigned and polarized party system has led to smaller election margins in presidential contests since 1984. This development complicates model estimation and may also have blurred differences between strong and weak forecasting models. The second issue concerns the presidential incumbency advantage. Forecasting research has shed greater light on this important electoral factor. The advantage of presidential incumbency appears to be largely restricted to first party-term incumbents. This exposes what might be seen as a blind spot to electoral accountability. Finally, the issue of forecast model credibility is raised. Several criteria beyond forecast accuracy should be taken into account in assessing whether a forecast model and its prediction are credible.

A Bayesian Prediction Model for the U.S. Presidential Election

It has become a popular pastime for political pundits and scholars alike to predict the winner of the U.S. presidential election. Although forecasting has now quite a history, we argue that the closeness of recent presidential elections and the wide accessibility of data should change how presidential election forecasting is conducted. We present a Bayesian forecasting model that concentrates on the Electoral College outcome and considers finer details such as third-party candidates and self-proclaimed undecided voters. We incorporate our estimators into a dynamic programming algorithm to determine the probability that a candidate will win an election.

Presidential Election Forecasts

This paper contrasts election forecasting models that include macro-economic and financial market variables with forecasts by Fair and the Iowa Electronic Market (IEM), as well estimates made by widely recognized pollsters. The Financial Market Model proposed here incorporates the impact of election year stock market changes, including information about the recent financial crisis. However, the Financial Market Model grossly overestimates the challenger's vote share. The IEM forecast of 53.0 percent, the Fair forecast of 51.9 percent, and the Macro-Economic Market model forecast of 51.5 percent were close to the actual vote share of 53.2 percent for the Democratic nominee, but the Financial Market forecast of 59.4 in this paper would have made the president-elect extremely happy.

The state of presidential election forecasting: The 2004 experience

This paper assesses the current state of U. S. presidential election forecasting, describing forecast methods and their predictive accuracies for the most recent election, 2004. Three types of forecasts were made for the election using the methods noted: 1) point forecasts of the popular vote (by campaign polls, futures contracts on candidates' performance, regression models, Delphi expert surveys, and a combination of forecasts from these methods); 2) point forecasts of the electoral vote (by regression models, probability models based on state polls, a compilation of median polls in states, and exit polls); and 3) dichotomous forecasts of the popular-vote winner (by a multi-indicator index, cut-points for single indicators, and bellwether states). Candidate futures provided the most accurate popular-vote forecasts. A state probability model and the median state poll technique were the most accurate electoral vote methods. All three dichotomous techniques successfully predicted the election winner.

Evaluating U.S. presidential election forecasts and forecasting equations

This article examines four problems with past evaluations of presidential election forecasting and suggests one aspect of the models that could be improved. Past criticism has had problems with establishing an overall appraisal of the forecasting equations, in assessing the accuracy of both the forecasting models and their forecasts of individual election results, in identifying the theoretical foundations of forecasts, and in distinguishing between data-mining and learning in model revisions. I contend that overall assessments are innately arbitrary, that benchmarks can be established for reasonable evaluations of forecast accuracy, that blanket assessments of forecasts are unwarranted, that there are strong (but necessarily limited) theoretical foundations for the models, and that models should be revised in the light of experience, while remaining careful to avoid data-mining. The article also examines the question of whether current forecasting models grounded in retrospective voting theory should be revised to take into account the partial-referendum nature of non-incumbent, open-seat elections such as the 2008 election.

US presidential election forecasting: An introduction

The modern era of election forecasting is now almost thirty years old. Like many significant developments in science, it began as separate streams of work occurring at around the same time: the investigations of Lee Sigelman (1979), Michael Lewis-Beck and Tom Rice (1982), Richard Brody and Sigelman (1983) on the relationship between presidential approval ratings and the vote; economist Ray Fair's (1978) work on the predictive strength of the economy and presidential incumbency for the vote; Steven Rosenstone's (1983) construction of a forecasting model for the presidential vote in the States; the identification by geophysicist Vladimir Keilis-Borok and historian Allan Lichtman (1981) of thirteen general indicators or “keys” for foretelling the outcome of presidential elections. Soon thereafter, Michael Lewis-Beck and Tom Rice (1984) combined the presidential approval ratings with an indicator of economic conditions to create a core forecasting equation based on retrospective voting theory. Alan Abramowitz (1988) later added to this core idea his “time for a change” consideration of the number of consecutive terms that the president's political

Deconstructing the 2004 Presidential Election Forecasts: The Fiscal Model and the Campbell Collection Compared

In this paper we compare the August 1st forecast of the 2004 presidential election generated with the fiscal model with those made with the seven models of what we call the “Campbell Collection,” after James E. Campbell, the editor or co-editor of several forecasting symposia that have appeared in the American Politics Quarterly (October, 1996) and in successive issues of PS: Political Science and Politics (October, 2004 and January, 2005). For reasons that will become evident, Ray Fair's “presidential vote equation” (Fair 2002a, 2002b) is also included in the comparison. First, given space constraints, we present only a brief summary of the theoretical model. (For a more extended discussion, interested readers are encouraged to pursue the relevant references.) Then, as we did pursuant to issuing a forecast for the 2004 election, we estimate the model over the 22 elections held between 1916 and 2000, showing the out-of-sample results. Finally, we evaluate the model relative to the Campbell Collection on operational and substantive criteria.

After the Vote: Evaluating a Prospective Forecasting Model of Presidential Elections

There are two components to my model of presidential election forecasting: Given the enormous amount of work on voting behavior that finds prospective assessments of the economy to be strongly related to vote choice (e.g., Abramowitz 1985; Kuklinski and West 1981; Lewis-Beck 1988; Lockerbie 1992), I make use of a prospective economic item from the Survey of Consumer Attitudes and Behavior that asks if the next year will be better, worse, or the same for the respondent. I take the average of the negative responses to this question from the first quarter of the election year as my economic measure. These data are available in late April.

Introduction—The 2004 Presidential Election Forecasts

This symposium presents seven presidential election forecasting models and their predictions of the popular two-party vote in the 2004 election. The modern age of election forecasting is now into its third decade. Models have been tested quite publicly in the heat of battle—with some doing well, others not quite so well, and still others making way for new models. In this introduction, I provide a brief overview of the models, a summary of this year's forecasts, and some thoughts about how the forecasts should be judged.

Temporal Horizons and Presidential Election Forecasts

In this article we present a simple forecasting model that has been successful at predicting past presidential elections. The two variables included in the model are cumulative per capita income growth and presidential approval. These "fundamental" variables predict the vote especially well when measured shortly in advance of the election, when the outcome is already becoming clear in the polls. Their predictive power drops quite quickly as one steps back from the election, however; readings of income growth and approval taken early in the election year only modestly predict readings of the same variables just prior to the election. Thus we turn to leading economic indicators that allow us to forecast presidential elections by giving advance indication of changes in the economy and approval during the election year. For 1996, our model offers a cautious prediction of a Clinton victory.