Article

Forecasting the 2013 Bundestag Election Using Data from Various Polls

Authors:
If you want to read the PDF, try requesting it from the authors.

Abstract

CDU/CSU 38.1%, SPD 28.2%, B'90/Die Grünen 13.5%, Die Linke 7.7%, FDP 5.4%, others 6.5% – these are our (point) estimates of the party vote shares at the upcoming Bundestag election. In contrast to the predominant academic approach to forecast incumbent vote shares from measures of government popularity, economic conditions and other fundamental variables (Jérôme, et al. 2013, Kayser and Leininger 2013, Norpoth and Gschwend 2013), we entirely rely on data from published trial heat polls. Opposite to common practice in the news media, we do not take isolated polls as election forecasts in their own right. Instead, we use historical data to empirically assess the relationship between polls and election outcomes, and combine extrapolations from 42 current polls in a Bayesian manner. A retrospective evaluation of our method compared against other approaches will be given after the election on September 22.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... Election forecasting in Germany was pioneered by Gschwend Selb and Munzert's (2013) approach is similar to the above mentioned ones in using data from prior elections to arrive at a forecast well ahead of the elections. They differ in that they exclusively rely on polling data. ...
... To provide a fair comparison of the accuracy of the models we compare their mean absolute errors (MAE). The model by Selb and Munzert (2013) has the greatest lead but an MAE of 2.82 percentages points. Our model has the smallest MAE of .75 percentage points but we also have the smallest lead of just two months and we only predict coalition vote share. ...
Conference Paper
Full-text available
Abstract In early August 2013 we published a forecast relying on a theory-driven empirical model of election outcomes that draws on previous election outcomes; characteristics of the government and of voters; and, most originally, the relative economic performance of Germany (“benchmarked” growth) in comparison to the three other most important economies in Europe, France, the UK and Italy. Our forecast put the outgoing coalition government of CDU/CSU and FDP at 47.05% of the popular vote. This makes our forecast one of the most accurate ones among the forecasts that have been published prior to the election. Our forecast was also on par with the last poll results collected only days before the election. In addition to the presentation of our model we offer a post-mortem analysis of our forecast comparing it to other forecasts, polling and of course the actual election results.
... The problem of understanding and predicting election outcomes has long been part of political science research. However, lack of pre-election poll data especially in developing countries is one of the main challenges associated with forecasting election outcomes [1][2][3][4][5][6][7][8]. Nevertheless, in developed countries opinion polls are now easily accessible through online polling. ...
... Leaving casual punditry aside, prognosticators can turn to structural models, survey polling – asking respondents whom they will vote for or who they think will win – political betting markets, poll averaging techniques and more recently, synthetic models (Erikson and Wlezien, 2014 ; Beck and Dassonneville, 2015 ) that mix structural forecasts and polls. The 2013 Bundestag election saw a welcome expansion in the number of election forecasts in a country that has historically had few. 3 In contrast to many of the other forecasts, we constructed a structural forecasting model that made no attempt at improving methods (cf., for example, Selb and Munzert, 2013; Graefe et al., 2014 ; Jérôme, Jérôme-Speziari and ); rather we employed simple linear regression with some of the most conventional vote-choice predictors identified in electoral research. We added only a single innovation intended to test whether voters use cross-national comparisons to assess the economy: in place of economic growth, we employed the deviation in German real GDP growth from that of its largest European benchmarks, the UK, France and Italy. ...
Article
Full-text available
Do voters judge their national economy relative to economic performance abroad? In 2013 we took advantage of the German Bundestag election to test this hypothesis predictively. Nearly two months prior to the election, we published an election forecast relying on a theory-driven empirical model of election outcomes that draws on previous election outcomes; characteristics of the government and of voters; and, most originally, the relative economic performance of Germany (" benchmarked " growth) in comparison to the three other most important economies in Europe, France, the UK and Italy. Our forecast put the outgoing coalition government of CDU/CSU and FDP at 47.05% of the popular vote deviating from the actual outcome of 46.3 by 0.75 points. This makes our forecast one of the most accurate in this election cycle. Despite one and a half months of lead time, our forecast performed on par or slightly better than the last poll results issued only two days before the election.
Article
In this paper I present an election forecasting approach to predict the vote share of the governing coalition in German national elections. The model is composed of two independent prediction components: the first is based on poll data, the second on fundamental variables. Both approaches have their advantages and disadvantages when used in isolation. The basic idea is to use both and find a better informed overall forecast. The predictions are combined using a shrinkage estimator, where the predictions are weighted by their respective prediction uncertainty. The uncertainty of the poll prediction is modelled time-dependent. The result is a dynamic model allowing for predictions longer before the elections highly relying on fundamental variables. With the elections coming closer predictions rely more and more on the polling data.
Article
Full-text available
In the run up to the German federal election of 2013, forecasts about who would win attracted considerable attention. Leaving casual punditry aside, the more serious forecasts were predominantly based on opinion polls. Polls offer a good snapshot in time, but when the election is distant, they predict outcomes less well – a point stressed by pollsters themselves but often ignored by the media and the wider public. However, the 2013 election also saw an unprecedented number of scientific forecasting models that explicitly aimed at forecasting the election result, in some cases months in advance. This overview of scientific election forecasting in Germany focuses on so called structural models. These utilize theory and empirics of electoral research for their predictions. Furthermore, survey aggregation methods as well as synthetic models that combine these with structural models are covered and the performance of such approaches is discussed in the context of the German general election in 2013 and address the question what, if anything, scientific election forecasting can contribute to electoral research.
Article
This paper outlines several methods for forecasting the next British general election on a daily basis from 20 months prior using opinion polls. It discusses their performance for previous electoral cycles and shows that the two models with the best historical record lead to substantially different predictions for 2015, but they can be averaged. The historical relationship between the polls and the vote suggests that government support rises substantially in the run up to elections, that Conservatives outperform but Labour underperform relative to the polls, and parties generally recover from low points or decline from high ones. Approximate prediction intervals and probabilities for key events are also generated. Despite the Conservatives trailing in the polls by 7 points in early October 2013, the models suggest a substantial Conservative lead at the 2015 election with a 64% chance of being the largest party and 42% chance of an overall majority. The estimated probability of a hung parliament (40%) is instructive for understanding the operation of the electoral system.
Article
Full-text available
The present study reviews the accuracy of four methods for forecasting the 2013 German election: polls, prediction markets, expert judgment, and quantitative models. On average, across the two months prior to the election, polls were most accurate, with a mean absolute error of 1.4 percentage points, followed by quantitative models (1.6), expert judgment (2.1), and prediction markets (2.3). In addition, the study provides new evidence for the benefits of combining forecasts. Averaging all available forecasts within and across the four methods provided more accurate predictions than the typical component forecast. The error reductions achieved through combining forecasts ranged from 5% (compared to polls) to 41% (compared to prediction markets). The results conform to prior research on US presidential elections, which showed that combining is one of the most effective methods to generating accurate election forecasts.
Article
Full-text available
Zusammenfassung??Unser Vorhersagemodell f?r Bundestagswahlen sagte das Stimmenergebnis der amtierenden Regierungskoalition (SPD und Gr?ne zusammen bei 42,3 %) bei der Wahl am 18. September bis auf 0,3 Prozentpunkte richtig voraus und das bereits lange vor der Wahl. Dieses Modell bezieht drei Faktoren ein: (1) die Popularit?t des amtierenden Bundeskanzlers, (2) den langfristigen W?hlerr?ckhalt der Regierungsparteien, und (3) ihre Abnutzung im Amt. Das Antreten der neunen Linkspartei machte eine Bereinigung der Kanzlerpopularit?t notwendig (unter der Annahme, dass die Anh?nger der Linkspartei ?berwiegend Kanzler Schr?der statt seiner Herausforderin w?nschen). Unsere Prognose sagte in jedem Fall den beiden regierenden Parteien gen?gend Stimmen voraus, um einen Sieg des Oppositionslagers (CDU/CSU und FDP) unrealistisch erscheinen zu lassen, solange die neue Linkspartei soviel Stimmen bekommt, wie die Umfragen signalisierten.
Article
Full-text available
Survey researchers have long speculated that there may be a link between nonresponse and measurement error—that is, people likely to become nonrespondents to a survey are also likely to make poor reporters if they do take part. Still, there is surprisingly little evidence of such a link. It could be that nonresponse is generally the product of one set of factors and reporting errors, the product of an unrelated set, or both nonresponse and reporting errors may be itemspecific so that no general relationship between the two is likely to emerge. Our study examined a situation in which we thought there would be a link between response propensities and the propensity to give inaccurate answers. We asked samples of voters and nonvoters to take part in a survey that included items about voting. Past research shows that nonvoters misreport that fact and that they are less likely than voters in general to take part in surveys. We thought we could heighten the differences between voters and nonvoters in both response rates and levels of misreporting if we characterized the survey as being about politics. However, only nonresponse biases were larger when the topic of the survey was described as political, and this difference was only marginally significant. These two ways of framing the study had even smaller effects on estimates derived from other items in the questionnaire.The overall biases in estimates derived from the voting items are very substantial, and both nonresponse and measurement error contribute to them.
Article
Full-text available
Die Gegenüberstellung von Wahlbörsenresultaten und Meinungsumfragen gehört zu den Ritualen, die in der Berichterstattung über Wahlen nach Bekanntgabe der Resultate einsetzen. Doch viele dieser Analysen beziehen sich nicht auf eigentliche Vorhersagen. In diesem Aufsatz stellen wir einen ähnlichen Vergleich an, beziehen uns aber auf Prognosen, die wir vor dem Wahlausgang für die Bundestagswahl 2005 errechneten. Die Analyse zeigt, dass die Wahlbörse Wahl$treet auch 2005 besser abschnitt als die kommerziellen Institute. Angesichts der außergewöhnlich großen Prognosefehler besonders der Umfrageinstitute diskutieren wir überdies im Licht der Social Choice-Theorie, welche Auswirkungen fehlerhafte Umfragen auf Wahlentscheidungen haben.
Article
A wide range of potentially useful data are available for election forecasting: the results of previous elections, a multitude of preelection polls, and predictors such as measures of national and statewide economic performance. How accurate are different forecasts? We estimate predictive uncertainty via analysis of data collected from past elections (actual outcomes, preelection polls, and model estimates). With these estimated uncertainties, we use Bayesian inference to integrate the various sources of data to form posterior distributions for the state and national two-party Democratic vote shares for the 2008 election. Our key idea is to separately forecast the national popular vote shares and the relative positions of the states. More generally, such an approach could be applied to study changes in public opinion and other phenomena with wide national swings and fairly stable spatial distributions relative to the national average.
Article
Opinion polls often give an indication of sampling error based on the standard conservative confidence intervals for a single binomial proportion. We give a lower bound for the simultaneous coverage probability when such intervals are applied to all of the proportions in a poll. This bound is equal to 1 – 2α. for small values of α. The same result is shown to apply to the standard confidence intervals for changes in proportions between surveys.
Article
This paper outlines and discusses techniques for three stages in forecasting parliamentary seats from British opinion polls: adjusting and aggregating published vote-intention figures from across different polls; forecasting how public opinion might change before election day; and predicting the seat totals from the forecasted election-day vote shares. Specifically, we consider a state-space model for opinion polls which correct for house effects and other sources of survey error, the estimation of the historical relationship between polls and the election-day share of the vote, and a probabilistic approach to predicting the winner in each constituency.
Article
Selten zuvor waren sich Medien und Politiker in ihrem Urteil so einig: Die Ergebnisse der Bundestagswahl 2005 waren ein “Debakel” (Süddeutsche Zeitung vom 20.09.2005, S. 2) der Demoskopie. In der gesamten vorangegangenen Legislaturperiode suggerierten die Zahlen der Wahlforscher, dass im Falle sofortiger Neuwahlen mit großer Wahrscheinlichkeit eine schwarz-gelbe Mehrheit zu erwarten sei. Sogar von einer möglichen absoluten Mehrheit der Union war kurze Zeit die Rede. Die erste Prognose am Wahlabend lieferte allerdings ein ganz anderes Ergebnis. Die Union erhielt das drittschlechteste Ergebnis ihrer Wahlgeschichte; die SPD schnitt besser ab als es die Vorwahlumfragen hatten vermuten lassen. “Die Bundestagswahl 2005 ist somit ein weiterer Tiefschlag für eine Branche, die bei Politikern und Medien vor allem deshalb beliebt ist, weil sie einen Blick in die Zukunft zu erlauben scheint”, schrieb der Spiegel. “Viele Demoskopen bedienen diese Erwartungshaltung gern, auch wenn sie leise hinzufügen, dass sie ja nur Momentaufnahmen lieferten” (Der Spiegel vom 19.09.2005, S. 63).
Article
There is little systematic research on the multitude of factors that influence the accuracy of poll results. This article examines six methodological factors directly under the survey researcher's control and two exogenous factors concerning the nature of public opinion as sources of survey error. Data for this study come from 56 ''trial heat'' polls conducted during the last month of the 1992 presidential election. The most important variables influencing survey accuracy were the number of days a poll is in the field, which increased total accuracy one-half of a percentage point per day; conducting interviews only on weekdays (and thus only during evening hours), which reduced overall accuracy rates by more than 1 percentage point; and conducting a ''tracking'' poll, which increased accuracy by about 1.5 points. Sample size was not related to accuracy rates. Results also indicated that sampling frames of ''likely voters'' (relative to ''registered voters'') tended to overestimate support for George Bush and underestimate support for Ross Perot, that interviewing only on weekdays led to overestimates of support for Bush, and that strict methods of defining a respondent as ''supporting'' a candidate hurt the two newcomers, Perot and Bill Clinton, more than Bush. In light of these data it is recommended that the common practice of reporting ''margins of error'' based solely on sample sizes be abandoned as misleading and replaced by a more empirically justifiable measure based more on response rates.
Article
Poll results vary over the course of a campaign election and across polling organisations, making it difficult to track genuine changes in voter support. I present a statistical model that tracks changes in voter support over time by pooling the polls, and corrects for variation across polling organisations due to biases known as ‘house effects’. The result is a less biased and more precise estimate of vote intentions than is possible from any one poll alone. I use five series of polls fielded over the 2004 Australian federal election campaign (ACNielsen, the ANU/ninemsn online poll, Galaxy, Newspoll, and Roy Morgan) to generate daily estimates of the Coalition's share of two-party preferred (2PP) and first preference vote intentions. Over the course of the campaign there is about a 4 percentage point swing to the Coalition in first preference vote share (and a smaller swing in 2PP terms), that begins prior to the formal announcement of the election, but is complete shortly after the leader debates. The ANU/ninemsn online poll and Morgan are found to have large and statistically significant biases, while, generally, the three phone polls have small and/or statistically insignificant biases, with ACNielsen and (in particular) Galaxy performing quite well in 2004.* An earlier version of this paper was prepared for the annual meeting of the Australasian Political Studies Association, University of Adelaide, 29 September–1 October 2004.
Article
We draw attention to a simple yet underappreciated way of forecasting the outcomes of elections involving two rounds of voting: surveying the voters’ candidate evaluations in first round exit polls, poststratifying the sample proportions of reported votes to official first round election returns, and redistributing the votes for eliminated competitors according to their supporters’ lower-order preferences among the viable alternatives in round two. We argue that the approach is likely to outperform standard pre-election surveys, due to its better coverage and reduced measurement error, and the possibility of correcting for sample selection. We set out the practical details of the method and demonstrate its usefulness by employing a recent German mayoral election as an empirical case. Thirteen candidates were competing in the first round, while there were six candidates in the decisive second round. The runoff result was forecast two weeks in advance with an average absolute error of less than one percentage point.
Article
Little is known about the evolution of electoral sentiment over the campaign cycle. How does the outcome come into focus as the election cycle evolves? Do voters’ preferences evolve in a patterned and understandable way? What role does the election campaign play? In this article, we address these issues. We translate general arguments about the role of campaigns into a set of formal, statistical expectations. Then, we outline an empirical analysis and examine poll results for the 15 U.S. presidential elections between 1944 and 2000. Our analysis focuses specifically on two questions. First, to what extent does the observable variation in aggregate poll results represent real movement in electoral preferences (if the election were held the day of the poll) as opposed to mere survey error? Second, to the extent polls register true movement of preferences owing to the shocks of campaign events, do the effects last or do they decay? Answers to these questions tell us whether and the extent to which campaign events have effects on preferences and, if so, whether these effects persist until Election Day. The answers thus inform us about what we ultimately want to know: Do campaigns have any real impact on the election outcome? The results of our analysis suggest that they do.
Article
To forecast an election means to declare the outcome before it happens. Scientific approaches to election forecasting include polls, political stock markets and statistical models. I review these approaches, with an emphasis on the last, since it offers more lead time. Consideration is given to the history and politics of statistical forecasting models of elections. Rules for evaluating such models are offered. Examples of actual models come from the United States, France and the United Kingdom, where this work is rather new. Compared to other approaches, statistical modelling seems a promising method for forecasting elections.
Article
A wide range of potentially useful data are available for election forecasting: the results of previous elections, a multitude of preelection polls, and predictors such as measures of national and statewide economic performance. How accurate are different forecasts? We estimate predictive uncertainty via analysis of data collected from past elections (actual outcomes, preelection polls, and model estimates). With these estimated uncertainties, we use Bayesian inference to integrate the various sources of data to form posterior distributions for the state and national two-party Democratic vote shares for the 2008 election. Our key idea is to separately forecast the national popular vote shares and the relative positions of the states. More generally, such an approach could be applied to study changes in public opinion and other phenomena with wide national swings and fairly stable spatial distributions relative to the national average.
Article
As most political scientists know, the outcome of the American presidential election can be predicted within a few percentage points (in the popular vote), based on information available months before the election. Thus, the general campaign for president seems irrelevant to the outcome (except in very close elections), despite all the media coverage of campaign strategy. However, it is also well known that the pre-election opinion polls can vary wildly over the campaign, and this variation is generally attributed to events in the campaign. How can campaign events affect people's opinions on whom they plan to vote for, and yet not affect the outcome of the election? For that matter, why do voters consistently increase their support for a candidate during his nominating convention, even though the conventions are almost entirely predictable events whose effects can be rationally forecast? In this exploratory study, we consider several intuitively appealing, but ultimately wrong, resolutions to this puzzle and discuss our current understanding of what causes opinion polls to fluctuate while reaching a predictable outcome. Our evidence is based on graphical presentation and analysis of over 67,000 individual-level responses from forty-nine commercial polls during the 1988 campaign and many other aggregate poll results from the 1952–92 campaigns. We show that responses to pollsters during the campaign are not generally informed or even, in a sense we describe, ‘rational’. In contrast, voters decide, based on their enlightened preferences, as formed by the information they have learned during the campaign, as well as basic political cues such as ideology and party identification, which candidate to support eventually. We cannot prove this conclusion, but we do show that it is consistent with the aggregate forecasts and individual-level opinion poll responses. Based on the enlightened preferences hypothesis, we conclude that the news media have an important effect on the outcome of presidential elections – not through misleading advertisements, sound bites, or spin doctors, but rather by conveying candidates' positions on important issues.
Article
Variance estimation techniques for nonlinear statistics, such as ratios and regression and correlation coefficients, and functionals, such as quantiles, are reviewed in the context of sampling from stratified populations. In particular, resampling methods such as the bootstrap, the jackknife, and balanced repeated replication are compared with the traditional linearization method for nonlinear statistics and a method based on Woodruff's confidence intervals for the quantiles. Results of empirical studies are presented on the bias and stability of these variance estimators and on confidence-interval coverage probabilities and lengths.Nous réexaminons dans le contexte d'échantillons tirés de populations stratifiées les techniques d'estimation de la variance pour statistiques non linéaires, telles que quotients, coefficients de régression et de corrélation, de měme que pour fonctionnelles, telles que quantiles. Nous comparons en particulier des méthodes de rééchantillonnages comme les méthodes d'auto-amorçage, de Quenouille-Tukey et de répliques équilibrées répétées avec la méthode traditionnelle de linéarisation pour les statistiques non linéaires et une méthode basée sur des intervalles de confiance de type Woodruff pour les quantiles. Nous présentons les résultats d'études expérimentales sur le biais et la stabilité de ces estimateurs de variance de même que sur les probabilités de recouvrement et les longueurs des intervalles de confiance construits à partir de ces derniers.
Article
We develop a method for efficiently using current poll data to update election forecasts based on historical relationships. The method is applied on an ex post basis to forecast Republican and Democratic shares of the national vote in U.S. Presidential elections from 1952 to 1992. Using poll data substantially improves the performance of forecasting models that rely solely on historical fundamentals. Moreover, when poll data are used appropriately, their information content dominates in the calculation of an optimal forecast. The method has also been applied on an ex ante basis to forecast the 1996 presidential election, producing a series of highly accurate predictions over the 2 month period before the election.
A Benchmarking Forecast of the 2013 Bundestag Election Die „Zauberformel “. Über das Erfolgreiche Scheitern des Prognosemodells von Gschwend und Norpoth bei der Bundestagswahl
  • Kayser
  • A Mark
  • Leininger
Kayser, Mark A., and Arndt Leininger. 2013. A Benchmarking Forecast of the 2013 Bundestag Election. Klein, Markus. 2005. Die „Zauberformel “. Über das Erfolgreiche Scheitern des Prognosemodells von Gschwend und Norpoth bei der Bundestagswahl 2005. Politische Vierteljahresschrift 46: 689-91
Sonntagsfrage Bundestagswahl. Wenn Am Nächsten Sonntag Bundestagswahl Wäre…
  • Matthias Cantow
  • Martin Fehndrich
  • Andreas Schneider
  • Wilko Zicht
Cantow, Matthias, Martin Fehndrich, Andreas Schneider, and Wilko Zicht. 2013. "Sonntagsfrage Bundestagswahl. Wenn Am Nächsten Sonntag Bundestagswahl Wäre…."
The Chancellor Model: Forecasting German Elections
———. 2010. "The Chancellor Model: Forecasting German Elections." International Journal of Forecasting 26: 42-53.
  • Markus Klein
Klein, Markus. 2005. "Die " Zauberformel ". Über das Erfolgreiche Scheitern des Prognosemodells von Gschwend und Norpoth bei der Bundestagswahl 2005." Politische Vierteljahresschrift 46: 689-91.