Zweitstimme.org

We offer a dynamic Bayesian forecasting model for multiparty elections. It combines data from published pre-election public opinion polls with information from fundamentals-based forecasting models. The model takes care of the multiparty nature of the setting and allows making statements about the probability of other quantities of interest, such as the probability of a plurality of votes for a party or the majority for certain coalitions in parliament. We present results from two ex ante forecasts of elections that took place in 2017 and are able to show that the model outperforms fundamentals-based forecasting models in terms of accuracy and the calibration of uncertainty. Provided that historical and current polling data are available, the model can be applied to any multiparty setting.

Wir berichten die Ergebnisse einer ex-ante-Vorhersage der Zweitstimmenergebnisse für die Bundestagswahl 2017. Dazu kombinieren wir Daten veröffentlichter Umfragen mit strukturellen Informationen. Das Modell trägt den Eigenheiten von Vielparteiensystemen Rechnung und erlaubt es, Aussagen über die Wahrscheinlichkeit bestimmter Wahlergebnisse zu treffen, z. B. der Mehrheit der Sitzanteile für eine Partei oder der rechnerischen Mehrheit für verschiedene Koalitionsoptionen. Die Vorhersagen des Modells werden auf der Plattform zweitstimme.org veröffentlicht und aktualisiert. Unser Ansatz hat dabei nicht nur akademischen Wert: Wir geben Journalisten, Experten und Bürgern Informationen an die Hand, die helfen können, tatsächliche Parteiunterst/utzung einzuschätzen und letztlich besser informierte Wahlentscheidungen zu treffen.

While election forecasts predominantly focus on national contests, many democratic elections take place at the subnational level. Subnational elections pose unique challenges for traditional fundamentals forecasting models due to less available polling data and idiosyncratic subnational politics. In this article, we present and evaluate the performance of Bayesian forecasting models for German state elections from 1990 to 2024. Our forecasts demonstrate high accuracy at lead times of two days, two weeks, and two months, and offer valuable ex-ante predictions for three state elections held in September 2024. These findings underscore the potential for applying election forecasting models effectively to subnational elections.

Existing studies show that aggregating citizens’ expectations about who will win can predict election outcomes in a majoritarian system. But can so-called citizen forecasting also successfully predict outcomes in mixed-member systems, where constituency results are less important? The existing evidence is mixed and limited in scope. We conducted, therefore, a citizen forecast of the 2021 German federal election by administering an original survey asking citizens who they thought would win in their constituency, what share of the vote each candidate would win in their constituency, and what share of the vote each party would win nationally. Citizens predicted constituency winners and vote shares more accurately than several benchmarks. However, our citizen forecast was based on a non-representative sample from an online-access panel. We conclude that citizen forecasting provides a simple and inexpensive way to predict the various relevant outcomes in mixed-member elections.

This article provides a forecast for the German Federal Election of 2025 using a combination of national-level forecasting models for party-vote shares and district-level models for candidate votes. By integrating these approaches, the authors generate predictions for coalition majorities in Parliament as well as the implications of recent electoral reforms that can lead to “vacant districts” where plurality winners may not obtain a Bundestag seat under the new rules. The forecasts offer insights into party vote shares, candidate outcomes, coalition probabilities, and the effects of the electoral law changes on representation.

Most democracies distribute parliamentary seats based on electoral districts. Due to a lack of district-level polls, forecasters project national voting trends to district-level outcomes – drawing substantial public attention from voters. We know little about how different assumptions about these models influence forecasts. We address this gap by comparing uniform and proportional swing models, alongside additional variants. Using data from the past eight German federal elections, we assess predictive performance and complement the analysis with a simulation study. The findings from Germany demonstrate that while differences between swing models are generally modest, model choice affects the precision of predictions. The simulation further reveals that uniform swing performs better with larger national swings, lower volatility between districts, and more parties.