Citizens’ Forecast for the 2021 German National Election
Mark Kayser Hertie School, Berlin
Arndt Leininger Chemnitz University of Technology
Andreas Murr University of Warwick
Lukas Stötzer Humboldt-Universität zu Berlin
23.09.2021
Who will win how many constituencies in the 2021 German national election? Although Germany uses a mixed-member-proportional system, that is a particularly pressing question in 2021. Not only because many constituencies are prominently contested – e.g., 061 Potsdam, where Chancellor candidates Scholz (SPD) and Baerbock (Greens) are vying for the so-called direct mandate –, some races are likely to be very close, the Greens are hoping for more and the AfD for regaining constituency seats, but also because the distribution of constituency seats will have a decisive influence on how large the 20th German Bundestag will be.1
Unlike for the second vote (Zweitstimme in German) with the so-called “Sunday question” (Sonntagsfrage in German), there are very few to no surveys for constituencies – where the first vote (Erststimme in German) decides the direct mandate. National polls cannot easily be used to forecast constituency results because, first, they usually only ask about the second vote, and second, there are at most a handful of respondents available for each constituency, while some constituencies cannot be covered at all. We, therefore, take a different approach with our citizens’ forecast.
We surveyed a non-representative sample of citizens in all 299 Bundestag constituencies – but we didn’t ask who they would vote for; we asked who they thought would win. Specifically, we asked at least 20 people per constituency about the winner of the district vote, the distribution of first votes in the constituency, and the distribution of second votes nationwide. The idea behind our approach is simple: even the most ardent supporter of the FDP, for example, will be able to realistically assess that their party has no chance of winning the direct mandate, for instance, because the candidates of other parties have been more successful in the past. Furthermore, they know the local area and can assess how the constituency has developed demographically and politically in the last four years. We, therefore, hope that our citizen prediction will fare better than the so-called “unit” or “proportional swing” approaches (see, e.g. zweitstimme.org) or simple heuristics like “The same party as last time wins.” Indeed, these cannot predict changes of direct mandates from one party to another that run counter to the general federal trend. This procedure has already been successfully implemented for election predictions in other countries such as the US or the United Kingdom, which use a pure majoritarian election system. How successful can citizens’ predictions be in Germany? This and more is what we endeavor to find out with this project. We will be a bit wiser after the election night and the announcement of the preliminary official results. In the following, we document our Citizens’ Forecast.
Citizens’ forecast of constituency winners
The map (Fig. 1) presents our forecast based on respondents’ expectations. In 286 constituencies, a relative majority of respondents’ expectations fell in favor of one single candidate. In 23 constituencies two candidates were tied in expectations (see also Tab. 1). But also in other constituencies, it must be emphasized that one can expect a close race. In many constituencies, the first-placed candidate is only slightly ahead of the second-placed candidate in our respondents’ expectations. The tooltip after a mouseover in Fig. 1 provides information on what proportion of respondents in the constituency predicted the respective candidate as the winner.
Fig. 1: Citizens’ forecast of direct mandate winners and runners-up in the constituencies for the 2021 federal election
According to the expectations of our respondents, the CDU is likely to win the most direct mandates, as in 2017 (see Tab. 1). However, it will fall short of its result of 185 direct mandates in 2017 with 129 direct mandates if the expectations of our respondents come true.2 Trusting our respondents, the CSU can again have hope for all 46 Bavarian direct mandates. Only in Fürth does a majority of respondents see the SPD candidate ahead. In 98 constituencies, the majority of respondents believe the SPD will win the direct mandate. That would be 39 seats more than in 2017 and almost double the number of direct mandates.
| Party | Constituency seats |
|---|---|
| CDU | 129 |
| SPD | 98 |
| CSU | 45 |
| Too close to call | 13 |
| Grüne | 7 |
| AfD | 5 |
| Linke | 2 |
| Andere | 0 |
| FDP | 0 |
The Greens can hope to defend their only direct mandate to date in Berlin-Friedrichshain-Kreuzberg - Prenzlauer Berg Ost and to win six more direct mandates. The majority of respondents in Kiel, Frankfurt am Main II, Stuttgart I, Esslingen, Heidelberg, and Freiburg expect a Green direct mandate in their constituency. Only respondents in Berlin-Lichtenberg and Berlin-Treptow-Köpenick see the Left in front, while a majority of respondents in Berlin-Marzahn expect the CDU to take the direct mandate from the Left. Two direct mandates will not be enough for parliamentary party status if the party fails to clear the five-percent hurdle nationwide. A majority of respondents in what are all Saxonian constituencies expect the AfD to win the direct mandate in their constituency. The FDP is not expected to win a direct mandate in any constituency by a majority of those surveyed there. The party is thus likely to remain the only parliamentary group without direct mandates.
In the sortable and searchable Tab. 2 the forecasts for all constituencies are tabulated again.
Citizens’ prediction of the nationwide second vote result
We also asked all respondents about their expectations for the national election result by second votes, which determines the distribution of seats in the German Bundestag. All respondents could distribute vote shares among the candidates of all parties represented in the Bundestag and the residual category “Others,” which had to add up to 100% in total. As a forecast of the election outcome, we calculated the unweighted arithmetic mean of all responses per party (see Tab. 3). We also calculated the median and, to represent the variance in respondents’ estimates, the standard deviation.
| Party | Forecast (Mean) | Prognose (Median) | Standard Deviation | N |
|---|---|---|---|---|
| SPD | 25.0 | 25 | 10.4 | 6448 |
| CDU/CSU | 23.4 | 22 | 10.4 | 6448 |
| Grüne | 16.4 | 17 | 8.1 | 6448 |
| AfD | 10.5 | 10 | 10.8 | 6448 |
| FDP | 10.4 | 10 | 6.8 | 6448 |
| Linke | 7.9 | 7 | 7.0 | 6448 |
| Andere | 6.4 | 5 | 10.8 | 6448 |
Our second-vote forecast is based on the responses of a total of 6448 respondents. We have additionally visualized the distribution of respondents’ answers in a so-called boxplot (see Fig. 2`).3.
Fig. 2: Citizens’ prediction of the nationwide second vote result
The authors
Mark A. Kayser, PhD is Dean of Research and Faculty and Professor of Applied Quantitative Methods and Comparative Politics at the Hertie School, Berlin. He works on party competition and electoral behavior and developed forecasting models for the 2013, 2017, and 2021 federal elections.
Arndt Leininger, PhD is an assistant professor of political science research methods at Chemnitz University of Technology. He works on political behavior and attitudes. He developed forecasting models for the 2013, 2017, and 2021 federal elections and advised YouGov in 2017 on the development of the YouGov Election Model.
Andreas Murr, PhD is Associate Professor in Quantitative Political Science at the University of Warwick (England). He works on political behavior and public opinion. He developed citizens’ forecasting models for the UK House of Commons elections 2015 and 2019 and the US presidential election 2020, among others.
Dr. Lukas F. Stoetzer is a research fellow at the Cluster of Excellence Contestations of the Liberal Script (SCRIPTS) at the Humboldt University of Berlin. He works on voting behavior and quantitative political methodology. For the last Bundestag election, he developed the prediction website zweitstimme.org.
Methodology
We commissioned respondi to survey a non-representative sample of eligible voters online about their expectations for the election outcome (field time: 6–22 September 2021). We asked about expectations for the winner of the direct mandate, the distribution of first votes in the constituency, and the nationwide second-vote result. The question about the direct mandate winner was, “Which candidate will win the most votes in your constituency in the Bundestag election?” The first-past-the-post question asked, “What percentage of first-past-the-post votes will the various candidates receive in your constituency?” - respondents could distribute vote shares among the candidates of all parties represented in the Bundestag and the residual category “others,” which had to add up to 100% in total. The question on second vote shares was, “What percentage of the vote do you expect the various parties will receive nationwide?” - again, respondents could distribute vote shares summing to 100%. Respondents were previously assigned to their constituency via their ZIP code to show them the names of the candidates of the respective parties in their constituency. At least 20 respondents were interviewed per constituency (\(\bar{N}=21.6\)). This number of respondents would be clearly too small for a survey on individual voting intentions, but if one asks about the expected outcome of the election, it has been shown in the U.S. and the U.K., among others, that good forecasts can be obtained even with a small number of respondents per constituency.4
The prediction of direct mandate winners is made using the modal value of named candidates, that is, the candidate named most frequently is predicted to be the winner. The map also indicates for first- and second-place candidates how many respondents (in %) in the constituency saw the respective candidate as the likely winner. The forecast for the national election result by two votes is calculated by the arithmetic mean (table 3) and, alternatively, the median (figure 2).
Forecasts for all direct candidates (of the parties represented in the Bundestag plus “Others”) in all 299 constituencies can be found in this Google Sheet. The raw data of the survey will be published at a later date as part of a scientific publication.
An explanation of German electoral law and why the total number of seats in parliament depends on the election result can be found at Deutsche Welle.↩︎
For the current distribution of direct mandates among the parliamentary groups in the 19th German Bundestag, see https://www.bundestag.de/abgeordnete/biografien/mdb_zahlen_19/direktmandate_landeslisten-529512.↩︎
The “box” visualizes the range in which 50% of our respondents’ answers lie, the vertical line in the box indicates the median value. See also https://de.wikipedia.org/wiki/Box-Plot for further explanation↩︎
See, e.g., Murr, A., Stegmaier, M., & Lewis-Beck, M. (2021). Vote Expectations Versus Vote Intentions: Rival Forecasting Strategies. British Journal of Political Science, 51(1), 60-67. doi:[10.1017/S0007123419000061](https://doi.org/10.1017/S0007123419000061)↩︎