Voter support for German political parties 2017–2021: A political data story on upcoming elections in a changing party landscape.
On 26 September 2021 Germany will hold its first general election since 2002 without Angela Merkel as a candidate to become prime minister. After 16 years (!) with the same prime minister (George W. Bush was still president of the United States when Merkel first took office) Germany is entering a new political era. Recent polls suggest indeed that the German party system is about to change after decades of two dominant political forces: CDU/CSU (centre-right) and SPD (centre-left). What is happening in German politics — known for its relative stability since 1949? Which impact had the COVID-19 pandemic?
This blogpost analyses voter support for German political parties based on a statistical model that combines data from 854 polls from 8 polling firms over the current legislative term (2017–2021).¹ This approach allows to take the uncertainty of individual polls (precision) and systematic deviations of polling organisations (bias) into account. Furthermore, this method enables more precise measurements of political parties’ electoral support by pooling over multiple polls providing also estimates during (short) time intervals where no polling data is available. The results indicate that voter support for political parties evolved substantially since 2017 — before and after the outbreak of the COVID-19 pandemic. The changing balance between political forces may also have consequences for Germany’s role in European and international politics.
Towards a new balance in Germany’s party system?
Throughout this paragraph I will describe how the (estimated) voter support for political parties evolved since the general election in September 2017. Figure 1 shows the development of voter support over the past 3.5 years for 7 political parties that are currently represented in the German Parliament (Bundestag).²
We can see three remarkable evolutions with regard to the electoral support of German political parties between 2017 and 2021 (in chronological order). First of all, it seems as if the (already historically small) difference between the two dominant parties CDU/CSU (Christian democrats) and SPD (social democrats) and the third largest party has completely vanished. The tipping point appears to be in 2018 when three parties (temporarily) shared the second position (which has never happened since 1949). While CDU/CSU and SPD had both attracted at least 10 percent more of the votes than the third largest party in 2017, at least one of the previously ‘small’ parties has broken their historical dominance. The trend of traditional parties in decline may not be over yet in Germany. Second, The Greens (Bündnis 90/Die Grünen) have substantially gained support among voters since the previous general election. Maintaining high levels of voter support over more than two years (exceeding their best electoral results so far), The Greens seem to have arrived at their (temporary) peak. Third, the COVID-19 pandemic could have an impact on the results of the 2021 general elections. Most importantly, the party of the current prime minister (Angela Merkel, CDU/CSU) has dramatically won support among voters since the classification of COVID-19 as a pandemic — mostly at the expense of The Greens. This development could be a sign of voters’ trust in Angela Merkel as a crisis manager (or in the authority of leading politicians more generally). Yet, that effect has disappeared with the increased containment measures after the second wave by the end of the year.
Who will follow Merkel as a new prime minister?
The recent changes in the German party landscape may not remain without consequences for the negotiations to form a new coalition that will elect a prime minister.³ In contrast with previous elections there are at least three political parties (CDU/CSU, SPD, The Greens) that have nominated a realistic candidate to become the successor of Angela Merkel as prime minister after the 2021 general election.⁴ Since the prime minister typically belongs to the largest party of the parliamentary majority, this blogpost focuses primarily on the three largest parties in order to answer the question who will eventually follow Angela Merkel as the new prime minister.
Figure 2 shows the increasing approximation of the three parties towards exceptionally close levels of voter support in a historical perspective. At the moment, CDU/CSU and The Greens can rely on the strongest support from voters followed by SPD. To get an idea how the electoral support for all three parties could plausibly evolve until the elections in September 2021, the figure also indicates potential future developments based on estimations from our model (light-shaded). The predictions take the polling data as well as a priori known historical records into account (trend over the last federal elections as well as midterm regional/European elections).⁵ Predictions based on this model suggest that CDU/CSU appears to have the best chances to receive the highest vote share and their candidate Armin Laschet hence to become prime minister. Nevertheless, also The Greens can still reasonably hope to end up on the first place with a predicted vote share between 15 and 24 percent. Finally, even the social democrats (SPD) may still have realistic chances to nominate the prime minister in a federal coalition without CDU/CSU if they arrive before The Greens.
While CDU/CSU has constantly been the largest party since 2005 (when Merkel first became prime minister), The Greens have never reached more than 10.7 percent of the votes during German federal elections. Despite some fluctuations, the rise of The Greens from the smallest political party in parliament seems to have been rather steady since 2017. (The development until the COVID-19 pandemic in March 2020 is pretty much linear with two bubbles following the allegedly increased media attention after The Greens had for the first time outperformed SPD in the polls and the European election). There might be three possible reasons for their sudden growth. First, the ‘constructive role’ during the ultimately failed government negotiations with two centre-right parties (CDU/CSU, FDP) after the 2017 general election proved to be a serious opening towards the centre of the electorate. Second, the election of the two party leaders in January 2018 resulted in the adoption of a less divisive style and high personal popularity ratings.⁶ Third, the increased voter support since April 2021 may eventually represent a short-term boost of popularity following the official nomination of a prime ministerial candidate at the party convention.
It is noteworthy, however, that the electoral support of The Greens appears to be strongly correlated with the two other ‘large’ parties. Although the aggregate data analysis does not allow any conclusion about the development at the level of individual voters, it seems as if The Greens may not only have gained support from centre-left voters of SPD but also from voters of the centre-right CDU/CSU. Given the large ideological divide between CDU/CSU and The Greens which made a federal coalition between both parties unthinkable over many decades, the fluctuation of voters between both parties would be more of a surprise. While the candidates for prime minister’s office from all three parties are well-known for their ideologically moderate positions within their own parties, the three political parties are far from ideologically homogeneous. From a systemic perspective, the open competition between three parties might therefore allow a better representation of opposing positions on central societal questions. The (indirect) competition about prime minister’s office between the three parties may for example lead to more precise choices of voters with regard to the transformation of the welfare state than the competition between two major political forces. This would allow voters who want to influence the election of the prime minister to adopt more precise positions on 1) the quantity of public spending (more or less government revenue/expenditure) and 2) the primary focus of public expenditure (long-term investments such as education vs. immediate social security such as unemployment schemes/pensions).
In any case, the race between three prime ministerial candidates has made the electoral outcome of the three largest political parties more unpredictable. The question who will become prime minister is hence for German standards (where never more than two candidates had realistic chances) extremely difficult to answer. In the context of the strong competition to become the largest political party at the 2021 general election, it might appear cynical to note that the election of Germany’s next prime minster may largely depend on absent voters⁷ and the electoral support of the remaining ‘small’ parties.
The impact of the ‘small’ parties
As the polls suggest that a federal coalition may require more than two political parties, the ‘small’ parties can expect to have some influence over the choice of a potential government coalition (and the prime minister). Depending on the respective vote share of the three currently somewhat smaller political parties (AfD, FDP, The Left) certain coalitions may become more or less likely.
While the far-right AfD has quickly grown after the rising numbers of asylum seekers in 2015, their support among voters has decreased since fall 2018 (see Figure 3). The declining electoral support for AfD has among others coincided with the public debate about their role in fascist violent outrages during demonstrations in 2018 and the seemingly established influence of an extreme-right faction within the political party at the 2019 party conference.⁸ Even though the electoral result of AfD may have an impact on future positions of other parties (notably CDU/CSU), all political parties currently represented in parliament exclude any formal collaboration with the far-right party. Consequently, the electoral support of the liberal party (FDP) and the left-wing populist party The Left (Die Linke) could be most decisive for negotiations to form a federal government after general elections in September 2021. The recently surged electoral support of FDP following the second lockdown to contain the COVID-19 pandemic in Germany starting in December 2020 might put the liberals into a particularly comfortable position as a kingmaker for several potential federal coalitions. However, the growth of the liberals makes a federal coalition that includes the currently governing CDU/CSU (eventually with The Greens or SPD) most likely given the strong strategic and ideological links between CDU/CSU and FDP.⁹ In contrast, The Left might be required as a potential coalition partner if The Greens and the social democrats (SPD) would strive to form a more leftist coalition. Since none of the federal coalitions that has previously governed since 1949 seems to gain sufficient electoral support, the elections may result in a substantially longer and more difficult negotiation period. Therefore, the German general election in September 2021 could also mark a further step towards the ‘normalisation’ of German politics in a European context where more volatile elections require political parties to form involuntarily more ideologically heterogeneous coalitions.
As the voter support for a political party cannot be directly observed, its estimation requires a more or less complex statistical model. In analogy to previous studies, this approach assumes that the result of every single poll i is a random sample from a normal distribution with the (latent) voter support alpha plus a systematic deviation (bias) of the poll delta as its mean and a standard deviation sigma.¹⁰ One possible approach to estimate systematic deviations of polling organisations is based on the final electoral outcome and the assumption that the electoral support of a political party at time t is informed by its electoral support at time t-1 plus a random error/external shock (random walk) (see e.g. Jackman 2005).¹¹ Since the election results of September 2021 are not (yet) known in our case, this model relies on the electoral outcome of the 2017 general elections to estimate systematic deviations of polling firms (this approach has also been used by Fisher et al. 2011).¹² Furthermore, this model assumes that there is also some information about the outcome of the 2021 general elections that is known a priori. Therefore, this model employs a moderately informed prior — a normal distribution with mean h and standard deviation sigma.h — for the electoral support of a political party at election day in September 2021 as well as a reverse random walk (see Linzer 2013). In order to estimate h and sigma.h I make use of a combination of the predicted values from 1) a linear model with the electoral result of a political party at federal elections since 2000 as dependent variable (linear long-term trend at the federal level) and 2) the deviation of the midterm election results at regional and European elections since September 2017 compared to the previous election result (medium-term trend at the regional/European level). The model has been estimated based on Markov Chain Monte Carlo estimation (MCMC) with two Markov chains with each 100.000 iterations after 100.000 adaptive iterations and 100.000 iterations burn-in.¹³
This political data story is not part of my research project at Ghent University but rather a personal hobby project in my free time (that arguably went a bit out of hand). I am solely responsible for the content of this article.
 I make use of polling data (‘Sonntagsfrage’) published at https://www.wahlrecht.de/index.htm. The analysis takes all polls into account that were published between 24 September 2017 and 26 May 2021.
 The Bundestag is the lower chamber of the German federal parliament that is directly elected by voters. While CDU/CSU are separate parties, they form one parliamentary party group and both parties have an electoral alliance choosing not to run against each other in the same constituency. Consequently, I consider both parties as (effectively) one political party for this blogpost.
 Germany is a parliamentary democracy where a majority of representatives from the directly elected parliament composed of multiple parties (proportional representation) elects the prime minister.
 Contrastingly with European elections, prime ministerial candidates (‘Spitzenkandidaten’) are typically not withdrawn after elections by German parties making them serious competitors for prime minister’s office.
 These predictions are still rather rough estimates to show plausible future trends. Still more accurate predictions might be possible by taking further factors into account, such as: financial and organisational resources for the electoral campaign, candidate popularity, contextual factors (socio-economic development, epidemiological evolution/anti-COVID-19 measures), strategic voting (to support a favoured government coalition instead of voting for the most preferred political party), etc.
 The changed style was particularly visible with regard to two aspects: 1) the abolition of the strict representation of dominant party factions, 2) an early German-wide campaign listening to citizens’ demands during town hall meetings by one of the two party presidents (Habeck).
 Since voting is not compulsory at German federal elections, the share of registered voters that did not participate during the two previous elections varied between 23,8% in 2017 and 28,5% in 2013.
 Right-wing extremism is a particularly sensitive issue in German politics due to its fascist history which has led to very low levels of electoral support for political parties of the far-right since 1949. The further radicalisation of AfD has also been noted by the German internal intelligence agency which is examining the observation of the political party after recently declaring the party faction ‘Der Flügel’ as a right-wing extremist threat for the constitutionally anchored free democratic order.
 CDU/CSU has clearly been the most preferred coalition partner of the liberals in the past.
 The variance of a poll depends on the vote share of the political party in a poll and the amount of responds that participated in the poll. Since polls often do not make use of a random sample (but make use of post-stratification weights) I control in analogy to e.g. Fisher et al. (2011) also for a potential design effect of the polls.
 Since the analysed period here stretches over several years, I employ weeks as time intervals where t=1 is election day in 2017 to facilitate more rapid model convergence.
 In this model I consider the 2017 elections as a poll at time t=1 based on a random sample of 1.000.000 respondents. This approach relaxes the assumption that elections are perfectly precise estimations of voter support.