1 Introduction

Strategic voting is commonly defined as voting for a candidate or party that may not be the voter’s first choice, but is chosen to get the best possible outcome (on definitions of strategic voting see, among various others, Cox 1997). Strategic voting can occur both under plurality and proportional-representation (PR). It is, thus, also commonly observed in Germany’s mixed-member proportional representation (MMP) system (Gschwend 2007). With the list vote, voters can choose not to “waste” their vote. This strategy makes sense when a voter’s most preferred party has no chance of passing the 5% threshold (e.g., Kroeber et al. 2021). Likewise, voters can choose to cast a “rental vote” to help a party that runs at risk of barely missing the 5% threshold (Gschwend et al. 2016). Strategic voting is even more common in the first tier of the MMP electoral system (Herrmann and Pappi 2008; Herrmann 2010). Because it only matters which candidate receives the most votes under first-past-the-post (FPTP), voting for a candidate who has no chance of winning the election is irrational from a purely strategic perspective (Downs 1957, 48). Instead, strategic voters can cast a vote for a less preferred candidate with the intention of impeding the election of a candidate they strongly dislike. In short, strategic voting under FPTP implies that voters should only support candidates with some reasonable winning chances in the district (Cox 1997, 30).

Historically, with the clear dominance of the CDU/CSU and SPD in the German party system, strategic voting at the district level was straightforward. For left-wing voters, it was usually rational to support the SPD candidate and more right-wing voters should support candidates of the CDU/CSU.Footnote 1 However, as the electoral strength of the CDU/CSU and SPD has decreased over the last decade(s) (Schmitt-Beck et al. 2022), strategic voting and electoral coordination have become more complex. The electoral decline of the SPD poses a challenge especially for left-wing voters. The German General Election of 2021 clearly demonstrated that candidates of the Greens are serious contenders in many electoral districts.Footnote 2 This development, however, can lead to a somewhat paradoxical situation in which the increase in support for the Greens might result in a lower probability of winning an electoral district for both the SPD and the Greens as the vote shares for both parties become more balanced. Thus, if left-wing party voters cannot unite behind a single candidate, it may increase the chances of a candidate from the CDU/CSU or AfD winning the electoral district.

In other words, the political left faces a collective action problem in which many individuals have to collaborate to achieve a common goal. One way to overcome this collective action problem is to form a “pre-electoral coalition” (Golder 2006; Gschwend and Hooghe 2008) in which multiple parties “jointly back single-member district candidates” (Ferrara and Herron 2005, 17). However, the probability of such coalitions to occur is lower when all parties are of equal strength and are similarly optimistic to win the mandate. When coordination between parties fails, other political actors can try enforce it. One such option is to run large-scale campaigns with the goal to unite certain voters behind a candidate with the highest chances of winning against the competing candidate from the opposing political block. While such campaigns are relatively well known in the US context where Super-PACs run large-scale campaigns for individual candidates (e.g., Brunell 2005; Scala 2014), such campaigns are relatively unknown in the German context. This changed in the German General Election of 2021 when the large left-wing campaign organization Campact ran strategic voting campaigns in six electoral districts.Footnote 3 The goal of this campaign was to impede the election of certain CDU/CSU or AfD candidates and to help the most promising candidate from left-wing parties to win the electoral district. The campaign received much media attention because it was also conducted in the electoral district 196 (“Suhl-Schmalkalden-Meiningen-Hildburghausen-Sonneberg”) in Thuringia, where Hans-Georg Maaßen (in the following: HGM) ran for parliament on behalf of the CDU. Being the former head of the Federal Office for the Protection of the Constitution, HGM has received a lot of criticism for many statements, which have been labeled as right-wing populist. Campact, thus, started a campaign against HGM and decided to support the SPD’s candidate in the electoral district 196, because it could be reasonably argued that the SPD candidate had the highest winning chances among the candidates of the SPD, Greens, and Left Party. The campaign specifically targeted the Greens and the Left Party to convince them to withdraw their district candidates and to publicly announce support for the SPD candidate. While none of the parties formally withdrew their candidates, the Greens indeed announced public support for the SPD candidate two weeks prior to the election.Footnote 4 The Left Party never supported the campaign. Instead, they criticized it as potentially antidemocratic.Footnote 5

The different reactions of the Greens and the Left Party to the Campact campaign indicate that such campaigns are highly contested, despite being a potential solution to the coordination problem among left-wing parties. Therefore, this paper is interested in analyzing which factors influence public support for such campaigns. Using the case of HGM as a publicly well-known example, I conducted a survey experiment on a sample of 988 Green voters to test whether (1) providing information about the electoral system and winning chances of the SPD candidate (information treatment), (2) providing information about the support of the Green party for the campaign (party cue treatment), or (3) a combination of both treatments increases support for the Campact campaign. The results indicate that the party cue treatment has a strong effect on the support for the strategic voting campaign. Providing voters with more information about the relevance of strategic voting under plurality shows no significant effects. Moreover, I analyze the effect of other voter characteristics on support for the campaign. I find that the campaign is seen as more favorable among more left-wing, well-educated, and politically interested voters.

The paper proceeds as follows. The next section formulates theoretical expectations regarding the factors tested in the experimental design. The third section describes the sample and the experimental design in more detail. The fourth section describes the results and the final section discusses the implications of these findings.

2 Theoretical expectations

In the following, two different factors that might influence support for electoral coordination campaigns are discussed. First, relying on research about voters’ lack of information about the electoral rules (Schmitt-Beck 1993; Behnke 2015; Jankowski et al. 2022), I discuss whether providing more information about the strategic incentives to coordinate voters under FPTP increases support for such campaigns. Second, drawing on the established literature on party cues, the impact of party signals is discussed.

2.1 Electoral system information

By having a plurality as well as a PR component, MMP electoral systems combine two different principles of electoral system design. Proponents of these systems argue that this combination provides “the best of both worlds” (Shugart and Wattenberg 2003) in electoral system design as it allows for regional representation, personalization, as well as proportionality in the election outcomes. However, by having two votes, MMP systems are often more complex than purely FTPT or PR systems (Schmitt-Beck 1993; Karp 2006; Behnke 2015). In fact, empirical research demonstrates that many voters either do not know how the electoral system works or even have a false understanding of how the MMP system works. Schmitt-Beck (1993) was the first to demonstrate that voters in Germany often do not understand which of the two votes is more important for the election outcome. Likewise, Behnke (2015) demonstrates that – after accounting for random guesses in the responses – less than 50% understand the importance of the first and second vote correctly. The study further shows that the false understanding of the first and second vote might lead voters to irrational choices, for example by splitting votes in a non-reasonable manner.

This research is important for the research question at hand because it suggests that some voters might form their preferences for strategic voting campaigns based on a lack of knowledge or even a false understanding of the German electoral system. Having a profound understanding of how the German electoral system works is a precondition for understanding why electoral coordination is necessary and especially relevant for identifying the candidate who has the highest winning chances. Thus, one way of increasing the support for strategic voting campaigns could be to increase voters’ knowledge about how the German electoral system works and which candidate has the highest chance of winning the district. By providing such information the necessity for strategic voting is explained to the voters and should increase its legitimacy. I refer to this as “Information Treatment” and it is the basis for the first hypothesis:

Hypothesis 1

Providing voters with information about the German electoral system and the winning chances of the candidates in an electoral district increases support for electoral coordination campaigns.

2.2 Partycue

A long-standing strand of research on public opinion formation has highlighted that many voters rely on cues when forming an opinion on certain issues (Campbell et al. 1960; Lupia 1994; Bartels 1996; Lau and Redlawsk 2006). This is particularly true for issues where voters do not have strong preferences (Bechtel et al. 2015). Instead of acquiring and processing various types of information on a certain issue, it is often more rational for a voter to rely on the recommendation by a trusted source when forming an opinion on an issue (Achen and Bartels 2016). Such behavior is less time consuming, and it often leads to the same outcome as if voters had fully informed themselves on an issue (Lupia 1994; Lupia et al. 1998).

Following this line of reasoning, it might be particularly relevant for a voter to know how their preferred party positions itself regarding the support of the electoral coordination campaign. Thus, the second hypothesis reads as follows:

Hypothesis 2

Providing voters with information about the support of the coordination campaign by their party increases support for the campaign.

3 Research design

3.1 Sample

To test the hypotheses, I rely on a survey experiment conducted one week prior to the German General Election of 2021. The experiment was conducted on a sample of 988 Green voters. Focusing on Green voters instead of a sample including voters of all parties is reasonable because Campact’s strategic voting campaign primarily addressed voters of the Greens and Left Party. These voters were the “target group” of the campaign and analyzing how they react to such a campaign is, therefore, particularly interesting. I did not include Left Party voters in the sample due to other survey questions that also specifically addressed the preferences of Green voters, i.e., the experiment was part of a more general project on Green voter preferences during the German General Election of 2021. The sample was provided by the company respondi, and respondents came from all over Germany. I used quotas for gender and age to make the sample balanced with regard to these characteristics. Green voters were identified based on a standard vote choice question at the very beginning of the survey. Only respondents who indicated that they will cast their second vote for the Green party or had already done so by postal voting were able to take the full survey. All other respondents were screened out and could not participate in the study. Of course, the sample is not fully representative of all Green voters. However, given that it is quite challenging to survey a large sample of voters of a single party, the sample should be of relatively high quality compared to alternative sampling strategies, such as collecting responses via social media, in which the composition of the sample can hardly be controlled. Descriptive statistics of the sample are provided in Table 1.

Table 1 Descriptive statistics of dependent and independent variables
Full size table

3.2 Survey experiment

The survey experiment started with a description of the political situation in electoral district 196. The exact wording (translated from German) was as follows:

In electoral district 196, Hans-Georg Maaßen is running for parliament on behalf of the CDU. Maaßen is the former Head of the Federal Office for the Protection of the Constitution. Many of his recent public statements are considered to be right-wing populist by political spectators.

The large left-wing campaign organization ‘Campact” has fielded a campaign in favor of the SPD candidate (Frank Ullrich) in the electoral district, and they ask the Greens and Left Party to support the candidate of the SPD.

How do you evaluate this campaign of Campact for the SPD candidate in electoral district 196?

Respondents could indicate their support on a 10-point scale, where 1 means “very bad” and 10 “very good”. Respondents in the control condition saw the question as described above, i.e. without any additional information. The information treatment extended the description from above by adding the following statement:

The justification for the campaign is that the candidate of the SPD has the highest winning chances against Hans-Georg Maaßen. In the first tier of Germany’s electoral system, only the candidate with the highest number of votes is elected in the electoral district. With this campaign, Campact wants to avoid a situation in the electoral district where votes are split between different left-wing candidates.

Thus, the treatment highlights that the SPD candidate has the highest winning chances in the electoral district. It also highlights that “splitting” votes among left-wing candidates is rather problematic for achieving this goal. Thus, the treatment provides information that explains the rationale for the campaign in more detail. The second treatment introduces a party cue and informs respondents about the support of the Green Party for the campaign:

The Greens support the campaign, and they ask their voters to support the candidate of the SPD with their first vote.

Finally, I also included a treatment condition in which both treatments were combined. The intention behind this combination is that both treatments should have a positive effect on the support for the campaign, and they might reinforce each other. In particular, one might assume that the electoral system information treatment effect is stronger when combined with the party cue treatment. The party cue might not only directly increase a respondent’s evaluation of the strategic voting campaign but also their trust in the provided information about the electoral system and the relevance of strategic voting.

3.3 Estimation

The dependent variable is the level of support for the campaign indicated on the 10-point scale with higher values representing a higher level of support. The main independent variables are the treatment indicators \(D_{\text{Info}}\) and \(D_{\text{Partycue}}\). To account for the combination of both treatments, I estimate the interaction between both treatment indicators. The control condition, in which no treatments were provided, is the reference category. Thus, the estimated OLS regression model takes the form of:

$$\begin{aligned}\displaystyle\text{Support for Campaign}=\alpha+\beta\times D_{\text{Info}}+\gamma\times D_{\text{Partycue}}+\delta\times(D_{\text{Info}}\times D_{\text{Partycue}})+\sum_{j=1}^{J}\lambda_{j}\times X_{j},\end{aligned}$$

where \(X_{j}\) denotes additional control variables that are included in the regression model. These additional variables are not strictly necessary for “controlling on observables” due to the random assignment of the treatments. Instead, they are included to provide further insights on which other characteristics of respondents might affect support for the strategic voting campaign.Footnote 6 Specifically, I include the left–right self-placement of respondents, which was measured from “very left” (1) to “very right” (10). One might expect that more right-wing respondents are less favorable towards the campaign as they might be less inclined to support a campaign from a left-wing organization against a CDU candidate. In addition, these more right-wing voters are probably more indifferent between the SPD and CDU which should also decrease their incentives to cast a strategic vote and, thus, lower their support for the campaign. Second, one might also expect that strong partisans are less inclined to support a campaign in favor of a different party. To account for this, I include a dummy variable that equals 1 if a respondent voted for the Greens in the election of 2017 and zero otherwise. Third, respondents who are favorable of the SPD are potentially more likely to support a candidate from the SPD. Therefore, I also include a variable measuring a respondent’s evaluation of the SPD on a 10-point scale from 1 (very bad) to 10 (very good). Fourth, strategic voting requires a certain level of knowledge about politics and, thus, respondents with high levels of political interest as well as a high level of education might be more likely to understand the purpose of the campaign.Footnote 7 Therefore, I include variables for measuring political interest (5-point scale) and education. Education is a dummy variable and has the value “high” if a person has a high school diploma (“Abitur”) and “low” otherwise. Finally, gender and age are included as socio-demographic factors.

4 Results

Before discussing the effects of the treatments, it is reasonable to look at the overall evaluation of the campaign among respondents (see Fig. 1). In general, the respondents are quite favorable toward the campaign. The average rating is 6.89. Around a quarter of all respondents evaluate the campaign with the highest value (10 – “very good”). In general, two-thirds of the respondents provide a rather favorable rating of the campaign with a value of 6 or higher (67.7%). Only 15% of respondents give a value of less than 5. Thus, the campaign is apparently not seen as particularly critical among most Green voters.

Fig. 1
figure 1

Evaluation of the campaign. Note: the dashed line denotes the average of 6.89 (\(N=988\))

Full size image

Table 2 reports the results of the OLS regression models. The first model only reports the treatment effects without the inclusion of any additional covariates. When interpreting the effects, it has to be kept in mind that the interaction term between both treatments is also included in the model. Thus, the “Information Treatment” coefficient only represents the effect of the “Information Treatment” in the absence of the “Partycue Treatment”. As can be seen, this effect is negative and not statistically significant. This finding suggests that providing additional information about the electoral system and the winning chances of the SPD candidate does not have a meaningful impact on the evaluation of the strategic voting campaign. In contrast, the “Partycue Treatment” does have a positive effect. There is a significant increase of approximately a half scale-point (0.48) in support for the campaign when it is mentioned that the Greens support the strategic voting campaign. Again, this is the effect of the treatment when the other treatment is not displayed simultaneously. When both treatments are displayed to a respondent, the effect sizes increase further by 0.27. However, the interaction effect is not significant. In more substantive turns, this means that the “Information Treatment” effect is positive (0.20) when the “Partycue Treatment” is also displayed, but the “Information Treatment” effect is still not significant. In contrast, the “Partycue Treatment” effect increases to 0.75 when it is displayed in combination with the “Information Treatment” and remains significant. These conditional treatment effects are also displayed in Fig. 2. Figure 2a shows that the “Information Treatment” is insignificant in both treatment conditions. Figure 2b demonstrates that the “Partycue Treatment” is always positive and significantly larger than zero. These estimates are slightly different when additional variables are included in the regression analysis, as displayed in Model 2 of Table 2. The main difference with regard to the significance of the estimates is that the Partycue Treatment is no longer significant in the absence of the Information Treatment (\(p=0.11\)). However, when Information Treatment is also displayed, Partycue Treatment is still significant (\(p<0.001\)).Footnote 8

Table 2 OLS regression results
Full size table
Fig. 2
figure 2

Conditional Treatment effects. Note: horizontal lines are 95% confidence intervals. Effects estimated based on Model 1 in Table 2

Full size image

Model 2 additionally reports the effects of the other included covariates that might explain why respondents evaluate the campaign differently. First, left–right self-placement has a negative effect. The more right-wing a respondent is, the more negative the evaluation of the campaign by the respondents. This finding is very plausible, as it can be expected that more right-wing respondents are less interested in supporting a campaign against a CDU candidate. Moreover, it is possible that more right-wing respondents are also critical of Campact, which describes itself as a clearly left-wing organization and was also described as such in the experiment.

Respondents who voted for the Greens in 2017 show a slightly more negative evaluation of the campaign. Given that the Greens gained a lot of votes in 2021, voting for the Greens in 2017 can be interpreted as an indicator of a rather long-standing party preference for the Greens. These “core voters” of the Greens appear to be a bit more critical of campaigns in favor of other parties. The evaluation of the SPD has a strong positive effect on the evaluation of the campaign. It is not very surprising that respondents who evaluate the SPD more favorably are also less critical of the campaign.

Political interest and education (measured as having the “Abitur”) both show large positive effects. This finding indicates that respondents with high levels of political interest and education evaluate the campaign substantially more positively than respondents with lower levels of political interest and education.Footnote 9 These results suggest that these respondents are more aware of the fact that strategic voting for the SPD is rational if one wants to impede the election of a strongly right-wing candidate. Finally, among the socio-demographic characteristics, gender has no significant effect. In contrast, age shows a significant positive effect.Footnote 10

To put the strength of the different effect sizes in comparisons, I compute first differences for all variables. These first differences denote the differences in the predicted values when an independent variable takes the value of its third and first quantile. Following King et al. (2000), I simulate 5,000 of such predictions for each variable by taking draws of the parameters from a multivariate normal distribution. All simulations are based on Model 2 of Table 2. This approach has two desirable characteristics: (1) it does not rely on extreme changes in the independent variable, and (2) it makes the estimates more comparable. For binary variables, such as gender, the effect is identical to the regression coefficient. However, as the treatment effects are interacted, the first difference approach averages over the interaction effect.

The results are displayed in Fig. 3 and demonstrate that left–right has the overall strongest effect. A change from the first to the third quartile in this variable leads to a one-scale point decrease in campaign support. The second strongest effect is the “Partycue Treatment”, which has an average effect of increasing support for the campaign by 0.56. Age and education are also relatively strong predictors for campaign support. Overall, this comparison of effect strength indicates that the “Partycue Treatment” is quite strong and of larger size than many other variables.

Fig. 3
figure 3

First differences between Q3 and Q1 with 95% confidence intervals

Full size image

5 Conclusion

Party systems are becoming increasingly fragmented. This development poses a challenge for party and voter coordination under FPTP because identifying and rallying behind the candidate with the highest winning chances becomes more difficult. Campaigns, such as the one by Campact during the German General Election 2021, might help to overcome such coordination problems by providing voters with the required information about the winning chances of the candidates and also by creating pressure for the parties to collaborate. However, the reactions to Campact’s campaign demonstrated that such campaigns can be evaluated quite negatively by party elites. Reflecting on these developments and using voters of the Greens as an example, this study demonstrates that (1) support for such a campaign tends to be quite high, and (2) that voters’ evaluation of the campaign increases substantially when the targeted party supports it. In essence, these results indicate that such coordination campaigns of non-partisan organizations will be most effective when they collaborate with the affected parties.

In contrast, no evidence was found that providing more information about the electoral system and the relevance of strategic voting increases support for the campaign. There are several potential explanations for this null finding. One is that the treatment itself might have been too vague or weak. Maybe a more detailed explanation of the electoral system and the relevance of strategic voting might have an effect. Another option is that voters are aware of the relevance of strategic voting but are unwilling to change their opinion on this issue as long as their preferred party does not support the campaign. Finally, the null effect might also be explained by the sample composition. The experiment was conducted with Green voters from all over Germany, and only very few of them, if at all, came from the relevant electoral district 196. As one reviewer has correctly noted, the preferences for casting a strategic vote might be stronger among Green voters from electoral district 196 because their votes really mattered for the election outcome. For the other Green voters the decision is rather hypothetical.

Of course, this study has a number of limitations. One is that it only focused on voters of the Greens. It would also be interesting to analyze how voters of the Left Party evaluated the campaign and how they react to the treatments, especially because the leadership of the Left Party was very critical of the campaign. In addition, the situation in electoral district 196 and the nomination of HGM by the CDU were quite prominently discussed during the election campaign. This could have diminished the effects of the information treatment as voters were well informed about the situation. In less prominent contexts, the treatment might have a different effect. Moreover, in many cases, the CDU/CSU does not nominate such controversial candidates as HGM. This raises the question of whether similar campaigns would enjoy an equally high level of support when the CDU/CSU candidate is less prominent or scandalous. Addressing such questions remains a task for future research.