1 Introduction

Voters both reward and punish politicians. They observe, for example, macroeconomic outcomes, such as economic growth, unemployment and inflation, and hold politicians responsible for that performance: when the economy is doing well (high economic growth, low unemployment and inflation), voters tend to reward politicians and return incumbents in office (on theories and empirical evidence of economic voting see, e.g., Nannestad & Paldam 1994; Lewis-Beck & Stegmaier 2013; Maloney & Pickering 2015).Footnote 1 Clearly, politicians often attempt to influence macroeconomic outcomes by, for example, implementing partisan fiscal policies that gratify the needs of their constituencies (e.g., Schmidt 1996; Potrafke 2017, 2018). Politicians cannot, however, influence macroeconomic outcomes fully. Voters understand that macroeconomic outcomes also are determined by exogeneous events and therefore evaluate politicians’ performances based on measures that politicians influence directly. It is conceivable that voters desire competent, honest and exemplary politicians and punish politicians who have been involved in scandals. The empirical evidence on how voters respond to politicians who had been corrupt, favored relatives or abused taxpayers’ money is quite mixed, however (e.g., Karahan et al. 2006, 2009; Hirano & Snyder 2012; Vivyan et al. 2012; Rudolph & Däubler 2016; Kauder & Potrafke 2015; Larcinese & Sircar 2017; Potrafke & Rösel 2019).

The rollcall votes cast by members of parliament (MPs) supply information on which voters plausibly reward and punish politicians (retrospectively).Footnote 2 Rollcall votes capture the behaviors of individual politicians in parliament; they often are conducted when politicians confront controversial issues. Examples include Brexit and same-sex marriage (e.g., Kauder & Potrafke 2019; Aidt et al. 2021). Political parties have collective views on such issues. Conservative parties are, for example, less likely to advocate same-sex marriage than liberal parties. Individual MPs may, however, disagree with their parties’ positions on controversial issues and consider voting against the party line. On the one hand, voters may punish individual MPs who defect from the party line when they believe that MPs should be loyal. On the other hand, voters may support defectors when they themselves disagree with the party line or wish to reward rebellion.

Empirical evidence on how MPs’ rollcall voting influences citizens’ voting behavior is mixed. In the United States, voters hardly pay attention to congressional rollcall votes (Ansolabehere & Jones 2010). Voters have punished politicians for being too partisan, but not for being ideologically too extreme (Canes-Wrone et al. 2002; Carson et al. 2010). In the United Kingdom, the policy accountability of MPs is both weak and general rather than issue-specific (Vivyan et al. 2012). Voters are likely to reward and punish MPs for their rollcall votes when the electoral system allows individual punishment, the vote received attention in the media, and voters had chances to learn about the votes cast by and the views of individual MPs (e.g., Stiers 2021).

The German electoral system is well suited for examining how voters punish and reward individual MPs because the so-called first vote in national elections is a personal vote.Footnote 3 MPs’ individual votes on same-sex marriage were public knowledge and some MPs were quite active in advertising their voting behavior (e.g., Tauber 2017).

The rollcall vote on same-sex marriage we study was held on June 30, 2017, in the German national parliament (Bundestag).Footnote 4 It was the only rollcall vote that day. That vote had a profound influence on the public debate about the social-political platforms of the conservative parties in Germany, the Christian Democratic Union (CDU) and its Bavarian sister party Christian Social Union (CSU) (see Wenzelburger & Fehrenz 2018).Footnote 5 Same-sex couples already had the opportunity to enter into legally recognized civil unions.Footnote 6 When the votes were counted, 393 members of parliament (MPs) favored and 226 MPs opposed same-sex marriage (four abstained and seven did not attend). The MPs of the social-democratic SPD, the Greens, and the left-wing party DIE LINKE voted nearly unanimously in favor of same-sex marriage; 225 of the CDU/CSU MPs voted against and 75 in favor of same-sex marriage (four abstained and five did not attend).Footnote 7 The voting behavior of the conservative CDU/CSU MPs was remarkable because MPs usually voted in party blocs in rollcall votes that dealt with moral policy issues, such as abortion and stem-cell research. Christian democratic MPs were more likely to vote for restrictive policies than members of other political parties over the 1992–2015 period (Engler & Dümig 2017).Footnote 8

Conservative MPs were concerned about their re-election chances when voting on same-sex marriage: the margin of victory for the incumbent in the previous election was a strong predictor of supporting same-sex marriage (Kauder & Potrafke 2019). When the winning vote margin increased by 1% point, the likelihood of voting in favor of same-sex marriage declined by around 1.32% points. Against the background of the re-election concerns of the conservative MPs in June 2017, we investigate how voting on same-sex marriage influenced vote shares in the national election in September 2017. The results of the econometric model show that, when comparing the outcomes of the 2017 and 2013 national elections, the vote share of conservative politicians who voted in favor of same-sex marriage was around 1.29% points higher (around 0.33 standard deviations of the change in the candidate’s vote share) than the vote share of conservative politicians who voted against same-sex marriage.

2 Empirical strategy

2.1 Institutional background and data

The German party system has changed quite a bit since the 1990s. The platforms of the established political parties CDU/CSU and SPD (Social Democratic Party) converged toward the ideological center. The then SPD-chancellor Gerhard Schröder moved the SPD towards the middle of the political spectrum in the early 2000s. Consequently, new political parties entered the political arena. In 2007, the leftist party DIE LINKE was established. DIE LINKE competes against the SPD to attract left-wing voters. As a result, SPD’s vote share fell in national and state elections. Angela Merkel, CDU chancellor and CDU chairperson, shifted the CDU towards the middle of the political spectrum. In 2013, the populist right-wing party Alternative for Germany (AfD) was founded. Afterward, the vote share of the CDU/CSU declined in national and state elections.

In national elections, voters cast two votes in a personalized proportional representation system. The second vote determines how many seats the individual parties receive in parliament. Each party that wins at least 5% of the second votes obtains a fraction of the 598 seats in parliament corresponding to the party’s second vote share. The first vote determines by simple majority the candidate who obtains a direct mandate in one of Germany’s 299 electoral districts. Candidates elected to the parliament on the first vote (direct mandate) gain their seats first. Candidates from state-specific party lists win the remaining seats. When the number of direct mandates exceeds the party’s vote share, the party acquires excess mandates, and the other parties obtain equalizing mandates to maintain parliamentary power relations.

The second national vote share of the CDU/CSU fell from 41.54% to 2013 to 32.93% in 2017. The first vote share of the CDU/CSU also declined (from 45.34 to 37.26%). We consider only those districts in which the 2013 candidate ran for office again in 2017; we also exclude districts whose boundaries changed between the two national elections. In the resulting sample of 198 CDU/CSU MPs, the first vote share declined from 45.96 to 38.03%. Figure 1 shows that conservative MPs voting in favor of same-sex marriage lost 2.63% points less support in 2017 than MPs voting against same-sex marriage.

Fig. 1
figure 1

Conservative MPs voting in favor of same-sex marriage lost 2.63% points less than conservative MPs voting against same-sex marriage. (Notes: The figure shows the difference in CDU/CSU MPs vote shares between the 2017 and the 2013 national election. The decline of the first vote shares of conservative MPs voting in favor of same-sex marriage (6.02% points) was 2.63% points smaller than the decline of the first vote shares of conservative MPs voting against same-sex marriage (8.65% points). This difference is statistically significant at the 1% level. Vertical lines show the 95% confidence interval)

2.2 Econometric model

The model has the following form:

$$ Change\,CDU/CSU\,first\,vote\,shar{e_i} = \alpha + \beta \,Supporting\,same - sex\,marriag{e_i} + \sum {_j{ \in _j}{X_{ij}} + {u_i}} $$

,

with i = 1, …, 198 (number of observations); j = 1, …, 10 (number of control variables).

The dependent variable Change CDU/CSU first vote share identifies the change in the vote share of CDU/CSU candidate i from the 2013 to the 2017 national election. Our explanatory variable of interest, Supporting same-sex marriage, takes the value one when a candidate voted in favor of same-sex marriage in June 2017 and zero when a candidate voted against same-sex marriage or abstained.

To deal with potential omitted variable bias (OVB) we enter j explanatory variables (in vector X) that are likely to be correlated with the change in the vote share between 2013 and 2017 and voting behavior on same-sex marriage. To control for economic voting, we enter Change in unemployment rate, measuring the difference between the 2017 and the 2013 unemployment rates in the candidate’s constituency. As the governing party, we conjecture that the CDU/CSU lost votes when the unemployment rate increased. It is conceivable that the new populist right-wing party AfD influenced the vote share of CDU/CSU candidates: in districts with many AfD supporters, we are likely to observe fewer votes for the CDU/CSU candidate. We thus include the variable Change AfD second vote share, measuring the change in the AfD’s vote share from 2013 to 2017. The role of the AfD in the German party system is complex. MPs and party members of the CDU/CSU have held different views about how the CDU/CSU should handle the AfD. For example, some believe that the CDU/CSU should denigrate the AfD strongly. Others believe that the CDU/CSU should counter the AfD’s political arguments. It is conceivable that voting on same-sex marriage and the CDU/CSU’s first vote share is related to the views of the individual CDU/CSU candidates and the views of the AfD’s candidate in the electoral district (if one was on the ballot).Footnote 9 We would like to measure the complex interactions between the CDU/CSU, the AfD and their candidates in more detail. No suitable data are available to do so, however. We acknowledge that controlling for the AfD’s second vote share is only a rough measure of its political influence.

We enter many other variables that describe the characteristics of the election district. East Germany has been shown to continue to differ from West Germany (see, for example, Bauernschuster et al. 2012; Potrafke 2013). The dummy variable East Germany assumes the value one if the electoral district is in East Germany, and zero if the district is in West Germany. The variable Change in population density describes whether the change in a candidate’s vote share differs between districts that have become more urban or more rural between 2013 and 2017. We expect a CDU/CSU candidate’s vote share to increase when a district has become more rural. The CDU/CSU’s candidates may have gained or lost votes when the share of young or old inhabitants has changed. The variables Change in share age below 18 and Change in share age above 60 measure how the shares of inhabitants below 18 and above 60 have changed between 2013 and 2017. We expect that CDU/CSU candidates gained votes when the share of the young has fallen and when the share of the elderly has risen. Change in share school low and Change in share school high describe how the shares of students that finished secondary school in the last year without graduating or with higher education entrance qualifications have changed from 2013 to 2017. German students have been shown to be more favorable to Social-Democratic and Green than of Christian-Conservative policy positions (see Fischer et al. 2017). We thus expect CDU/CSU candidates to lose votes when the share of skilled graduates increases.

We finally enter two variables describing the personal characteristics of the candidate: Candidate age in 2017 and a dummy variable Candidate female that assumes the value one for female candidates. Survey data show, for example, that young and female citizens were more likely to support same-sex marriage than old and male citizens (Der Tagesspiegel 2017). Clearly, considering the ten control variables in the econometric model does not solve the issue of potential OVB. (Un)observed confounders may be correlated with voting on same-sex marriage and the change in first vote shares. We estimate an OLS model with standard errors robust to heteroscedasticity (Huber-White-sandwich standard errors—see Huber 1967; White 1980). Table 1 reports descriptive statistics.

3 Results

3.1 Baseline

Table 2 contains our baseline regression results. In discussing them, we focus on our preferred specification in column (4). The estimate indicates that voting in favor of same-sex marriage gave rise to a smaller decline in the CDU/CSU’s first vote share (the vote shares of all 198 MPs in our sample shrank). The coefficient estimate is statistically significant at the 5% level. The numerical meaning of the effect is that candidates who voted in favor of same-sex marriage were better off by about 1.29% points (around 0.33 standard deviations of the change in the candidate’s vote share).

Table 1 Descriptive statistics

The change in the unemployment rate does not turn out to be statistically significant and thus does not support the theory of economic voting. The coefficient on the change in the AfD’s vote share indicates that a 1% point increase in it gave rise to a 0.68% point decline in the CDU/CSU candidate’s vote share (statistically significant at the 1% level). An increase in the share of the population aged above 60 by 1% point (around one standard deviation) gave rise to a 1.30% point increase in the CDU/CSU’s vote share, corroborating that the conservative CDU/CSU is popular among the elderly (statistically significant at the 1% level). An increase in the share of students that finished school with the highest possible graduation credentials reduced the vote share of the CDU/CSU candidate (statistically significant at the 5% level). The numerical meaning of the effect is that an increase in the share of skilled graduates by 1% point (around 0.17 standard deviations) reduced the vote share of the CDU/CSU candidate by about 0.07% points, indicating that German students prefer parties other than the CDU/CSU. The effect is, however, numerically small. The other variables describing the electoral district do not turn out to be statistically significant: whether the district is in East Germany or its changes in population density, in the population share younger than 18 and in the share of students leaving school without graduating do not seem to matter.

From the two variables describing the candidates’ characteristics, only age is statistically significant (at the 5% level), indicating that older candidates lost more votes than younger candidates. The numerical magnitude of the effect is, however, small and indicates that being one year older reduced the candidate’s vote share by about 0.06% points. The coefficient estimate of the candidate’s sex, by contrast, lacks statistical significance at conventional levels.

Table 2 Regression results

3.2 Robustness and heterogeneity tests

3.2.1 Additional control variables

GDP per capita in the electoral districts is available only for the year 2017. We wish to consider the change in economic performance from 2013 to 2017. That is why we relied on the change in unemployment and not GDP per capita in our model. The population shares of migrants likewise are available only for 2017. When we enter GDP per capita and the share of migrants as measured in 2017, inferences regarding the relation between voting on same-sex marriage and changes in the CDU/CSU’s first vote shares do not change.

3.2.2 Jackknife tests

Jackknife tests do not suggest that inferences change when an individual MP is included or excluded. When dropping individual MPs one at a time, the coefficient of same-sex marriage ranges between 0.0106 and 0.0146 and is statistically significant at the 5% or 1% level.

3.2.3 East and West Germany

We examine whether the effect of voting on same-sex marriage differs in East and West Germany. In East Germany, the decline in the first vote shares of conservative MPs voting in favor of same-sex marriage (7.33% points) is 5.46% points smaller than the decline in the first vote shares of conservative MPs voting against same-sex marriage (12.79% points). In West Germany, by contrast, the decline in first vote shares of conservative MPs supporting same-sex marriage (5.69% points) is 2.25% points smaller than the decline in the first vote shares of conservative MPs voting against same-sex marriage (7.94% points). When we enter an interaction term between voting on same-sex marriage and running in East Germany as well as our baseline model’s controls, the marginal effect of voting on same-sex marriage does not differ statistically between East and West Germany.

3.2.4 New candidates in electoral districts

Our baseline model includes only the candidates who ran for election in both 2013 and 2017. In 37 electoral districts, a new candidate was running for the conservatives in 2017 rather than the incumbent conservative. We examine whether voting in favor of same-sex marriage also gave rise to a positive effect on the first vote share in those 37 electoral districts. When replicating columns (1) and (2) of Table 2, the results indicate that new candidates also benefitted from their predecessors’ vote in favor of same-sex marriage (replicating columns (3) and (4) is not suitable because they take account of personal characteristics). A spillover effect thus materializes from the previous MPs’ votes on their successors in the constituency.

4 Conclusions

Voters often retrospectively reward incumbent politicians when the economy was booming (economic voting). Rewarding politicians for economic performance is difficult, however, because politicians do not directly influence macroeconomic outcomes such as growth and employment.

Rollcall votes in parliament are a suitable way of evaluating politicians’ performances. We have examined a controversial rollcall vote in the German Bundestag: voting on same-sex marriage. Same-sex marriage was approved by the German Bundestag on June 30, 2017. That rollcall vote attracted considerable media and public attention. Around 75% of the conservative MPs opposed same-sex marriage, 25% of the conservative MPs favored it. Conservative MPs running in contested electoral districts likely attracted left-wing voters by advocating same-sex marriage (Kauder & Potrafke 2019).

An interesting question is whether voters rewarded conservative MPs for supporting same-sex marriage in the national election on September 24, 2017. They did. We have investigated how voting on same-sex marriage influenced changes in vote shares between the 2013 and 2017 national elections. The (unconditional) difference in vote shares of MPs who voted in favor of same-sex marriage was 2.63% points larger than the difference in vote shares of MPs who voted against same-sex marriage. That difference declines to 1.29% points once we control for many potential confounders, such as the vote share of the populist right-wing party AfD, the unemployment rate and the age of the individual MP. The change in the unemployment rate did not predict changes in conservative MPs’ vote shares, a result indicating that same-sex marriage was more salient to voters than economic issues. In many industrialized countries, economic performance has been strong since the 2008–2010 financial crisis (at least until the Covid pandemic). Voters in 2017 apparently focused on issues other than economic ones to evaluate politicians’ performances in office.