To adapt or to defend? Comparing position shifts among Bundestag candidates between 2013 and 2017

Abstract In this research note, candidate survey data from the German Longitudinal Election Study (GLES) is used to analyse positional shifts of German Bundestag parties between 2013 and 2017. Two developments make Germany a particularly interesting case: (1) the liberal but also controversial policies of the Merkel cabinet during the European refugee crisis and (2) the change of leadership within the right-wing populist party Alternative für Deutschland (AfD). Applying scaling techniques to locate candidates of both elections in the same two-dimensional policy space, the analysis demonstrates that in 2017 the AfD took a distinct radical right position in the party system of Germany. Moreover, the study finds that almost all parties moved to the right on the cultural left–right dimension in 2017, whereas for the economic left–right dimension this has not been the case. Contrary to the mantra of an ideological delineation against right-wing populism, there has been a robust socio-political conservative shift in the German party system.

some have considered it Germany's duty to take care of refugees, others believe that the large share of young Muslim males is a potential threat to public security, especially against the background of sexual assaults in Cologne on New Year's Eve 2015 and terrorist attacks in Paris, Nice, and Berlin (Mushaben 2017: 528-9). Today, there is no doubt that the refugee crisis has had an enormous and lasting impact on the problematical perception of German voters. 1 Most obviously, the rise and ideological positioning of the Alternative f€ ur Deutschland (AfD -Alternative for Germany) needs to be seen in the context of the refugee crisis. While the AfD responded to the crisis with a radical conservative shift, combining anti-refugee and other authoritarian positions (Patton 2017;Schmitt-Beck 2017;Siri 2018), the ideological response of other parties is less clear. On the one hand, there are signs that the refugee crisis and the success of the AfD provoked established parties into defending their ideological worldviews against populist rhetoric. On the other hand, the debate before and after the 2017 election hints at ideological shifts among other major parties. We take this conundrum as a starting point and examine parties' ideological shifts between 2013 and 2017. The main objective is to answer the following three questions: First, is there a robust rightward swing in AfD candidates' ideal points? Second, did the ideal points of candidates from other parties remain stable? Third, does candidate replacement provide a valid mechanism to explain ideological shifts?
In order to provide an answer to these questions, we use data from the Candidate Campaign Survey of the German Longitudinal Election Study (GLES). Nowadays, various approaches for analysing party positions and their shifts exist (Laver 2014). While expert surveys or the coding of party manifestos (Manifesto Research on Political Representation project -MARPOR) are the most prominent approaches, we opt to use candidate survey data as a basis for our analysis. Just like any other method for obtaining party positions, candidate survey data has certain advantages and shortcomings. We consider candidate surveys to be important as candidates are the link between parties and voters (Manow 2015: 107). Candidates are responsible for communicating and promoting their parties' positions to the electorate. However, they have leeway in how they communicate these positions and therefore their ideological positions matter, particularly in cases in which parties remain ambiguous or are internally divided (Eder et al. 2017). Moreover, candidate surveys allow for the analysis of intraparty heterogeneity, which is quite challenging or impossible to obtain by other standard measures of party positions (Steiner and Mader 2017). 2 Our methodological approach expands on previous research that uses candidate survey data to analyse the ideological positions of parties. Jankowski et al. (2017) demonstrate the applicability of scaling methods to candidate data for the case of Germany in the election of 2013. They apply the basic space ('blackbox') scaling technique developed by Poole (1998) to locate candidates in a two-dimensional policy space and to analyse the position of the (back then) newly founded AfD. Carroll and Kubo (2017) use the same methodological approach to highlight how combining candidate surveys allows researchers to locate candidates from different surveys in the same policy space. We adopt Carroll and Kubo's (2017) approach, but in contrast to analysing candidates from different countries, we locate candidates running for parliament in the German federal elections of 2013 and 2017 in the same policy space. This enables us to identify how German parties' ideological positionsbased on the stated preferences of their candidates on 12 different issueshave developed between the two elections.

Candidates' ideological shifts
The year 2015 marks a watershed for the AfD, in terms of both factional infighting and ideological reorientation. In the federal election of 2013, the AfD presented itself primarily as a critic of Chancellor Merkel's political course in the European debt and currency crisis (Arzheimer 2015;Berbuir et al. 2015;Jankowski et al. 2017). The AfD, a Eurosceptic and anti-euro party, led by the economics professor Bernd Lucke, failed by only a slight margin to enter the German Bundestag in 2013. Most studies about the AfD under Lucke come to the conclusion that at that time it was not necessarily a radical right-wing party (Arzheimer 2015;Berbuir et al. 2015;Jankowski et al. 2017). However, research into the candidates of the AfD in 2013 has demonstrated that they s already held strong populist, anti-elite attitudes that distinguished them from established parties (e.g. Ceyhan 2016; Lewandowsky et al. 2016). 3 In the subsequent European and regional elections in 2014, the AfD managed to achieve respectable results, while at the same time the European debt and currency crisis lost both political urgency and public attention. Consequently, the party shifted its focus to immigration issues (Schmitt-Beck 2017: 140).
The ideological rightward shift of the AfD coincides with the 'refugee crisis' which vice federal spokesman for the AfD, Alexander Gauland, referred to as a 'gift' for his party because the societal and cultural conflicts centring around the issue of immigration (Mushaben 2017: 528-9) have been the basis of the AfD's political resurgence in 2015 and paved its way into the German Bundestag in 2017. The ideological rightward shift of the AfD after 2015 denotes a fundamental change in its positioning, shifting from ordo-liberal economic views toward nationalist conservative and anti-migration views (Patton 2017: 165-6). This shift has been identified in manifesto data from the MARPOR project (e.g. Schmidt 2018) and public opinion data (Goerres et al. 2017). Additionally, survey data shows that anti-immigration and anti-Muslim attitudes were salient among the electorate (e.g. Czymara and Schmidt-Catran 2017;Jacobsen et al. 2017;Marx and Naumann, 2018). On these grounds, we also expect to find a shift in the ideal point of AfD candidates.
The second, more puzzling question is how the ideological position of candidates from other parties have shifted between 2013 and 2017. Spatial models and research on democratic representation put forward that parties, MPs and candidates respond to shifts in public opinion, to political events (e.g. economic crises), as well as to the emergence of new competitors with ideological and policy adjustments (Abou-Chadi 2016; Adams and Somer-Topcu 2009; Adams et al. 2004Han 2015). In this respect, it appears plausible that the problem perception of voters and the success of this populist radical right party (see Mudde 2007 for a definition) created a conservative shift among all candidates. Bale et al. (2010) and Han (2015) suggest a pattern of how established parties respond towards emerging populist right-wing parties. After a first phase of ignorance, followed by a phase of political contempt, established parties partially adopt positions of right-wing populist parties in order to convince former voters. A content analysis of newspaper articles by K€ onig (2017) indicates that all parties altered their stance towards refugee immigration in the period between August and November 2015. If candidates for the German Bundestag shifted their ideological position in the context of refugee crisis and immigration topics, this should have taken place on the libertarian-authoritarian, i.e. cultural left-right dimension (Jankowski et al. 2017: 708) rather than the economic left-right dimension. Likewise, Abou-Chadi (2016) argues that mainstream parties are likely to adopt a more anti-immigrant position when facing a successful radical right challenger party. He also argues that this shift should be particularly strong for more mainstream right-wing parties and finds empirical support for these assumptions.
The third question is: what accounts for the position shifts of parties and candidates. One potential explanation is that 'new' candidates hold more radical positions compared to candidates who ran for parliament in the previous election. For the AfD, this assumption is very plausible. In 2013, the AfD was led by Bernd Lucke, who tried to position the AfD as an anti-EU party but not necessarily as a populist radical right party (Arzheimer 2015). With the change in party leadership in 2015 and the exit of Lucke and his sympathisers from the AfD, as well as the new focus on immigration and nationalist issues described above, it can be assumed that the AfD's comparatively moderate candidates had not been re-nominated in 2017 or that they left the party. 4 However, we can also expect that other parties respond strategically to the rise of a populist radical right challenger party (Abou-Chadi 2016; Schumacher and van Kersbergen 2016) by selecting new candidates with a more right-wing position. As we can expect that established candidates and those running again are less likely to change their positions, this implies that candidates who did not run in the previous election should have a more right-wing position on the cultural left-right dimension in the election of 2017.

Data and methods
While many sophisticated measures for estimating party positions existsuch as expert surveys or the MARPOR datasetrecent research promotes the use of candidate survey data as a useful resource for estimating candidate and party positions. As Carroll and Kubo (2017: 1) argue, standard measures of party positions 'exist only at the party level and only loosely reflect the preferences of actual party elites'. Using candidate survey data allows us to overcome these limitations, as they provide insights into the positions and heterogeneity of party elites.
To derive the positions of candidates and parties in 2013 and 2017, we follow the existing work on scaling candidate survey data by using Poole's (1998) basic space ('blackbox') scaling technique (see Armstrong et al. 2014: chapter 3 for a comprehensive introduction). In principle, blackbox scaling is comparable to a factor analysis. However, the technique has some specific advantages over factor and principal component analysis as it allows for the presence of missing data and because it 'analyzes the data matrix directly without any intervening transformations of the original data' (Poole 1998: 954). It is therefore specifically designed for estimating positions of respondents from a set of issue scales (Poole et al. 2016).
We use data from the GLES Candidate Campaign Survey for the federal elections of 2013 (Rattinger et al. 2014) and 2017 . Both surveys are part of the Comparative Candidate Survey (CCS) and include an identical set of 12 policy statements. 5 We combine both datasets and then apply the basic space scaling technique. The Pirate Partywhich was part of the candidate survey in 2013 but not in 2017has been excluded from the analysis due to its decreasing relevance in the German party system. Moreover, the interesting case of the CSU (the 'sister party' of the CDU, which runs only in Bavaria, where the CDU does not run for election) cannot be reliably analysed, as we have only very few observations for both elections for this party (N ¼ 24 for the election of 2013 and N ¼ 14 for the election of 2017). Therefore, we treat the CDU/CSU as one party in the analysis. 6 By combining both datasets, we can locate the candidates from both elections in the same policy space. The positions of the candidates and parties are thus comparable between the two elections. 7 As previous theoretical and empirical work suggests, a two-dimensional model is most appropriate to capture candidates' preferences on the economic (state vs. market) and cultural (authoritarian vs. liberal) left-right dimensions (Jankowski et al. 2017).

Empirical results
Interpreting the policy space Table 1 reports the results of the scaling procedure. The first column reports the ID of the question in the two surveys. The statement column is a translation of the German statement to which the candidates could respond on a five-point scale ranging from 'fully disagree' to 'fully agree'. The column 'N' reports the number of responses for each of these statements. The number of responses varies slightly, as not every respondent answered all questions. As described above, basic space scaling can still locate these respondents if the number of missing responses is not too high. 8 The other columns describe how much variation in the responses to the respective statements can be explained by the first or first and second dimensions. The last column describes the increase in the R 2 values between the two dimensions and thus indicates for which items the second dimension captures important variation. Items are ordered by the goodness-of-fit measure for the first dimension.
Our results confirm findings from previous studies that the first dimension captures candidates' preferences on the economic left-right dimension. This becomes visible by the high R 2 values for items that are closely related to the economy. The second dimension, however, captures items that are more closely linked to the cultural left-right, i.e. liberalauthoritarian, dimension. Items traditionally connected to this dimension (same-sex marriage, immigration, abortion) show strong increases in the degree of variation explained by the second dimension and show the lowest explanatory power for the first dimension.
While the interpretation of the two dimensions as being about the economic and cultural left-right dimension seems to apply, it should be noted that not every item loads as it would be expected from a theoretical perspective. For example, environmental politics and some items that are connected to gender equality load strongly on the first dimension. One could have also expected them to load more strongly on the second dimension. As a consequence, the meaning of the second dimension is a bit narrower than in standard conceptualisations of the cultural left-right dimension, as it mainly captures the issues of immigration, integration, and law and order in this case.

Comparing positions between 2013 and 2017
Having identified this two-dimensional policy space, we now turn to the analysis of positional shifts of candidates between the two elections. The results are depicted visually in Figures 1, 2, and 3.  Table 2, we report statistical tests that compare both distributions for each party. Finally, Figure 3 reports the marginal effects for the election year for each party on both dimensions obtained from an Ordinary Least Squares (OLS) regression in which the party and the election year have been interacted. This regression also included control variables for gender, age, and type of candidacy (list candidacy, district candidacy or candidacy on both tiers). The full results of this regression model can be found in this article's online appendix (see Table A1). 10 As can be seen in the left panel of Figure 1, there have been almost no changes in the positions of the candidates on the economic left-right dimension as the density plots are basically identical for almost all parties. The only clear exception to this pattern are the AfD candidates, who show a higher density at a more right-wing position for the election in 2017. In order to test whether these distributions are statistically different from each other, we conducted a Kolmogorov-Smirnov test (KS test) for equality between the two distributions of each party (see Table 2 for the exact numbers). The null hypothesis that both samples come from the same underlying distribution can clearly be rejected (p < 0.001) but only for AfD candidates. For all other parties, we cannot reject the hypothesis that the candidates held similar positions on the economic left-right dimension between 2013 and 2017. These results are also confirmed by a t-test that compares the average party positions between the two elections. Only the positions of the AfD candidates are significantly different (p < 0.001) between the two elections. The same holds true for the results of the regression analysis (see Figure 3).
A different picture is found for the distribution of positions on the second dimension that captures the cultural left-right dimension. Only the candidates of the Left Party and the Greens show a very similar distribution of their positions. For all other parties, there seems to be at least some evidence that the candidates are now located more to the right of the dimension. Importantly, it is not so much the form of the distribution that has changed between the two elections (for example from a unimodal Notes: The plots displays the average position of candidates on two ideological dimensions estimated using the basic space (blackbox) scaling procedure (Poole 1998). Different shades of grey identify the two elections. Symbols identify candidates from different parties. Figure A1 in the appendix to this article shows the scatterplot for all candidate positions.
to a bimodal distribution). In contrast, the distributions have simply shifted to the right. An exception is the distribution of the CDU/CSU that has become more bimodal in 2017, a pattern that might point to the internal struggle of this party. The change in the distributions is confirmed by the KS test (see Table 2). Only for the Left Party, the null hypothesis of similar distributions cannot be rejected based on the KS test. For all other parties, the KS test rejects the null hypothesis. For the Greens, the p-value is 0.021. In contrast, the t-test and the OLS regression find significant results for all parties. 11 Importantly, however, the shifts are considerably stronger (weaker) for right-wing (left-wing) parties. The AfD shows the largest shift, followed by the FDP and CDU/CSU, while the shifts for the Greens and Left Party are modest. The shift of the SPD is between both these camps.
Another interesting finding is that the point estimate is positive for all parties, indicating that all parties have shifted their position. Under the null hypothesis of no position shift between the two elections, observing shifts to the left or to the right should be equally likely. Therefore, finding position shifts in the same direction for all six parties is very unlikely: 0:5 6 ¼ 0:016: 12 When taking a look at Figure 2, it becomes clear that the landscape in Germany has changed significantly between 2013 and 2017. 13 First, the parties that have traditionally been described as 'right-wing' (CDU/CSU, FDP, and AfD) have moved significantly to the right on the second dimension in particular, while the shifts have been modest for left-wing parties. This is in line with previous studies arguing that right-wing parties respond more strongly to the rise of radical right parties (e.g. Abou-Chadi 2016). The second important finding from Figure 2 is that the AfD has now positioned itself in a unique place in the German policy landscape. While in 2013 the AfD could not directly be distinguished from the CDU/CSU candidates based on the 12 issue statements, the new AfD in 2017 shows a clear national-conservative radical right position. In this regard, it is crucial to mention that the 12 issue items only contain a small number of statements regarding immigration and no item that directly addresses attitudes towards Muslims or refugees. This highlights that the 'new' AfD provides a more fundamental conservative policy platform.
Our results clearly indicate a shift of the AfD to the right on both dimensions and a shift to the right on the cultural left-right dimension for the majority of the other parties. Our final question is, however, whether candidate replacement accounts for these position shifts. This assumption is supported by the fact that the AfD in 2017 shows the lowest proportion of candidates in the sample that claim to have been running for parliament in the previous election. Only 16.3% of the AfD candidates stated that they also ran for parliament in 2013. The high share of newly selected candidates for the AfD does not come as a surprise given the turnover in the party leadership between the two elections. It can be assumed that those AfD candidates who claim to have also run for parliament in 2013 were part of the radical faction within the AfD that later took over the party leadership. The share in the survey of candidates running again is higher for the other parties: approximately 25% for  the FDP candidates and Greens, 31% for the Left Party, 44% for the SPD, and 49% for the CDU/CSU. While these shares are higher compared to the AfD, they are still quite low and might indicate that the response rates of the candidate survey is lower among more established candidates. 14 In Figure 4, we display the results of our analysis in which we compared the average positions of first-time candidates and those running again in the 2017 election. We display the predicted position on the first and the second dimension on the y-axes. The x-axes denote the candidate's party. First-time candidates and candidates running again are distinguished by different shades of grey and symbols. As can be seen, there is no difference between candidates running for the first time and candidates running again for parliament as the positions are very similar for both groups. The only statistically significant difference is between firsttime and re-running candidates of the SPD on the first dimension. This difference is very likely the result of chance, given that the remaining 11 comparisons are not significantly different.
These results contrast with our assumption. We expected that candidates running again would be less likely to change their position as their political attitudes should resemble the sample of candidates running in 2013. Candidates running again are more established 15 and they should be less likely to change their attitudes during their political career. We also assumed that the shift in the average positions might be caused by new candidates with more conservative attitudes entering the electoral Notes: The plots display the predicted position of candidates from the 2017 election on two ideological dimensions estimated using the basic space ('blackbox') scaling procedure (Poole 1998). Grey points identify estimates for candidates that were also running for parliament in 2013.
arena. This does not seem to be the case. Instead, this obvious null finding on what explains position shifts suggests that these are a result of a change in candidates' attitudes.

Conclusion
At a time when many EU countries were following a restrictive migration policy, Germany's so-called 'open border policy' received much attention and praise but also criticism at both the international and national levels. Particularly, the response to the refugee crisis has been one of the most important accelerators for the increasing success of the populist radical right party AfDa party that underwent serious internal struggles, but still managed to enter all state parliaments and in 2017 also the German Bundestag as the largest opposition party.
In this research note, we have analysed how the position of candidates in Germany shifted between the elections of 2013 and 2017, of which the latter was heavily dominated by debates on the European refugee crisis. The empirical results can be summarised in three points. First, the finding reported in Jankowski et al. (2017) that the AfD candidates of 2013 did not take a unique position in the political space raised the question of whether this result indeed reflects a moderate ideological position of the AfD in 2013 or whether scaling approaches are not able to reveal the 'true' ideological position of the AfD. In line with our expectations, we find that the AfD of 2017 now indeed takes a much more radical position compared to 2013, one that is clearly distinguishable from the CDU/CSU and all other parties represented in parliament. This result strongly suggests that the findings from 2013 are not caused by a methodological shortcoming, but that the AfD candidates in 2013 were indeed quite similar to those of the CDU/CSU. In this regard, our article contributes to the ongoing research about the development and position shifts of the AfD. It is also noteworthy that the policy space identified by the scaling procedure remained very stable between the elections. While the first dimension captures various issues on a general left-right dimension, the second dimension is quite narrow and captures issues about migration and law and order. This contrasts with broader definitions of the cultural left-right dimension. Comparative studies could therefore investigate whether such a narrow definition of the second dimension is found in other countries as well.
Second, we observe an overall shift to the right on the cultural leftright dimension by almost all parties in 2017. On average, not a single party occupies a more leftist position on this dimension compared to their average position in the election of 2013. The shift to the right is especially strong for right-wing parties, while it is only modest for those on the left. As could be expected, the Social Democrats clearly showed the largest shift among the left-wing parties. Importantly, we do not find this pattern for the economic left-right dimension. In sum, these findings suggest that candidates systematically moved to a more conservative position in 2017, as a response to the controversial liberal migration policy of the Merkel cabinet in 2015 and the simultaneous rise of the AfD.
Third, the shift cannot be explained by candidate replacement because candidates running for the first time in 2017 and those who already ran in 2013 do not differ. This finding supports the assumption of a true ideological shift. However, it should be noted that this analysis would have benefited from using panel data, which is not available as the candidate survey is cross-sectional and does not allow the tracking of individual candidates between the elections.
Besides the candidate survey data utilised in this research note, there are many alternative party position measures that scholars employ to study party positons, including manifesto codings (e.g. Franzmann and Kaiser 2006), expert surveys of party positions, party placements by rankand-file voters in national election surveys (also see Benoit et al. 2016;Volkens 2007), as well as content analyses of newspaper articles (e.g. K€ onig 2017). In this respect, it becomes a question of particular relevance whether the shifts in German candidates' expressed beliefs across 2013-2017 are also reflected in shifts in the policy rhetoric the parties published in their manifestos. For example, Schmidt (2018) used data from the MARPOR project and applied the method developed in Franzmann and Kaiser (2006) to compare position shifts among German parties between 2013 and 2017. The analysis also finds that all parties, except for the Left Party, have moved to the right on the left-right axis. 16 Moreover, the shifts are particularly strong for parties on the right, while the shift was moderate for parties from the left of the spectrum. Thus, German candidates' expressed beliefs across 2013 and 2017 appear to be reflected in shifts in the policy rhetoric the parties published in their manifestos. Although we cannot answer the causal relationship between candidates' ideological shift and shifts in parties' manifesto positions, there is consistent evidence that candidates' ideological shifts are reflected in the parties' officially stated policy positions in their election manifestos. At least we can presume a mutual influence between candidates' beliefs, the parties' official policy positions as enunciated in their manifestos, press statements, the party leaders' speeches, and the policies these parties actually enact in government. However, more research is definitely desirable on how party positions obtained from different sources are connected, how these measures reflect different aspects of party positions, and how empirical results depend upon which measure is being used.
In methodological terms, our research note is in line with the findings in Carroll and Kubo (2017), as it highlights the opportunities for scaling respondents from different elections into the same policy space. The robust results for 2013 and the meaningful shift in 2017 suggest that candidate survey data is a useful resource for party and candidate research. It does not only allow us to locate parties in a two-dimensional ideological space and track positional changes, but also helps to depict intra-party heterogeneity.
However, a final caveat must be mentioned: we cannot identify whether candidates of other parties moved to the right due to the rise of the AfD as a competitor, because they aimed to capture the current mood in public opinion, or whether this ideological shift had other reasons. Furthermore, the GLES candidate surveys deploy a cross-sectional design which does not allow for inferences on individual shifts. Nevertheless, we can say that one of the common mantras towards the emergence of rightwing partiesdo not try to imitate the new kidappears to be untrue for the case of Germany as all parties shifted to the right.

Data sharing
Stata and R scripts are available for replication purposes. Notes 1. Online appendix Figure A4 demonstrates how issue salience has shifted dramatically between the two elections based on monthly survey data about the 'most important problem' in Germany (Forschungsgruppe Wahlen 2018). 2. Another challenge with the MAPOR data is that under certain electoral conditions, parties may prefer to present ambiguous statements in their manifestos for strategic reasons (see Lo et al. 2016). The candidate survey, on the other hand, asks participants to comment on a number of concrete policy issues. 3. Various other papers have analysed the positions of the AfD and its voters.
Recently, for example, Bieber et al. (2018) highlighted that the AfD has gained support among voters who identify clearly as right-wing. Jankowski and Lewandowsky (2018) analyse the positions of the AfD's Members of the European Parliament and find evidence for a position change after Lucke and his supporters left the party. 4. The Liberal Conservative Reformers (previously ALFA), founded and led by Bernd Lucke after the AfD split in 2015, has been practically irrelevant. 5. The 2017 wave of the survey includes two additional statements at the end of the item battery. We do not use these items in the analysis. Please also note that some items changed between the GLES candidate surveys 2009 and 2013, which is why we restrict our sample to the survey waves of 2013 and 2017.
6. In the online appendix to this paper, we provide the positions of the CDU and CSU candidates separately. For both parties, the results indicate a shift on the cultural dimension of similar magnitude. 7. Only when the data is combined and estimated jointly are the positions are comparable. The positions are not comparable when estimated in two separate models. However, in order to control whether both elections were structured by a similar policy space, we estimated the models separately for each election and plotted the correlations between the positions. The results suggest a very strong correlation between the positions. The results are displayed in the online appendix in Figure A3. 8. We decided that a candidate must have answered at least 10 of the 12 statements to be included in the analysis. 9. For example, from an unimodal to a bimodal distribution. 10. We ran the regression analysis with and without control variables and also with and without using the candidate weights provided in the survey. The results are not affected by these different model specifications. 11. For the Left Party the p-value is < 0.1 and for the Greens < 0.05. 12. The level of significance for the individual point estimates is not relevant for this calculation. See Lo (2013) for a comparable example. 13. In this paper's online appendix, Figure A1 displays the raw positions for all candidates. 14. Unfortunately, we have no access to the actual rates of those running again based on all candidates, as the German Federal Election Office (Bundeswahlleiter) does not publish the list of all candidates running in an election due to privacy laws. 15. For example, candidates running again claimed that they had a higher chance of being elected. One question in the survey asks the candidates how they considered their chances of being elected (on a five-point scale with 1 ¼ 'very unlikely' and 5 ¼ 'very likely'). The average value for re-running candidates on this scale is 3.7, while it is only 2.2 for first-time candidates. 16. Note that the data does not allow us to separate between an economic and cultural left-right dimension.

Disclosure statement
No potential conflict of interest was reported by the authors.