Loyal lists, distinctive districts: how dissent-shirking and leisure-shirking affect mixed-candidate selection

Abstract In parliamentary systems, party leaderships accomplish party unity by promising positive incentives to their members of parliament (MP) and threatening them with negative incentives. Regarding reselection, ‘loyal’ and active MPs should be rewarded with promising list positions, while rebellious and ‘slacking’ MPs should be punished with worse list positions or even denied renomination. Although this assumption is central to explaining party unity induced by party discipline, empirical evidence for this practice has been scarce. The study combines data on roll-call voting, parliamentary speeches and questions in the German Bundestag from 1990 to 2017 with novel data on renominations including renomination failures. Investigating list position changes and renomination failures in the German mixed-member system, the results indicate that selectorates of list candidates punished dissenting voting behaviour, while selectorates of district candidates considered the degree of parliamentary activity. The findings have broader implications for the representatives’ accountability in parliamentary systems.

was a small surplus, however, in a straw poll of the CDU/CSU parliamentary group eleven MPs voted against the party line and two abstained. The general secretary of the FDP, Christian Linder, affirmed party unity and only expected four dissenting or abstinent votes in the parliamentary party group (Zeit-Online 2011). Then, in the actual RCV in the Bundestag, 26 members of the coalition government (10 CDU, 6 CSU, 10 FDP) (Wimmel 2014) voted against their government's bill and thereby withdrew the majority from the government. The Chancellor Angela Merkel and her cabinet could not maintain their majority to get the bill enacted on their own and needed the support of the opposition. Finally, the bill was enacted, but the political aftershock for the dissenters did not stay away for long. While the rebelling incumbents of the CDU and CSU were all elected in a single-member district, and only dependent on their sub-regional party branch in their renomination, dissenters of the FDP who entered the Bundestag via the party ticket were confronted with rival candidates at the regional delegation conferences for list nomination. Some of the rebellious MPs failed to regain nomination for the party list at all, others received a worse list position (Wimmel 2014). The politicians attributed their poor performance in the list selection to their dissenting voting behaviour. For example, Nicole Bracht-Bendt, FDP Lower Saxony, was on list position 6 in the 2009 federal election, applied for list position 5, but was challenged 'completely against all agreements' (Handelsblatt 2013) at the regional party list convention and was not selected for this position. After the defeat, she did not apply for lower list positions. Three months before, she had been reselected as a candidate in the single-member district by her sub-regional party organisation, just like other prominent dissenters-e.g. Wolfgang Bosbach (CDU), who not only voted against the party line in the ESM vote but was among the top 5 dissenters of his party in that electoral period and generally known for his non-conforming stances. Bosbach won his district race and remained MP while Bracht-Bendt did not succeed and moved back to local politics.
This anecdotal example of MPs violating party unity and being deselected for the next election poses the question of whether an MP's loyalty and more generally, parliamentary behaviour, plays a systematic role for their renomination prospects. Moreover, the example shows that there might be differences by the type of the selection process. To enhance our understanding of the underlying mechanisms creating high levels of party unity in parliamentary democracies (Sieberer 2006), we concentrate in this article on the question of what type of parliamentary behaviour is actually sanctioned in candidate selection in single-member districts (SMD) and closed-list proportional representation systems (CLPR). Since parties do not only have to overcome the collective action problem of party unity but also of perceived party vitality, i.e. creating a public image of an active and assertive agency, MPs' degree of parliamentary activity may influence the party selectorate's decision on promotion and demotion as well. Therefore, we consider both loyalty in parliamentary votes as well as the activity level of MPs in parliament. Using data on various parliamentary activities (roll-call votes, speeches, parliamentary questions, and parliamentary interpellations) in seven electoral periods of the German Bundestag from 1990 to 2017, our study sheds light on the question, whether parties consider parliamentary (in)activity ('leisure-shirking') and/or (dis)loyal voting behaviour ('dissent-shirking') in candidate selection in parliamentary democracies and how the consideration of different parliamentary activities varies between the nomination processes in the two tiers of the German mixed-member system.
We chose the case of the German Bundestag, because the German mixed-member system provides us the opportunity to investigate whether the regional list selectorates take different kinds of parliamentary behaviour into account than the even more decentralised sub-regional district selectorates. We find that despite low variance in party loyalty, list selectorates systematically punish rebellious MPs and reward loyal MPs in candidate selection, while attendance rate in RCVs and the number of parliamentary questions and interpellations not affect reselection on the list. However, in the more decentral district selection process, we do not find effects of party loyalty in votes but of the number of parliamentary interpellations, which can be explained by different motivations of selectorates. The number of held speeches shows significant effects on both selection processes. This latter finding may hint at the attribute of the number of speeches being connected to both, positional party loyalty and visibility, since access to the speaker floor is gate-kept by the PPG leadership and, as we show, dependent on party loyalty in RCVs.
The remainder of the article is structured as follows. In the second part, we present the state of the literature on the effect of parliamentary behaviour on candidate reselection and describe the ways in which our study contributes to the research literature. We then derive hypotheses on relationships between the MP's parliamentary behaviour and renomination prospects. In the third part of the article, we describe the empirical case, candidate selection in Germany, and the data we use. In the fourth part, we present our empirical findings. In the fifth part, we discuss our findings in the light of previous studies. In the sixth part, we summarise our contribution and consider future research avenues.

State of the art: leisure-shirking vs. dissent-shirking as deselection criteria
From the perspective of the democratic delegation chain on representation in parliamentary democracies, political parties are principals who delegate the task to represent the party in parliament to MPs, who act as the party's agents (Müller 2000). In order to pursue party goals in terms of policy, votes and office (Strøm 1990) most efficiently, the party leadership is expected to reselect MPs that share the same preferences as the leadership and/or have proven to be loyal and hard-working agents, but deselect 'lost' agents. The problems of agency loss can be distinguished into leisure-shirking and dissent-shirking (Müller 2000: 320-322). Leisure-shirking reflects the problem that MPs do not contribute to the party's goals as much as they could. Dissent-shirking means that MPs have a different policy preference than the party's preferred position and undermine party unity. The puzzle that remains is which kind of agency loss plays a more pivotal role for the reselection of incumbents.
Building on the principal-agent theory's concepts of leisure-and dissent-shirking, several studies have investigated the effect of parliamentary behaviour on the reselection of incumbents in parliamentary systems. Previous research on the link between parliamentary behaviour and reselection can be distinguished by two main dimensions: the type of shirking behaviour of MPs and the type of electoral system which renders the centrality of the candidate selection (Hazan and Rahaṭ 2010).
Leisure-shirking is the predominant concept investigated in the empirical literature on effects of parliamentary behaviour on candidate selection. The underlying argument is that MPs who contribute less to a party's policy-making through their parliamentary work are punished by the party leadership 1 with a worse list position or complete deselection. Studies found evidence for an effect of parliamentary activity on renomination prospects in closed-list proportional representation (CLRP) systems with strong centralisation of the candidate selection (Borghetto and Lisi 2018;Louwerse and van Vonno 2022;Marangoni and Russo 2018;Wilson et al. 2016;Yildirim et al. 2019) and in flexible list proportional representation (FLPR) systems (Däubler et al. 2018;Put et al. 2022). However, the level of activity appears to have no effect on the incumbent's renomination prospects in the Hungarian first-past-the-post (FPTP) system with single-member districts and a decentralised candidate selection (Papp 2019). For the effect of MP's level of parliamentary activity on list positioning, studies yield mixed findings. Some scholars find a positive effect of the level of activity on better list positions in CLPR systems (Frech 2016;Hermansen 2018;Louwerse and van Vonno 2022;Papp 2019;Yildirim et al. 2019), while others find no such effect in CLPR (Borghetto and Lisi 2018) and in FLPR systems (Däubler et al. 2018;Put et al. 2022 only for oral plenary questions).
Literature on dissent-shirking shows mixed results for the effect of deviating legislative behaviour on reselection prospects. Here the argument is that MPs violating party unity are deselected or punished with a worse list position. For elections to the European Parliament, where the national party leadership is the main selectorate, but organisationally detached from the PPG (leadership) (Hermansen 2018), scholars find that loyalty to the national party in roll call votes proves to be uninfluential on the list positioning of MEPs in CLPR systems (Frech 2016;Wilson et al. 2016). Baumann et al. (2017) find the expected effect of dissenting behaviour in parliamentary speeches on MP's list position in the German CLPR system only for opposition MPs (Baumann et al. 2017). However, there is no effect of dissenting behaviour in parliamentary speeches on the presence of competitors in renomination in single-member districts (Baumann et al. 2017).
Our study contributes to our understanding of how parliamentary behaviour affects decisions of selectorates in several ways. First, in contrast to previous studies, we control why names of the MPs do not appear on the list again. One reason for their disappearance could be that MPs voluntarily do not seek re-election, for instance to pursue a different (political) career or to retire. These MPs are also more likely to be less active and less loyal in parliament (Bailer and Ohmura 2018;Snyder and Ting 2003). This could confound their covered relationship between parliamentary behaviour and renomination prospects. We address this problem in this study by an extensive investigation of every single MP's reason for not being listed again as a candidate on the party list. This allows us to include nomination failures and exclude retirement in our explanandum. Second, almost all previous studies have concentrated on either leisure-shirking or dissent-shirking on reselection. To our knowledge, there is only one study that analyses dissent-shirking and leisure-shirking at the same time. Galasso and Nannicini (2017) find in Italy that higher party loyalty in RCVs was rewarded with a safer list position and MPs' degree of parliamentary activity has no effect on their reselection. However, like the other previous studies, these findings do not give a clear answer to our puzzle because it does not account for last-term effects. Third, by analysing nomination failures in the CLPR tier and the FPTP tier of the German case, we can approach the effect of varying institutional settings on the link between parliamentary behaviour and reselection.

Reasons for investigating both leisure-shirking and dissent-shirking
We argue that in order to gain more comprehensive knowledge about the effect of parliamentary behaviour on candidate selection, it is useful to investigate both leisure-shirking and dissent-shirking.
There are several reasons to be sceptical about an influence of parliamentary activity on decisions by the selectorate. For one, it is unclear which exact type of parliamentary activity is actually signalling MP's effort in the legislative arena to the selectorate. It is argued that speech-making is more visible to the selectorate and therefore has a stronger positive impact on list promotion (Louwerse and van Vonno 2022). However, speech-making in parliament is often institutionalised in a way that allows the party leadership to restrict who speaks about what issue. The PPG leadership can assign specific MPs to hold a speech on the floor due to strategic reasons, like generating more visibility for frontbenchers (Giannetti and Pedrazzani 2016;Proksch and Slapin 2012). This holds especially true for the German Bundestag where ideologically close MPs are more likely to be admitted to the floor (Proksch and Slapin 2012). Thus, the problem of endogeneity can be countered to a certain degree by controlling for the level of dissent-shirking in roll-call votes. Second, previous empirical findings suggest that not only the amount of floor time might affect reselection, but also further attributes of the speech like topic and visibility (Louwerse and van Vonno 2022;Yildirim et al. 2019) as well as the policy position a MP takes on the floor (Baumann et al. 2017). Moreover, the level of parliamentary activity and the type of parliamentary work varies over the career stages of MPs (Bailer and Ohmura 2018). During their legislative career MPs learn to manage their scarce sources and spend effort differently to parliamentary work. While leisure-shirking can be a serious problem for political parties (Müller 2000: 320-1), little is known about how party leaderships deal with varying levels of parliamentary activity, for instance in different career stages of MPs.
In contrast, dissent-shirking appears to have far more severe consequences for political parties. Citizens vote for (candidates of) a certain political party to implement specific policies, which is why parties try to appear uniform in their policy agenda in the legislative arena. Therefore, political parties try to vote as unanimously as possible on legislative bills. The most costly form of non-uniform appearance is dissenting voting behaviour of MPs. For this reason, the party leadership should monitor particularly deviant behaviour in RCVs. The main criticism against investigating dissent-shirking through deviating behaviour in RCVs is that in parliamentary systems party unity in parliamentary votes is very high and thus not suited to explain punishment or promotion in candidate selection (Baumann et al. 2017: 984;Louwerse and van Vonno 2022: 5;Yildirim et al. 2019: 795). Notwithstanding, our study takes into account that dissent-shirking may play an important role in the reselection of MPs, as suggested by the principal-agent theory literature on party unity. As is usual for parliamentary democracies, party unity in the German Bundestag is very high (Sieberer 2006). Nevertheless, there is some variation in the degree of party loyalty in RCVs among MPs with up to four percent of all individual votes against the party line in some PPGs (Sieberer et al. : 1142. Since defection from the party line in a RCV is the most visible and most harmful form of expressing dissent (Proksch and Slapin 2015: 26), we expect party leaderships to consider indeed especially behaviour in RCVs when it comes to the decision over candidate reselection, as has been suggested by our introductory example.

Dissent-shirking
Party unity is crucial to pursue the party's goals effectively. Party unity refers to the 'observable degree to which members of a group act in unison' (Sieberer 2006). It can result without further coordination from cohesion (shared serious preferences) or in an organised way from party discipline (strategic incentives). Ex-post controls, like (denied) reselection, and their anticipation induce party discipline, because elected MPs fear to be sanctioned by deselection for disrupting party unity or hope to be rewarded by promotion to higher offices for loyalty to the party line. If party unity is not given by preference homogeneity (cohesion), ex-post control via candidate selection must be exerted (discipline). We thus expect: H1a: The more loyal an MP is in parliamentary votes, the less likely she will be deselected from the party list and the less likely she will receive a worse list position in the next election.
H1b: The more loyal an MP is in parliamentary votes, the more likely she will receive a better list position in the next election.

Leisure-shirking
Parties in parliament are based on a strong division of labour and policy specialisation. Therefore, parties depend on their MPs to do their part in fulfilling the party tasks, such as representing the party position in plenary debates, voting on legislative bills, or asking questions to the government parties or the coalition partner. MPs who do not contribute to these tasks jeopardise the achievement of the party's collective goals, and therefore will be punished by the party leadership in the next candidate selection. In contrast, MPs who stand out for an excessive amount of parliamentary activity should be rewarded with a better list position (Louwerse and van Vonno 2022: 5-6).
H2a: The more active an MP is in her parliamentary work, the less likely she will be deselected from the party list and the less likely she will receive a worse list position.
H2b: The more active an MP is in her parliamentary work, the more likely she will receive a better list position in the next election.

Difference between types of candidate selection
According to neo-institutionalist approaches to candidate selection and parliamentary behaviour, institutions shape the preferences of the selectorate and representatives. Regarding preferences of MPs, scholars have formulated that 'where nominations are controlled centrally, we might expect to find that deputies follow the party line faithfully in parliament, as disloyalty will mean deselection' (Gallagher and Marsh 1988: 15). Along the same lines, Benedetto and Hix (2007: 762) also expect that '[t]he more decentralized candidate selection and/or deselection is, the more likely that rebels will be protected from sanctions from their leadership' , however they do not investigate this claim empirically. These expectations are backed by the finding that deviating position-taking in parliamentary speeches influences the German Bundestag PR-tier nomination process but not the more decentral FPTP-tier nomination process (Baumann et al. 2017). Thus, regarding dissent-shirking in RCVs, we expect that MPs' loyalty especially pays off in in the more central CLPR-tier nominations rather than in the more decentral sub-regional FPTP-tier nominations.
Another line of argumentation regarding varying impact of parliamentary behaviour on renomination between the two tiers is based on district magnitude rather than centrality of the selectorate. 2 For SMDs, both central and decentral selectorates may favour MPs with a personal reputation through higher activity in visibility increasing parliamentary behaviour, because in SMDs the personal characteristics of the candidate are more important for winning election (Carey and Shugart 1995). In contrast, in CLPR-systems the visibility of candidates may be less important for both centralised and decentralised selectorates, because voters cannot vote for individual candidates on the party list. Moreover, MPs can increase their visibility and name recognition in the general population by pursuing 'personal vote' strategies in RCVs (Cain et al. 1987;Carey and Shugart 1995), which is why a certain degree of dissent in roll-call voting behaviour might be even embraced by more decentral selectorates. Thus, we expect that parliamentary behaviour that increases an MP's visibility, like holding speeches and engaging in parliamentary questions and interpellations, play a bigger role in the more decentral selection process in SMDs than in the more central regional nomination procedure for the CLPR tier.
The German mixed-electoral system with its varying levels of candidate selection centralisation represents the typical combination of electoral institutions and centrality of candidate selection, as SMD selection is decentralised and CLPR selection is more centralised (Shomer 2014: 539). Thus, because district magnitude co-varies with the level of centralisation of candidate selection, we cannot disentangle the effect of centralisation of candidate selection procedures from the effects of electoral institutions (district magnitude) on candidate selection outcomes in our single country study. Nevertheless, as both proposed mechanisms lead to the same expectation, we derive the following hypotheses: H3a: Loyalty to the party line is more beneficial in renomination for closed party lists than in renomination in the single-member district.
H3b: Parliamentary activities that increase a legislator's visibility are more beneficial in renomination in the single-member district than in renomination for closed party lists.

Candidate selection in Germany
We chose Germany as an unlikely case for disciplining and rewarding through candidate selection because of its territorially decentralised and relatively inclusive candidate selection process (Itzkovitch-Malka and Hazan 2017). The electoral system is a mixed-member system with two different party selectorates in the candidate selection process. About half of the MPs are elected via single-member districts (SMD) in the First-Past-The-Post (FPTP) tier and the rest of the MPs is elected via 16 Länder (regions) closed lists in multi-member districts (MMD) with different district magnitudes in the closed-list proportional representation (CLPR) tier. In a first step, the candidates for the FPTP tier are selected and in a second step, the candidate selection for the CLPR tier takes place. The national party leadership is formally neither part of the CLPRnor the FPTP-candidate selection process.
In the FPTP tier, the Kreis (county) and local party leadership in the single-member district makes a proposal for the candidacy. Other aspirants may also stand for nomination, but this rarely leads to the overthrow of the incumbent and therefore rarely occurs if the incumbent seeks renomination (Baumann et al. 2017: 986-7;Schüttemeyer 2002: 151). In all political parties, either party delegates or party members decide on the proposed nominees on a sub-regional level (Hellmann and Höhne 2020: 9-11). The national party leadership has no formal role in the candidate selection in the FPTP tier and there are no known cases of informal interventions of the national party leadership (Schüttemeyer 2002: 151). However, the Land party leadership has the legal right to raise an objection against a nominated candidate, which requires the candidate selection to be repeated (Hellmann and Höhne 2020) but this rarely occurs.
In the CLPR tier, the Land party leaderships mostly make the proposals for the list rank order, but in a few Land party branches, actors from lower levels of party organisation are involved additionally. Informal rules-and in some cases also formal rules (Hellmann and Höhne 2020)lead to advantages for candidates who have already been nominated as FPTP-candidates by a sub-regional party branch in its constituency. They thus oftentimes take the top positions of a list. This is a solution to the demand for regional quotas, which are known to be very important in the list making process (Reiser 2019). Other existing formal quotas, like gender quotas as well as non-formal quotas like party faction quotas are obstacles for list manipulation by the party leadership, too. Moreover, the list proposal has to be finally decided by a general meeting (list convention) of potentially all Land party members or a delegates' meeting on the Land level and can be changed at will (Hellmann and Höhne 2020).
Looking closely at the list candidate selection process reveals severe obstacles for party leadership control (Hazan and Rahaṭ 2010). In terms of territorial centralisation, the national party leadership has formally no influence on the candidate selection in the CLPR and FPTP tier, since all decisions are taken either at the regional Land level or sub-regional district level. In terms of inclusiveness, even the regional Land party leadership's control over the final list rank order is heavily constrained. Although the Land party leadership has control over the first proposal, the final list rank order is determined by formal and informal institutional arrangements, most importantly party statutes prescribing the Land party list convention as final decision maker, but also quotas. If we nevertheless can observe an influence of parliamentary behaviour on list position change, we can be quite confident that these results can be generalised to parliamentary democracies with territorially more centralised nomination processes, less inclusive selectorates and a pure PR electoral system, since these institutional arrangements enhance concentration of power in the hands of party leaderships.

Data
Our dataset consists of pooled observations from seven full electoral periods. Our units of analysis are individual MPs and our observation period starts at the point of the first Bundestag elections after the German Reunification in 1990 and ends with the Bundestag elections in 2017.

Dependent variables
We use four different binary variables as dependent variables. The dependent variables PR-tier nomination failure and FPTP-tier nomination failure indicate whether the MP in question failed to be nominated by the selectorate in the PR-tier and in the FPTP-tier. For the PR-tier nomination failure variable, we considered nomination failures as MPs who applied for a list position or to be the SMD candidate, but the selectorate decided not to nominate them (for the demanded list position or as SMD candidate) and thus they do not appear on the ballot for the next election. If an MP was a dual candidate in the last election and failed in the renomination in the FPTP-tier, we consider the MP as an anticipated nomination failure in the PR-tier because candidacy in SMD is often crucial to be renominated on a promising list position (Hellmann and Höhne 2020). These nomination failures are essential for our analysis of decisions taken by the selectorate because MPs who are completely deselected from the party compose a critical part of the variation of our explanandum. This information cannot be obtained from official records, because these former MPs do not appear in the candidate register of the subsequent election. The information could only be gathered by extensive research from reports about Land and district party conventions in local and regional newspapers (for a detailed description of the data collection see online Appendix B).
Regarding the other two dependent variables, we measure the MP's change in her list position in the CLPR tier derived from the Federal Returning Office and the BTVote dataset . Our dependent variable measures changes in the list position of an MP of party P in a state S between two subsequent elections t and t + 1. Our measurement strategy differs in two aspects from previous studies. First, whereas previous studies of mere renomination only consider the MPs whose names reappear on a party list, we also include nomination failures. At the same time, we exclude cases of voluntary departure, e.g. retirements, from our sample. Second, since the selectorate is the central actor in our theoretical perspective, we consider changes of absolute list positions. We choose 'change in the absolute list position' as basic measure, because it reflects the decision taken by the selectorate most accurately and independent from electoral results. In contrast, a change in the relative list position, e.g. relative to the number of won seats (Baumann et al. 2017), is not only contingent upon the selectorate's decision but also upon voters' decisions.

Key independent variables: Individual parliamentary behaviour
We use several indicators of individual parliamentary behaviour to investigate the effect of leisure-shirking and dissent-shirking on the probability to fail nomination, to receive a worse list position and to receive a better list position. In order to avoid endogeneity problems, all behavioural indicators are measured only up to one year before the election date. At that time, candidate selection is usually done by the regional and sub-regional party conventions.
We use the BTVote dataset ) as well as data from official records for the 18th legislative period to measure each MP's degree of party loyalty in roll call votes. It is measured by the share of votes in which the MP has voted with the party line compared to the number of all votes she participated in. We thus end up with an agreement rate ranging theoretically from 0 (voting behaviour equals party line in not a single vote) to 1 (voting behaviour equals party line all votes). We treat 'soft' (party line: yes, MP vote: abstention) and 'hard' (party line: yes, MP vote: no) deviation equally. We exclude free votes, i.e. votes where at least one party has publicly announced to relax party discipline (usually moral issues like embryo research or medicide). Since we look at the agreement rate between an MP and her party, we conceptually do not capture the degree of parliamentary activity, but the degree of expressed policy conflict between the MP and her PPG's position. 3 We operationalise parliamentary activity through a wide range of different parliamentary activities. We include measures of these activities as single variables in our models to be able to distinguish between their distinct effects. 4 Using the same database as for measuring party loyalty, we extract the absence rate in roll call votes for each MP as a first indicator of the degree of parliamentary activity. It is measured by the share of roll call votes, where the MP was absent and not officially excused, compared to all votes she could have participated in.
Further indicators of individual parliamentary activity are parliamentary questions and interpellations. Data on the usage of both instruments is taken from the 'Every single word' database (Remschel and Kroeber 2022). 5 It should be noted that these parliamentary tools are used primarily by the opposition to scrutinise the government. However, it can be used by government MPs as well e.g. to demand the implementation of the coalition treaty or monitor the coalition party's ministers (Höhmann and Krauss 2022;Höhmann and Sieberer 2020). Thus, if a government MP asks a parliamentary question to the government, this is not necessarily viewed as a rebellious act by her PPG leadership or her party's list selectorate. It might be even rather interpreted as a signal of compliance with the party line by asserting the party position against the coalition partner.
Another indicator of parliamentary activity is the number of speeches given by an MP. However, it is actually unclear how this measure should be interpreted conceptually. Since access to the speaker's floor in the German Bundestag is guarded by the PPG leadership, we must expect that especially MPs following the party line faithfully will be granted time to speak for the party (Proksch and Slapin 2012). Thus, this measure is not necessarily an indicator of how hard an MP works, but to a considerable part also an indicator of how faithful she has behaved in the past or ideologically proximate she is to the PPG leadership. Data on the number of speeches given is extracted from the ParlSpeech dataset (Rauh et al. 2017). In order to discriminate between interjections and actual parliamentary speeches, we count all days on which an MP appears with more than 1000 words in the dataset and divide it by the number of days she has been in office.
We have standardised the behavioural measures using the mean and standard deviation of the MP's PPG in the respective legislative period. By this procedure we take potentially varying party cultures in terms of non-conformity in RCVs and in terms of preferences for certain control instruments over others into account. Furthermore, this approach adjusts our measures for potential period-specific effects on the PPG level that may stem from institutional factors, because e.g. parties might reveal a lower base level of party unity and ask more questions to the government in times of opposition, or specific intra-party conflicts occur in a certain legislative period.

Control variables
We control for a range of variables that could confound our relationship of interest. First, we control for the degree of competition for a list position, because undesirable behaviour in terms of collective benefit might be less sanctioned when more incumbents decide to not run for re-election and MPs might anticipate this in their behaviour. We operationalise the degree of competition for list positions in the upper range of the party list through the share of elected MPs, who ran on party list in the last election and do not seek re-election in the current election subtracted from 1. For this measure, we consider legislators who voluntarily leave office and their exit is not related to the nomination or election process. We furthermore control for the MP's relative list position at t, because improvement is logically more likely for MPs with worse list positions and deterioration is more likely for MPs with better list positions and MPs on safe and rather vulnerable list positions may differ structurally in their parliamentary behaviour. Moreover, the inclusion of the previous list position mitigates the endogeneity problem of well-connected and charismatic MPs receiving both better list positions and more floor time for speeches and interpellations to a certain extent. 6 For this purpose, it is reasonable to use the list position relative to the number of list mandates won by the party in the respective SMD rather than the absolute list position. This is, because an absolute list position of e.g. '4' can be extremely safe or extremely vulnerable depending on the context of district magnitude and party strength in the respective SMD. In contrast, the list position relative to won list mandates controls for exactly this context.
We also include if a MP holds a government office (0 for no government office, 1 for minister or junior minister) or a party office (0 for no such office, 1 for [vice-]president of chamber, [vice-]chair of a parliamentary committee, [vice-]head of PPG, whip) as these MPs are both prominent within the PPG (and therefore less likely to be deselected) and have been shown to deviate from the party line less likely (Becher and Sieberer 2008;Saalfeld 1995;Sieberer 2010;Sieberer and Ohmura 2021).
Another confounding variable is the MP's type of mandate (either elected via district tier or list tier), because MPs elected via party ticket might address a different constituency (and thus [s]electorate) with their parliamentary behaviour than MPs elected in single-member districts and were found to be less likely to deviate in roll call votes (Sieberer 2010). As the mixed-electoral system allows MPs to run in two tiers simultaneously, we take into account if a MP runs only on a list position or as a dual candidate in the last election (0 list candidate, 1 dual candidate) and subsequent election, because this electoral feature has also been shown to influence loyalty to the party line (Sieberer and Ohmura 2021). MPs elected in the FPTP tier are often also placed on the party list as an 'additional safety net' (Manow 2007).
Furthermore, it is important to control for seniority, because more seasoned and merited MPs might both behave differently in parliament and be less prone for a deselection. We also include the MP's gender (0 female, 1 male), the MP's age at the beginning of the term in order to take quotas and potentially varying parliamentary behaviour by gender and age into account. Furthermore, we control for whether the MP was elected in West (0) or East (1) Germany, because especially in the earlier electoral periods in our sample, party organisation dynamics and parliamentary behaviour might differ between these regions.
Moreover, we include party fixed effects to control for effects resulting from a specific party feature (stemming from party culture or party organisation) which is stable over time. Additionally, we include fixed effects for each electoral period in all models to account for specific features of the electoral periods in question (e.g. a Bundestag electoral reform in the 14th electoral period or snap elections in the 15th electoral period). The summary statistics of all variables are displayed in the online appendices (see Table A.1).

Descriptive results
Table 1 displays the descriptive summary of our dependent variables-the MP's failure of renomination in the FPTP-tier SMD and the MP's change of the list position by political party. The first column shows that only incumbents of the two largest parties, the SPD and the CDU/CSU failed nomination as a SMD candidate. Since nomination in a SMD usually does not have consequences for the personal composition of the PPG, the benefits of (informally) intervening into a local party branch's decision may not outweigh the costs for (the national or) regional party leadership. In contrast, nomination failures of incumbents in the PR-tier occur even less often in the larger parties of SPD and CDU/CSU. Regarding changes in the list position, incumbents of all political parties more often experience a promotion (40.2%) than a demotion (21.4%) in their list position. The higher share of better list positions can be explained by voluntary exits (of mostly advanced career MPs due to retirement), which create opportunities to achieve a promotion in list positioning for re-election seeking MPs. MPs of the larger parties SPD and CDU/CSU are more often nominated on relatively better list positions than in the smaller parties of FDP, Greens and Left.

Method of analysis
The unit of analysis is an MP within an electoral period. Thus, if an MP has served more than one electoral period, she will appear multiple times in the dataset, which is why we cluster standard errors by MP. To evaluate our hypotheses, we make use of four separate logistic regression models for failing nomination in the CLPR-tier, for receiving a worse list position, for receiving a better list position and for failing nomination in the FPTP-tier. In the models with 'better list position' as dependent variable, we excluded MPs who received the first list position (i.e. 'Spitzenkandidaten') at t, because they logically were not able to receive a better list position and thus counting these cases as 'stable' list position would not be appropriate. We excluded 6 MPs who run for the same party but in a different region. Moreover, we dropped 88 MPs who participated in less than 50 RCVs to arrive at reliable estimates of the parliamentary behaviour indicators. We excluded 42 further MPs who had standardised voting agreement values below three standard deviations from the PPG mean to make sure that our results are not driven by some single very unusual mavericks. Overall, our modelling strategy with fixed effects for parties and electoral periods as well as many relevant control variables and exclusion of unusual cases produces conservative estimates for the parliamentary behaviour indicators, so that we are confident to rather under-than overestimate their effect size.

Multivariate results
We now turn to the results of our multivariate specifications. Figure 1 reports the estimates of the three CLPR-models (for the full regression table, see models 1 to 3 of Table A.2 in the online appendices). The estimates represent the average marginal effect of each variable on the probabilities to fail, to receive a worse list position and to receive a better list position while holding all other variables at their observed values (Hanmer and Kalkan 2013). It displays the 95% as well as the 90% confidence intervals of every point estimate.
As described above, the variables indicating parliamentary behaviour were standardised by the PPG mean and standard deviation in each electoral period for reasons of comparability. Our models show that an increase by one standard deviation in the roll call voting agreement rate 7 on average decreases the probability of failing nomination by 0.74 percent (p < 0.05) and of receiving a worse list position by 3.29 per cent (p < 0.01), while it increases the probability of receiving a better list position by 3.11 per cent (p < 0.05). An increase by one standard deviation in the number of parliamentary speeches held per day in office 8 decreases the probability of failing nomination on average by 1.69 per cent (p < 0.001) and to receive a worse list position by 3.78 per cent (p < 0.001). There is no effect of the number of parliamentary speeches on the probability of receiving a better list position significant at conventional levels of statistical significance. According to our models, the attendance rate in roll call votes, the number of parliamentary questions and the number of interpellations do not have a significant influence on the list reselection at all.
We test our results for robustness by analysing different operationalisations of the dependent variable applied in the literature (see Table A.3 in the online appendices). While the findings for parliamentary questions and absence rate are overwhelmingly robust, the findings for loyalty in RCVs and number of parliamentary speeches vary in their level of statistical significance across different operationalisations but not in their substantial direction. This makes us confident that our inferences are not biased by the operationalisation of our dependent variable. Utilising list positioning relative to list length (see Baumann et al. 2017)-as a proxy for expected list mandates won-shows a significant effect of deviant voting behaviour at a significance level of p < 0.1 for both worse and better list positions (models 5 and 6 of Table A.3.1). As a second robustness test, we analyse the list position at the election t + 1 relative to the won list mandates while controlling for relative list position at election t (see Louwerse and van Vonno 2022). Here, only the number of parliamentary speeches and the number of parliamentary interpellations are statistically significant predictors of list position change (model 7 of Table A.3.2). Voting behaviour in RCVs is not statistically significant, which may however be explained by the fact that we are not able to incorporate the crucial cases of nomination failures within this approach, so that these essential cases are missing. This may also explain why previous studies did not find significant effect for party loyalty in RCVs on the selection process. All results discussed so far remain substantially robust when using panel logistic regression with random effects for MPs (see models 8 to 11 in Table A3.3 in the online appendices).
In hypothesis H3a and H3b, we had formulated the expectation that selectorates for lists in the PR-tier value loyalty to the party line more strongly, while selectorates for SMDs in the FPTP-tier value parliamentary activities that create visibility more strongly. In order to evaluate these claims, we compare the predictive power of our parliamentary behaviour indicators for the probability to be deselected in the more central PR-tier and in the more decentral FPTP-tier. Figure 2 shows the average marginal effects of the behavioural indicators (see model 4 of Table A.2). As already discussed, loyalty to the party position in RCVs indeed yields a statistically significant effect on the deselection probability in the more central PR-tier. However, there is no such effect in the more decentral FPTP-tier, which corroborates H3a. In contrast, the number of parliamentary interpellations only has a significant effect on the FPTP-tier nomination process, but not on the PR-tier process. 9 Both observed differences are in favour of our divergence hypotheses, which had suggested different motivations of the selectorates of candidates in the different tiers. The regional list-selectorate is tied closer to the national party leadership both in organisational terms and electoral goals than the sub-regional district-selectorate and thus has a comparably higher interest in party unity. In contrast, the sub-regional district-selectorate is more interested in candidate visibility, which is why loyalty to the party line is less influential and parliamentary activities that create higher visibility in the general population, like submitting parliamentary interpellations to the government, have a higher influence on the decision to reselect an MP. Figure 2 moreover shows that the more parliamentary speeches an MP held in a legislative period, the lower is her probability of failure in both nomination processes. An explanation for this observation could be that the number of parliamentary speeches is correlated to both loyalty to party position and visibility (see discussion).

Discussion
What do these findings mean with regard to our theoretical claims? The interpretation of the effects of party loyalty in RCVs on list positioning is straightforward: Our models support the claims derived from the spatial model and the neo-institutional party loyalty literature in terms of dissent-shirking. The selectorate tends to punish rebellious MPs and reward MPs toeing the party line in the allocation of list positions. Our findings thus establish a comprehensive empirical foundation of the often stated but rarely investigated assumption about the genesis of comparably high levels of party unity in parliamentary democracies via selective incentives by the party (leadership) at the candidate selection stage. The results are thus in favour of our first set of hypotheses, which stated that MPs following the party line faithfully are less likely to be deselected and to receive a worse list position (H1a) and more likely to receive a better list position (H1b). This finding is surprising insofar as variance in party loyalty is very low because of the high general level of party unity in parliamentary systems. This had been the main criticism against the explanatory potential of party loyalty in RCVs for punishment or promotion in candidate selection in the previous literature (Baumann et al. 2017: 984;Louwerse and van Vonno 2022: 5;Yildirim et al. 2019: 795). Nevertheless, even this low variance has shown significant effects on sanction and promotion in candidate selection in our study.
Concerning our second set of hypotheses, H2a and H2b, we only find limited evidence for the expected effects of parliamentary activity on reselection. The findings for the different measures of the parliamentary activity level do not unanimously speak for the concept of leisure-shirking as criterion for reselecting more hard working MPs and deselecting slacking MPs. The only measure of the parliamentary activity level that shows significant effects on the probability to fail nomination in both the FPTP-tier and the PR-tier and to receive a worse list position in the PR-tier is the number of parliamentary speeches. This indicator is, compared to roll-call voting loyalty, much fuzzier in terms of conceptual interpretation. One possibility, in line with the concept of leisure-shirking, is that the correlation is caused by being visibly active via holding many speeches and thus signalling the selectorate commitment to the party goals. However, since we know that access to the floor is gate-kept by the PPG leadership and depends on ideological proximity to the PPG leadership (Proksch and Slapin 2012), we might also observe a confounding effect of being ideologically close to the party position and therefore being granted both a good list position and access to the speaker floor (Louwerse and van Vonno 2022: 2).
To investigate this possibility further, we calculate a model with the standardised number of speeches held per day in office as dependent variable. Studies have shown that there are considerable correlations between the ideological distance from the party mean and rebellious behaviour in roll call votes even in parliamentary democracies with high levels of party discipline (Willumsen and Öhberg 2017), which makes loyalty in roll call votes a suited proxy for ideological proximity. Our results show that the less often an MP votes against the party line in RCVs, the more speeches she holds on average (see models 12 and 13 of Table A.4 in the online appendices). Mediation analyses combining each of the equations of model 1 and model 2 with the equation of model 13 affirms that considerable parts of the effect of parliamentary speeches on nomination failure (13%) and worse list positioning (9%) are mediated by ideological proximity (see Table A.5 in the online appendices). Nevertheless, even when controlling for party loyalty in roll call votes, there is a substantial partial effect of holding speeches in parliament on the probability of receiving a worse list position and also of failing renomination completely, as could already be seen in our models 1, 2 and 4.
Thus, the results show that the number of parliamentary speeches has a positive effect on reselection probabilities beyond ideological proximity to the party leadership position. Whether this effect can be attributed to choices made by the MPs is however beyond the scope of this study. MPs may be selected because of their rhetorical talent or popularity in general both by the party leadership for parliamentary debates and by the selectorate for promising list positions. While the other indicators of the degree of parliamentary activity (RCV attendance, parliamentary questions and interpellations) are more reliable measures of actual individual MP behaviour, the number of speeches is partially rather reflective of decisions taken by the PPG leadership than of decisions taken by the individual MP. Thus, we cannot assert that leisure-shirking, which by definition is individual MP behaviour, influences reselection in the PR-tier based on our results.
However, our results show that reselection in the more decentral FPTP-tier is affected by the degree of parliamentary activity, although only one of our indicators (parliamentary interpellations) showed a significant effect. Arguably, this effect should be interpreted in terms of rewarding visibility rather than contribution to the national party goals, because it has not shown an influence in the more central PR-list selection process. 10 We have assumed largely the same incentivising function of all three kinds of selectorate actions-positional improvement, positional deterioration and deselection. Since our results indicated that less speeches make a worse list position more likely, but more speeches do not make a better list position more likely, a further potential avenue for future studies could be to consider different causal mechanisms that may underlie these kinds of actions more closely.

Conclusion
This study contributes in various ways to the existing literature on the link between parliamentary behaviour and candidate reselection in parliamentary democracies. For the first time, MPs who failed to be renominated are considered in the study of the link between parliamentary behaviour and candidate reselection. Therefore, we cover the full variation of the phenomenon to be explained and do not miss crucial units of observation in our analysis. Our findings suggest that dissent-shirking plays a more pivotal role in the reselection of incumbents. MPs diverging more from the party line in roll call votes are more likely to be deselected and more likely to experience a deterioration in their list position, while toeing the party line faithfully is rewarded with a better list position. According to our results, leisure-shirking by MPs has less clear-cut consequences for renomination probability as well as on list positioning. Only for the number of parliamentary speeches we find a significant effect on renomination prospects and worse list positioning, but not on promotion in list positioning. However, the finding of the number of speeches influencing reselection does not unanimously prove the influence of leisure-shirking on reselection, because doubts can be raised in how far the number of parliamentary speeches given by an MP is an accurate measure of leisure-shirking. MPs who give fewer speeches might be either not selected because they are 'bad' (in a spatial loyalty sense) or 'dreadful' (not rhetorically talented) or do not volunteer or even refuse to speak, because they are 'slacking' (leisure-shirking). Within our study we could only show that there is an effect beyond the 'bad' category, however we are not able to disentangle the 'slacking' and 'dreadful' category with our data. This demands further investigation in future studies.
Overall, our study shows the potential of researching the 'selectoral connection' from the selectorate's side to enhance our knowledge about intra-party dynamics in the strategic pursuit of collective goals. From a democratic theory perspective, this is especially important in parliamentary systems with closed lists, because here, the accountability of representatives to the citizens is entirely organised through parties (Müller 2000). If individual parliamentary actions that blur the party position were not sanctioned by the selectorate, citizens would not be able to rely on the party label as an informational short-cut and doubts about the integrity of democracy would have to be raised. However, since we find that dissent-shirking does affect candidate selection at least in the list selection process, we can assert that individual accountability through parties overall functions as proposed by the democratic chain of representative delegation in the parliamentary system of Germany between 1990 and 2017. Since we find this mechanism to be at work in the unlikely case of decentralised candidate selection for regional lists in German Bundestag elections despite potential contamination through the FPTP-tier of the German mixed-member system, we are confident that dissent-shirking should also play a crucial role in candidate selection in parliamentary systems with more centralised candidate selection and a pure PR electoral system. However, as our results concerning the even more decentralised SMD candidate selection in Germany indicate, in less centrally organised processes, we may encounter a democratic deficit in this respect. Cross-country comparisons from the selectorate's perspective that include systems with more centralised (e.g. Spain and the Netherlands) and more decentralised candidate selection (e.g. UK and Ireland) could further disentangle the effect of candidate selection rules (decentralisation, inclusiveness) and electoral system properties (district magnitude, ballot structure) on the kind of parliamentary behaviour that is taken into account by selectorates at different levels.

Notes
1. There is a certain tension between the theoretical derivation and the empirical reality. In game theoretic terms, candidate selection is a power transferred to the central authority 'party leadership' to overcome the collective action problems of party unity and party vitality. In the democratic delegation chain, candidate selection serves as an ex-post control mechanism available to the principal 'party leadership' to ensure loyal agents (MPs) (Müller 2000). In empirical reality, however, the actor in candidate selection is the party selectorate, which may share certain goals with the party leadership, but is not necessarily personally identical with the party leadership. Depending on the organizational and preferential overlap between party leadership and the party selectorate, we can expect the party selectorate to act more or less according to the incentive logic for providing public goods. We thank an anonymous referee for pointing us to this clarification. 2. We thank an anonymous referee for pointing out the potentially independent effect of electoral institutions on the selectorates' preferences in candidate selection. 3. Conceptually, we do not understand this measure as proxy for the general preferential distance between the MP and her party, but as measure of how often an MP has expressed dissent with the party line at the most visible stage possible: recorded votes in parliament (Proksch and Slapin 2015: 26). Consequently, the critique of selection bias because of the restriction to recorded votes (Carrubba et al. 2006;Hix et al. 2018;Hug 2010) does not apply to our study. We do not expect dissenting behaviour in non-recorded votes to affect party leaderships' reselection considerations, because this kind of dissenting behaviour is very unlikely to be noticed by the public and thus not harmful for the party label. 4. We abstain from calculating an overall index of parliamentary activity because this would assume to either consider every parliamentary activity as equally important for the selectorate or require a weighting of the different activities, which is hard to define, since the weight of each type of activity for the selectorate is unclear. Moreover, this weight might vary across the career stages of MPs (Bailer and Ohmura 2018). 5. In a plausibility check of our data, we noticed that cumulative numbers of parliamentary interpellations and questions from the Every Single Word dataset slightly differ from records published by the parliamentary documentation service of the German Bundestag in the electoral periods 16 to 18. We did not detect systematic deviations alongside PPGs and thus assume the missing occur randomly. 6. We thank an anonymous referee for this point. 7. Since this measure was standardized by PPG per electoral period, one standard deviation amounts to dissent with the party line in 0.7 percent up to 7.5 percent of all RCVs (mean value is 2.0 percent). 8. See note 7; one standard deviation amounts to 1.2 up to 3.6 speeches per year. 9. All results remain substantially robust when using panel logistic regression with random effects for MPs. The only difference is that in our FPTP-tier failure model, the coefficient of parliamentary interpellations becomes insignificant, while the coefficient of parliamentary questions becomes statistically significant (p < 0.1), which does not hurt our interpretation concerning the theoretical mechanisms. 10. Our findings suggest that success in reselection in the different tiers is partly contingent on different types of parliamentary behaviour, which both contribute to the collective party goals of party unity and party vitality. This may be a potential benefit for parties in mixed-member systems contrary to pure (s)electoral systems. We thank an anonymous referee for this point.
'Candidates and Careers of MPs' and of the Comparative Politics Colloquium at the University of Bamberg, and especially Ulrich Sieberer, Elena Frech, Radoslaw Zubek, Tom Louwerse as well as the West European Politics reviewers for their valuable comments and suggestions. We moreover thank Lennart Vogt and Natalia Mleczko for excellent research assistance. All errors remain are own.

Data availability statement
Data and code to replicate the results in this article will be made publicly available at the Harvard Dataverse.