Division of labour and dissenting voting behaviour of MPs in a ‘ working parliament ’

In the literature on the determinants of party unity, one pathway has remained largely neglected: division of labour. Given their workload, members of parliament (MPs) are only thoroughly concerned with a subset of policies. We argue that this results in MPs casting fewer dissenting votes on matters within their area of specialization since they have had the opportunity to shape the party line there. Regression analyses using data for the German Bundestag support this hypothesis, including four important refinements: Not only the current but also past membership in the responsible committee reduces an MP ’ s defection probability. Additionally, this pattern is more pronounced for policy spokespersons and for less consequential, i.e., non-legislative votes as well as for issues less salient to the MP ’ s party. The results have implications for our understanding of MPs ’ legislative behaviour, the functioning of parliaments as institutions and for the relationship between parties, MPs and voters.


Introduction
Party unity is an important prerequisite for much of what happens politically in parliamentary democracies (Andeweg and Thomassen, 2011). Theories of party competition, e.g., on issue ownership and how parties make use of it in campaigns, treat parties as monolithic blocks that are bound to their party platforms (Budge, 2015). In elections, citizens often identify with, but at least vote for parties whose relative strength usually decides on the composition of parliament. Thereby, voters expect party representatives to feel committed to the policy positions with which they ran for election ('promissory representation', Mansbridge, 2003). Party unity is thus considered a precondition for responsible party government (Bowler et al., 1999). A lack of party unity is punished at the ballot box (Lehrer et al., 2022) and, on the member of parliament (MP) level, with lower chances of career advancement (Schobess and de Vet, 2022). Moreover, hypotheses on the formation of minimal winning coalitions would be meaningless if parties could not rely on the support of all their MPs. Finally, in order to exert measurable effects on public policies (Hibbs, 1977;Wenzelburger and Zohlnhöfer, 2021), parties have to act cohesively, both vis-à-vis coalition partners and the opposition. Therefore, exploring what drives or impedes the unified legislative behaviour of parties and their MPs points far beyond parliamentary research.
Division of labour has been described as one of several pathways to party unity, meaning a factor conducive to MPs of the same party overwhelmingly voting together in parliament (Andeweg and Thomassen, 2011). On the individual level, division of labour means that MPs are not concerned with all policies in detail but specialize on certain issues which they work on in parliamentary committees and preparatory bodies of their party group. On the institutional level, committees are considered the 'workhorses' of the legislative process (Siefken, 2021: 117). They fulfil various functions in representative democracies: For opposition parties, committees are a means of introducing new ideas, criticism or modifications of bills in the decision-making process that is, apart from that, dominated by governing parties (Strøm, 1990). Governing parties use committees for shadowing, monitoring and overriding their coalition partners and to modify government bills (Kim and Loewenberg, 2005;Fortunato, 2019). In both respects, committee deliberations serve as a 'test stage' for the plenary debate and decision (von Oertzen, 2006). Moreover, committees and their corresponding bodies within the party groups are targets of interest group influence (Cross et al., 2021). Finally, committees are concerned with scrutiny and oversight activities (Siefken, 2018;Norton, 2019).
Division of labour and MPs' issue specializations have implications for their roles within their party groups and for their legislative behaviour, including their varying propensity to toe the party line. However, empirical tests of this argument using observational data, i.e., measures of actual MP behaviour, are rare. In this paper, several behavioural manifestations of this argument are deducted and transferred into testable hypotheses. In a first step, we argue that MPs have lower defection rates when motions are on the floor which they were concerned with during their committee work (policy-shaping hypothesis). In a second step, several refinements are discussed: whether also past committee memberships are effective and whether the committee membership effect is conditioned by having a leadership position or by the importance of the vote for an MP's party. After outlining the study design, the results of panel regression analyses using data for the German Bundestag (2017-2021) are presented. The article concludes with a discussion of the findings' implications for the nexus between parties, MPs and voters.

State of the art: party unity and parliamentary committees
The question of why MPs overwhelmingly vote together in parliamentary democracies has been subject to empirical studies especially for the last two or three decades. Theoretically, the literature identified a set of 'pathways to party unity' (Andeweg and Thomassen, 2011), among them (1) homogeneity of preferences, (2) loyalty and (3) discipline. Voting unity is not determined by preference similarity among the MPs alone (Willumsen, 2022), although MPs' own ideological stances demonstrably affect their voting behaviour in certain policy areas (e.g., Degner and Leuffen, 2016). The loyalty path to unity has been conceptualized as the result of MPs' party socialization (Rehmert, 2022;Mai and Wenzelburger, 2023). To capture the varying effectiveness of party discipline, the career-related dependence of MPs on their party, compared to other principals (e.g., voters), has been approximated. Empirically, holding or ambitions to reach powerful offices (Kernecker, 2017;Bhattacharya and Papageorgiou, 2019;Zittel and Nyhuis, 2019) strengthen, whereas an MP's need for personal votes in the general election (Sieberer, 2010;Sieberer and Ohmura, 2021) or in intra-party primaries (Sozzi, 2023), personalized campaign styles (Zittel and Nyhuis, 2021), local ties (Tavits, 2009), economic interests of an MP's constituency (Stiller, 2023), outside earnings (Mai, 2022), a low prospect of being promoted to higher office (Benedetto and Hix, 2007), career ambitions at another political level (Meserve et al., 2009), party switching (Gherghina and Chiru, 2014), impending retirement (Mai et al., 2023, but Willumsen andGoetz, 2017 for non-results) or being a 'career politician' (Heuwieser, 2018) weakens an MP's propensity to toe the party line, arguably due to a changed effectiveness of party discipline compared to the baseline MP.
In contrast, empirical implications of division of labour as a fourth pathway to unity have rarely been included in observational studies of vote defections (Andeweg and Thomassen, 2011;van Vonno 2019). The division of labour principle in parliaments can be observed most clearly in MPs' committee work. MPs' issue specialization has effects on their re-election chances (Frech, 2016), bill co-sponsorship patterns (Baller, 2017), participation in parliamentary debates (Fernandes et al., 2019), communication with voters (Meyer and Wagner, 2021), further career advancement (Cirone and van Coppenolle, 2018), attention towards particular issues (Borghetto et al., 2020) and their overall level of activity (Louwerse and Otjes, 2016). However, few studies focus on a connection between committee work and party unity. Based on interviews with German and Dutch MPs, Mickler (2019) describes how the intra-party working groups develop their positions and why the whole party group usually follows them. Fujimura (2012) illustrates, for the Japanese case, that committee assignments are used to reconcile the party's need for unity and MPs' electoral concerns. For the U.S. case, Kanthak (2009) shows that MPs having received plum committee assignments toe the party line more frequently than other MPs. Additionally, Grimmer and Powell (2013) find that legislators who were involuntarily removed from preferred committees have significantly higher defection rates. Conversely, MPs overwhelmingly toeing the party line receive more attractive committee seats or even chair positions, compared to more rebellious colleagues (Leighton and López, 2002;Cann, 2008;Whitaker, 2019; but see Chiru, 2020 andFernandes et al., 2022 for non-results).
Although those findings are related to division of labour, not the MPs' specialization itself but a changed effectiveness of party discipline after having gained or lost desired committee seats seems responsible for the variation of party unity. Committee assignments are one of the 'carrots and sticks' the party leadership uses to enforce discipline (Bailer, 2018). Only Willumsen and Öhberg's (2017) study of the Swedish case explicitly links MPs' issue specialization to their voting behaviour and finds that MPs vote less frequently against the party line concerning issues that have been dealt with in the parliamentary committee they belong to. We will complement their argument in three crucial respects: First, we argue that not only the current but also past committee memberships are likely to affect individual legislative behaviour. Second, we expect behavioural differences between policy spokespersons and ordinary committee members. Third, we posit that the effect of committee membership is conditional on the importance of the vote. In the subsequent section, we theorize on those aspects and derive five falsifiable hypotheses.

Theory
In its 19 th term (2017-2021), 870 draft bills and hundreds of other motions (e.g., amendments, resolutions) were introduced and (most of them) adopted or rejected during votes in the German Bundestag. The MPs were provided with more than 31,000 documents as a technical basis for their decisions. However, they are only able to take note of a small subset of those documents and to form their opinion on only some of the motions brought to the floor (Ismayr, 2012). Most obviously, time constraints prevent them from doing soas described by a leading MP cited in Searing (1995: 680): 'The volume is so great that most of the time they would be completely lost, they wouldn't know which way to vote'. In order to keep the amount of information to process manageable, MPs usually join one or two standing committees that, broadly, resemble the jurisdictions of the government departments and deal with bills and other motions related to them (Siefken, 2021). Within each parliamentary party, the committee structure is reflected by working groups (Arbeitsgruppe) in which the MPs collaborate in defining their party's position. In smaller parties, the working groups consist of MPs from multiple committees with similar topics, e.g., foreign, defence and development aid policy (Arbeitskreis; Mickler, 2019). The positions of the party's working groups mark an important predecision for a motion's prospect to be passed. Suppose MPs or party factions aim to modify a proposal: In that case, not the plenary, the committee or the party group meeting but the responsible working group of the parliamentary party is the suitable arena for such an undertaking (Schöne, 2010). If its members have agreed on a common position, a discussion in the whole party group is 'preferably avoided' (Mickler, 2019).
Consequently, membership in those working groups is the most promising avenue for backbench MPs to pursue their policy-seeking goals.
There are several reasons why MPs do not often disagree with the other policy specialists of their party regarding their own topics. Besides other factors, committee assignments and, consequently, MPs' 'own' topics often reflect their educational or occupational background (Mickler, 2018). In the German parliament, for example, members of the agricultural committee are to a disproportionately higher share farmers themselves; most of the members of the justice committee are lawyers (Ismayr, 2012). Additionally, according to the distributive theory of legislative organization, MPs systematically join committees related to constituency characteristics: MPs from poor districts are often members of social policy committees, whereas members from constituencies with an increased demand for construction activity join infrastructure committees (Martin and Mickler, 2019 with further references). Both the personal and electoral reasons for their committee membership might favour a common understanding of problems and, possibly, even a rough homogeneity of preferences among the policy specialists of a party group. Within the cooperative structures of their working group, the single MPs are described as 'busy, well-informed, often experienced, more or less knowledgeable policy workers who develop and evaluate parliamentary motions' (von Oertzen, 2006: 256, own translation). Within their role, they are able to introduce their personal preferences into the decision-making process of their party groupin some cases, even proactively and not only related to substantiating, modifying or impeding government motions (Schöne, 2010). Consequently, because those policy specialists are involved in shaping their party's position, they could reduce the policy distance between them and the responsible working group of the party and, finally, the whole party group. Since the MPs thus helped to shape the party line within their area of specialization, they have fewer reasons to vote against it (Willumsen and Öhberg, 2017). Dingler and Ramstetter's (2021) inverted finding whereby MPs have higher defection rates within their focus area might be rooted in that the authors measure MPs' interest (by parliamentary questions) and not their capability of shaping policies.
The negative effect of committee membership on defection rates could be amplified by the fact that among the policy specialists who have often worked together for many years there is an increased expectation of mutual loyalty and not to distinguish oneself by dissent from the responsible working group (Schöne, 2010). After some Christian Democratic MPs had voted against measures to resolve the Eurozone crisis in 2015, the leader of the party group threatened them with their removal from the respective committees: Those who voted 'No' cannot keep their seats in committees where it is essential to keep the majority, e.g., in the budget or the EU committee [ : : : ]. The party group sends MPs into committees so that they represent their party's position there. (Die Welt, 2015, own translation).
In the year before, another member of this party lost his rapporteur position after publicly proposing social policy reforms that were not coordinated with his working group (BILD, 2014). Both examples illustrate that the leadership expects MPs not to take deviating stances on topics which they work on in their respective committees.
Hence, most dissenting votes are likely to happen outside their areas of specialization. MPs not familiar with a policy field might lack the expertise to have their own view on a particular motion (van Vonno, 2016). Besides that, MPs might also not have the same interest in every decision and may ignore certain topics (Schöne, 2010). If they do not have an opinion on an issue, MPs take voting cues from the policy specialists of their party. However, if MPs actually have an opinion and are not members of the respective committee and intra-party working group, they have a greater leeway to dissent. Additionally, there are more substantive reasons for defections, as these MPs were not involved in shaping the party line.
HYPOTHESIS 1 (policy-shaping hypothesis): MPs have a lower probability of voting against the party line in matters inside their area of specialization.
That said, we expect that not only the current committee membership but also which topics MPs have specialized on in the past influences their legislative behaviour. Former committee membership indicates interest in the policy area, and MPs keep at least parts of the expertise gained in the past. Therefore, if MPs have been committee members one or two legislative terms ago and, after that, turned to another policy area, they most likely still have an opinion on topics of their former committees. Additionally, during their former membership in a committee and the respective intra-party working group, a convergence might have taken place between the MPs' preferences and the party line which those MPs used to shape in former times. Lastly, MPs might still feel a sense of loyalty towards their former working group and its members whom they know from their prior collaboration. To conclude, we expect the policy-shaping hypothesis to be valid not only for MPs' current but also for their past committee membershipswhich has not been empirically tested before.
HYPOTHESIS 2 (former specialist hypothesis): MPs have a lower probability of voting against the party line in matters which they were specialized on in former terms.
As insiders of the parliamentary business (von Oertzen, 2006), the main tasks of policy spokespersons are to lead the working group concerning a particular policy area and to express the working group's position vis-à-vis the committee, the media and, secondarily, the policy spokesperson of the coalition partner (Ismayr, 2012;Mickler, 2019). One of their primary duties is to advocate the position of their working group within the whole parliamentary party (von Oertzen, 2006). In this respect, policy spokespersons have, compared to ordinary MPs, significantly better means to shape the party line. Acting as a first filter, they can push certain initiatives or delay others (von Oertzen, 2006). Additionally, since they are members of the party group leadership, they have an informational advantage and a certain leeway in settling conflicts with other intra-party working groups. Their informational advantage is also rooted in the fact that they have privileged access to the parliamentary staff of their party and are able to use its expertise for their purposes (Petersen and Kaina, 2007). If the working group is divided on an issue, the policy spokesperson most likely has the authority to resolve conflicts with their decisive vote. This is reflected by the MPs' own perceptions: 68% of the German MPs who took part in the survey support the statement that the spokespersons define the position of the (whole) party group (van Vonno, 2016). However, coalition agreements, the party platform or (rarely) a resolution of the whole party group constrain the spokesperson's room for manoeuvre.
In addition to the power of significantly defining the party's position, Schöne (2010) describes that, in exchange for the privileges those policy speakers have (e.g., media attention), they are expected to strictly represent their party group's position and not dissenting opinions of their own. A recent example illustrates possible role conflicts: Whereas his party group was sceptical, the Social Democratic spokesperson for defence policy, Fritz Felgentreu, supported the claim of its coalition partner to purchase armed drones for the German military. After his party had postponed a decision once again, he resigned from office in 2020 and explained his decision on Twitter as follows (own translation): Either I could stick to the position [of my party] vis-à-vis the public and the military although everyone knows that I have another opinion [ : : : ] or I could dissociate myself from the parliamentary group and my party. As a member of both, I expect more loyalty and more solidarity with the leadership and the majority. Therefore, I have resigned from my office as the spokesperson for defence policy.
Consequently, both the spokespersons' greater influence on the decision-making process and their stronger attachment to the majority position leads to expect the following: HYPOTHESIS 3 (spokesperson hypothesis): The unity-inducing effect of votes inside their area of specialization is stronger for policy spokespersons compared to ordinary committee members.
Many studies find that defection probabilities differ by policy issues (e.g., Skjaeveland, 2001;Stecker, 2015;Bergmann et al., 2016). However, this issue-based approach is only an approximation of where those differences actually stem from: a varying salience of the respective issue either for the party's brand name (Traber et al., 2014) or the electorate (Ohmura, 2014). Accordingly, we expect that the importance of an issue conditions the explanatory power of division of labour as a pathway to unity. We focus on two dimensions of importance: the type of the vote and issue salience.
Not every vote is equally consequential. Owens (2003) argues that the politics of party unity is affected by the type of document which is put to the vote. In votes that are crucial for the functioning of the government (e.g., budget), party unity is expected to be higher than in less momentous votes (Rahat, 2007). According to Stecker (2015), the vote type moderates also the determinants of unity, and he differentiates between legislative and non-legislative matters. Whereas the former exerted direct policy implications through changes in legal rules, the latter had a more symbolic value by being political expressions of intent without direct consequencesalthough the documents vary in importance within those categories as well.
Beyond vote type, the literature on issue ownership assumes that topics are not equally important to parties, their activists and voters (Budge, 2015). This connection between parties and particular issues affects their strategies in election campaigns (Wagner and Meyer, 2014), legislative agenda-setting (Green and Jennings, 2019) and in government (Jensen and Seeberg, 2015). Since MPs are not only agents of their constituency but, primarily, of their parties whose success decides, for the most part, on their electoral and other career-related fortunes, the connection between parties and issues might also affect its MPs. Consequently, we expect a behavioural difference between high-and low-salience topics. The latter include those hardly covered in the election manifesto or technical issues difficult to grasp without specific policy knowledge (Andeweg and Thomassen, 2011). Those topics are unlikely to be politicized since they do not relate to the party's identity or the core interests and values of its voters. In such a setting, less constrained by the party manifesto and without creating facts by legislation, the policymaking process leaves enough room for parliamentary actors to substantially influence those policies. Accordingly, we expect division of labour to be particularly effective here, with lower defection rates of committee members.
In contrast, especially when consequential or salient issues are on the floor, parties must reckon with their actions being closely watched by voters. Electoral research has shown that voters indeed react to MPs' legislative behaviour (Papp and Russo, 2018;Wagner et al., 2020;Duell et al., 2023). Disunity as a negative valence signal endangers the party's issue ownership and, for government parties, their policy-making capability and will ultimately be punished by voters (Greene and Haber, 2015;Lehrer et al., 2022). Accordingly, the party group leadership is inclined to use the whip to a stronger extent during consequential and salient votes in order to prevent harm to the party (Owens, 2003). Consequently, for legislators, party unity hinges much more on a trade-off between their preferences, the interests of their constituency (Stiller, 2023) and the incentives of party discipline. In such a setting where parties and MPs have a clear opinion, taking voting cues from policy specialists becomes largely meaningless (Andeweg and Thomassen, 2011;van Vonno, 2016). Moreover, it is less necessary since intra-party preferences are more homogeneous than on less salient issues (Sieberer, 2010). What is more, for legislative matters and salient issues, policy formulation transfers from the bottom-up process through committees more into a top-down process driven by top representatives of the coalition parties and the government bureaucracy (Schindler, 2019). Possible conflicts are then resolved elsewhere, which further limits the scope of committee members for shaping the party line. This leads to another two conditional hypotheses: HYPOTHESIS 4 (vote type hypothesis): The unity-inducing effect of votes inside their area of specialization is stronger in non-legislative compared to legislative votes.
HYPOTHESIS 5 (issue salience hypothesis): The unity-inducing effect of votes inside their area of specialization is stronger for low-salient compared to highly salient issues.

Data and methods
The propositions are tested quantitatively using data from the German parliament. First, the Bundestag is regarded as a 'party group parliament' (Ismayr, 2012), i.e., a parliament with powerful party groups which dominate parliamentary work. Second, it is considered a 'working parliament' with a high degree of issue specialization among the MPs (Steffani, 1979;Lord, 2018) and with the most policy-influential committees in Western Europe (Zubek, 2021). Therefore, Germany is a suitable (and certainly a most-likely) case for an argument based on the nexus between parties and committees. However, the findings should be generalizable to other parliaments in Europe with strong parties and at least a moderate influence of committees (and preparatory bodies of the party groups) on parliamentary decisions.
We collected data on all roll-call votes in the most recent 19 th parliamentary term (2017-2021). Despite the discussion on their representativeness and its consequences for the observed level of unity (Ainsley et al., 2020;Yordanova and Mühlböck, 2015), roll-call votes are the only source of individual-level voting behaviour for our case. Free votes, i.e., votes where MPs were released from party discipline, are, in principle, not restricted to morality policies but could be tactically imposed by the party group leadership if the party line is difficult to enforce (Ohmura and Willumsen, 2022). In the legislative term under study, three unwhipped votes were held on a reform of the organ donation system and, thereby, a classical issue of conscience. Since the determinants of MPs' behaviour systematically differ between whipped and unwhipped votes on morality policy issues, an exclusion of the latter from the sample is justified (see  on MPs' voting behaviour in those votes). Finally, the models consist of 241 roll-call votes on various kinds of documents (e.g., bills, amendments, motions or resolutions).
The unit of analysis is an MP's voting decision in a single recorded vote. This highly disaggregated data structure is necessary given that the main independent variable, 'own committee: at the time of the vote', measures dichotomously whether an MP votes on a document that has been deliberated in a committee he/she belongs to. Consequently, this variable varies both on the MP and the vote levelsuch variance could not be explored if aggregated unity scores for MPs or votes were used. Initially, this variable is coded 0 for all votes on motions that were not assigned to at least one committee or, in the case of amendments, are not related to a bill assigned to a committee. This concerns about a quarter of all votes. For the remaining motions, on the vote level, we only consider the leading (federführend) committee since only in those committees, a thorough discussion takes place, possibly including expert hearings, and a voting recommendation for the plenary is resolved ( §63 Bundestag rules of procedure). Other committees having only an advisory function in the deliberations are not considered in the main analysis. On the MP level, we only take ordinary committee memberships into consideration for our main models. Even though most MPs are deputy members of additional committees, they only attend their meetings when ordinary members are absent. More importantly, those deputy members are usually not involved in the discussions within the intra-party working groups. Consequently, our policy-shaping hypothesis does not apply to deputy members. In robustness checks, we will explore how the results change if those two restrictions are relaxed (Appendix A3). For now, at the descriptive level, 5.4% of the observations are cases where MPs vote on a document that has been dealt with in their committee(s).
For the variable 'own committee: before the time of the vote', the ordinary membership in the respective leading committee during the past two legislative terms, as well as in the current term until the day before the vote, is counted. We chose two past terms as a cut-off point because we assume that policy knowledge and connections to the former working group shrink the longer an MP is not a member anymore. In order to disentangle the effects of past and current membership, the variable is coded 0 if an MP is still a member of the respective committee. The dichotomous variable 'policy spokesperson (all subjects)' takes value 1 if an MP at the time of the vote leads a working group of the parliamentary party which is related to the work of one or more committee(s)irrespective of whether the vote is pertaining to those issues. If the party has elected deputy group leaders who are responsible for certain (broader) topics, they are counted as spokespersons as well since they are actively involved in settling conflicts between different working groups and, as a connection between the leadership and working level of the parliamentary party, have significant powers in the decision-making process (von Oertzen, 2006). To compare the spokespersons' behaviour across votes, we interact this variable with the variable 'own committee' in model 3. For investigating the conditional hypothesis 4, we dichotomized the votes by the respective documents into legislative (bills and amendments) and non-legislative matters (various kinds of motions and resolutions), following Stecker (2015). In order to test H5, we interacted committee membership with issue salience for the MP's party. We draw on Manifesto Project data (Lehmann et al., 2022) which quantifies the percentage of 'quasi-sentences' a party dedicates to one of several dozens of issues in its election programme. Its focus on relative issue emphasis aligns with the assumptions of saliency theory (Budge, 2015) and thus renders it the most suitable data source for our purposes. For 'issue salience', we summed up all manifesto items that are linked to the committee's jurisdiction (irrespective of the party's position on the issue). For the main analysis, we accept the huge differences among the salience scores since they represent the overall importance of the topics for the parties. Another option, which we pursue in the robustness checks, is to z-standardize the salience scores among the parties. The result would indicate then which party emphasizes a topic to an above-or below-average extentirrespective of whether it is generally an important topic. The salience variables were coded 'missing' if no manifesto items fit to the committee's jurisdiction. This reduces the number of observations in model 5 compared to models 1-4.
In order to isolate the hypothesized effects, we control for other factors that affect MPs' defection rates. Most of them are approximations of the varying effectiveness of party disciplinesince the national-level party leadership is the major principal for progressively ambitious MPs (Sieberer and Müller, 2017). To capture electoral incentives, we include whether an MP was elected in the nominal or list tier of Germany's mixed-member electoral system (Sieberer, 2010). Especially, list candidates are punished by party selectorates for dissenting votes through less promising list positions (Schmuck and Hohendorf, 2022), whereas MPs, as individual candidates, tend to profit from vote defections, especially if justified with representing voter concerns (Duell et al., 2023). We also include dichotomous variables measuring whether an MP holds an executive office (chancellor, minister, junior minister) or another important parliamentary office (party group leadership, whip, Bundestag presidency, committee chair 1 ) to account for higher disciplinary pressures (Zittel and Nyhuis, 2019). Moreover, we consider parliamentary experience which was regarded as a proxy for MPs' parliamentary socialization (Delius et al., 2013), policy 1 Whereas committee chairs are said to hold the more prestigious office, compared to policy spokespersons, they do not exert any comparable influence on policies (Petersen and Kaina, 2007). Sieberer and Höhmann (2017) report one of the lowest power scores for German committee chairs, compared to other Western democracies. Therefore, their office is only included in the controls instead of being equated with spokespersons.
influence (Tavits, 2009), policy convergence with one's party group (Saalfeld, 1995) or expertise within 'their' policy area (von Oertzen, 2006). MPs' gender and age are also included especially since recent studies detect, under particular circumstances, lower defection rates for women MPs (Clayton and Zetterberg, 2021;Dingler and Ramstetter, 2021). Finally, parties differ in their defection ratesa pattern which is also reflected in our data. Whereas the Christian Democrats show an average defection rate of 0.58%, the Green MPs cast 2.86% of their votes against the party line (Appendix A2, figure 3). Those inter-party differences were attributed to, among others, party group size (Bergmann et al., 2016), candidacy eligibility criteria (Rehmert, 2020), intra-party democracy (Close et al., 2019), party organizational strength (Tavits, 2012) or party family (Close, 2018). Since only one term is analysed and thus sufficient variance is lacking, we include partyfixed effects to account for possible party differences.
Our dichotomous dependent variable takes value 1 whenever an MP's voting behaviour differs from the majority of their party group. We do not differentiate between different degrees of dissent (i.e., between voting 'abstention' or 'no' if the party votes 'yes') since, in both mentioned cases, the MP does not ultimately support the party line. If an MP does not attend a vote, we code this as a missing observation in the main analysis. Although some absences might be strategically motivated in order not to dissent visibly from the party line (Font, 2020), they probably mostly have non-political reasons (e.g., time constraints, illness). Due to the dichotomousness of our dependent variable and a data structure that is clustered into multiple voting occasions of MPs, we estimate logistic panel regressions with standard errors clustered by MP.
Details on the measurement, data sources and descriptive statistics of all variables are listed in Appendix A1 (Table 1). Distributional graphs show that the defection rates vary strongly both among MPs and votes (Appendix A2, figures 12). Whereas about 50% of the MPs and 22% of the votes do not exhibit a single defection, some MPs have a much higher propensity to defect (up to 27%). Among the 20 votes with the highest defection rates (up to 10%), there are numerous decisions on military missions, but also on measures against the pandemic and other domestic policies (Appendix A2, Table 3).

Results
To evaluate our hypotheses, we estimated five regression models (Table 1). According to model 1, MPs defect less frequently if policy matters are concerned which they dealt with during their committee and party group work. This effect is statistically highly significant and supports the policy-shaping hypothesis (H1). However, this does not imply that MPs do not take any cues from policy specialists on other topics. Actually, it shows that the effect of shaping a motion within their issue specialization or following the position of their own working group outweighs the cue-taking effect on other matters if tested against each other.
According to model 2, not only the current but also former memberships in a committee related to the vote affect MPs' voting behaviour. Both exert a statistically significant negative effect on casting a dissenting vote. Whereas the coefficient is even larger for past membership, the statistical significance of the current membership is higher. This supports our former specialist hypothesis (H2).
In the third model, the defection probability is compared both between policy spokespersons and ordinary committee members as well as between subjects within/beyond MPs' issue specialization. The interaction term is not statistically significant at conventional levels (p = 0.11). However, both constitutive variables, which were also included in the models (Brambor et al., 2006), exert statistically significant effects. To explore its substantial size, we plotted the effect of voting on a topic within one's area of specialization both for all MPs and separately for MPs holding a spokesperson function and those who do not (Fig. 1). Overall, voting on a topic that an MP is familiar with reduces the defection probability by 37%. In addition to the lower baseline level, this pattern is considerably more pronounced for spokespersons: Their defection probability is 65% lower regarding issues which they are responsible for. The confidence intervals do not overlap in this comparison. For MPs without a spokesperson position, the difference between the defection probabilities is 'only' 30%. This substantial difference between spokespersons and ordinary committee members clearly supports H3.
Model 4 tests whether the committee membership effect is conditional on the type of document which is voted on. Figure 2 shows that committee membership exerts an effect on MPs' defection rates only in votes on motions which do not result in direct legal changes. Non-legislative matters include also votes on military missions abroad. Even if they do not have legal consequences for the citizens, they constitute meaningful decisions both for the soldiers' lives and the federal budget. Predicted probabilities of voting against the party line are shown, with 95% confidence intervals displayed by spikes.  Regarding those votes, it has been reported that the leadership of at least some party groups relaxes party discipline for those who cannot support those missions for conscience reasons (Delius et al., 2013). On those non-legislative matters, MPs who are responsible committee members have statistically as well as substantially significant lower defection rates. In contrast, when bills and amendments are put to the vote, committee membership is no longer a statistically significant predictor of MPs' behaviour. In those highly consequential votes, division of labour might be replaced by other pathways to unity, most likely discipline imposed by the party group leadership. A similar conditional effect is found for issue salience. Again, the interaction term is not statistically significant (model 5). However, a marginal effects plot shows that the effect of voting on an issue within the jurisdiction of one's committee indeed varies with issue salience (Fig. 3): The unity-inducing effect of committee membership is strongest for topics which are hardly salient to the MP's party. With increasing salience, the difference in the defection probabilities between committee members and non-members becomes weaker and, ultimately, statistically insignificant. Supporting H5, it appears that the division of labour principle applies mainly to issues of minor importance to the party, whereas parties monitor and guide the parliamentary behaviour of their MPs more closely on issues that are fundamental to the party's identitywhich, in turn, constrains the committee members' ability of shaping policies and thus the major reason for toeing the party line.
Regarding the controls, we see that holding parliamentary and, even more, executive offices goes along with lower defection rateswhich could be interpreted as a discipline effect. Both a longer parliamentary experience and a higher age correlate with a higher probability to defect. Additionally, we see significant party differences: All parties (except for the Social Democrats) show higher defection rates than the Christian Democrats, with the Greens and the right-wing populist AfD showing the highest defection probabilities. In contrast, whether the MP was elected via the district or the list tier and his/her gender do not matter.
Robustness checks show that the results do not depend on the operationalization of certain controls or the treatment of MPs' non-participation in roll-call votes. Additionally, they reveal that the more the measurement of MPs' issue specialization moves from active participation to mere formal consultation, the lower its explanatory power (Appendix A3). Although a correlative analysis cannot ultimately prove it, this substantiates our causal argument that shaping policies and, thus, potentially minimizing the policy distance to the party line is the plausible reason for lower defection rates of committee members.

Conclusion
We investigated how the issue specialization of MPs affects their probability of casting votes against the party line. Like other so-called 'working parliaments' (Lord, 2018), the Bundestag delegates large parts of its work to standing committees. Since MPs are members of only one or two committees, they are concerned with a subset of all policies in detail. Consequently, there might be differences in MPs' voting behaviour between issues which they are specialized on and others which they do not get in touch with before the party group meeting or the plenary vote. Empirically, the analysis of roll-call vote data for the German Bundestag (2017-2021) supports our policy-shaping hypothesis, i.e., that MPs have lower defection rates concerning issues inside their area of specialization, probably because they actively participate in determining what later becomes the party line (H1). Proceeding from this, four refining propositions were deducted that had not been empirically tested before and are now backed by the data: First, MPs who formerly were members of the responsible committee show lower defection rates as well (H2). Second, the negative effect on vote defections is stronger for policy spokespersons than for ordinary committee members since the former have enhanced possibilities to shape the party line (H3). Third, the explanatory power of the policy-shaping hypothesis is lower for important decisions, meaning for votes that have a direct effect on legislation (H4) or that are highly salient to the MP's party (H5). The described effects support the understanding of division of labour as a powerful pathway to party-compliant behaviour, especially for MPs responsible for the given topic. Thus, the findings add to the more prominently featured discipline-and loyalty-related accounts of how parties reach unified action in parliament.
The results have several implications for the relationship between parties, voters and MPs. First, if votes against the party line become more unlikely for MPs who are concerned with a proposal, this might be used strategically by the party group leadership. Instead of denying an MP with diverging views a seat in the respective committee (which would resemble the partisan theory of legislative organization, Cox and McCubbins, 1993), it is also conceivable to dissuade the MP from vote defections by involving him/her in the substantive policy work. This would provide an explanation for Mickler's (2018) finding that policy distance is not a predictor of committee assignments in the German case. However, disciplining by involvement only works for MPs ideologically not too far away from the party mean who occasionally use deviations to raise their own profile. In contrast, the involvement of staunch preference outliers in the committee work could weaken the overall party position more than just one dissenting voice on the floor. Second, the findings add to the literature on issue competition and issue ownership in that defection rates are not equal among all policy issues. The more salient an issue is for a party, the more centralized appears the policy-making process and the more party discipline restricts the policy-shaping possibilities of individual MPs. Third, that the behavioural differences between committee members and other MPs disappear in the highly consequential legislative votes points to the dominance of political parties and the executive in public policy-making within parliamentary systems like the German one. However, although they centralize policy-making on important matters, legislative parties might also take an independent stance towards their government from time to time and our results underlined that MPs are quite capable of shaping the party line on some issues (von Oertzen, 2006). Fourth, our findings could be read as dissenting votes are reduced when the division of labour principle is invigorated. This carries two different implications for the major theoretical debate on representation. On the one hand, reducing votes against the party line mitigates an agency problem within the constitutional chain of delegation (Müller, 2000) and strengthens promissory representation (Mansbridge, 2003), i.e., that MPs support the party on whose platform they were elected to parliament. It would also support the notion that parties increase the transparency and predictability of policy outcomes and are thus the best means of facilitating accountability between voters and their representatives (Kölln, 2015). On the other hand, citizens explicitly prefer constituency representation over partisan representation (Bøggild, 2020), and the division of labour principle has been alleged to impair MPs' responsiveness to their constituents (von Oertzen, 2006). Their specialization on a few issues necessarily results in a lack of overview and knowledge on other issues. However, the inputs MPs receive from their constituency are not limited to their area of specialization. Since it is considered bad style to chime in when other topics are concerned (Schöne, 2010;Ismayr, 2012), MPs' responsiveness is restricted to 'their' topics. It is true that MPs can pass on concerns from the constituency to responsible committee members. However, the chances that these will be taken up are low. Therefore, the insufficient internal responsiveness of the specialized MPs to each other weakens the external responsiveness of parliament to society (von Oertzen, 2006). This problem gains additional weight in times of declining classical cleavages and partisan dealignment, because supporters and members of parties (including MPs) are increasingly less united on policy core beliefs and, at the same time, the parties' policy specialists shield themselves from each other.
Our findings provide several avenues for future research. Because this study focused on the 'division of labour' path to unity, we conceptualized MPs' connection to particular issues by their involvement in the committee and intra-party group deliberations. However, this might not be the only way in which MPs' defection probabilities vary among issues. First, MPs might have an interest in topics not related to their committee membership, be that based on their former occupation or personal curiositywithout them joining the respective committee or despite being denied the assignment to those committees. Second, MPs might also connect to particular topics based on earlier stages of their political career, be that executive offices at the federal or subnational level, committee memberships in other parliaments or engagement in NGOs. Future studies ought to shed light on those possible advances or replicate our findings for further parliaments.