Keeping tabs through collaboration? Sharing ministerial responsibility in coalition governments

Abstract

Moving past the conventional focus on ministerial portfolios, this paper investigates how coalition governments allocate and share ministerial responsibility for individual policy issues. Sharing responsibility induces coalescing parties to collaborate on policy issues, which addresses the problem of ministerial autonomy. Consequently, I argue that incumbent parties in coalition governments share ministerial responsibility for contentious and salient policy issues. This claim is corroborated based on a newly elicited dataset of over 30,000 ministerial policy responsibilities from Denmark, Germany, and the Netherlands. The findings have important implications for scholarship on coalition governments, as they demonstrate that incumbent parties can use the design of ministerial portfolios itself to insulate a coalition compromise from partisan deviations.

In the German chancellery, a junior minister once complained that “in many cases, there was no complete clarity about the precise jurisdictions of the ministries” (Hoffmann, 2003, 28). Although just anecdotal, this complaint is at odds with the stylized standard notion of coalition governance, which rests on a clear and exclusive allocation of ministerial competences. From a ministerial government perspective, the executive's structure and design shape a government's policy legacy. As the task of policy design is commissioned to the responsible ministry—and thus indirectly the party furnishing the respective minister—the allocation of ministerial portfolios to parties is crucial and formative for the design of future coalition policies (Laver and Shepsle, 1996). Hence, any sort of coalition agreement is essentially void, because, once in power, parties cannot enforce those parts of the deal that fall outside their portfolios. This corollary sparked a line of research on how parties in coalitions can nevertheless “keep tabs on partners” (Thies, 2001). These studies unraveled a toolkit of instruments coalitions can use to ensure the enactment of policy deals. Within the executive, parties can deploy junior ministers in “hostile ministries” to curb the informational advantage of the minister in charge (Thies, 2001). Farther down the line of legislation, parliamentary amendment rights (Martin and Vanberg, 2004, 2005, 2011) and scrutinizing parliamentary committees (Kim and Loewenberg, 2005; Carroll and Cox, 2012; Fortunato et al., 2017) can be effective in reeling back in politically biased bills toward the initially agreed compromise.

These commonly researched ex-post controls of ministerial discretion are largely reactive. Instruments of parliamentary review can only be used to scrutinize and amend bills after the government has already set their general course. Junior ministers ring the alarm bells after they have spotted their bosses going native. Even hierarchy within the government, be it the PM or any sort of “inner cabinet,” is mostly used to react to policy initiatives drafted by the responsible minister. In contrast, the perspective presented here is qualitatively different from other coalition governance tools, as it is not built upon monitoring ministerial policy proposals. Rather, it contents that coalitions motivate and exploit inter-ministerial coordination during the phase of policy formulation to curb ministerial discretion. In other words, incumbent parties in coalition governments do not need to rely on ex-post policing of policy bargains but, drawing on shared responsibility, can directly coordinate among each other while policy initiatives are being drafted by the ministerial bureaucracy. Hence, whenever incumbent parties disagree on salient policy issues, responsibility ought to span across ministries run by different parties.

To this end, the paper explicitly breaks with the assumption of exclusive ministerial portfolios, which denotes the idea that ministers do not interfere in their colleagues’ affairs (Andeweg, 2000). This assumption both underpins the work by Laver and Shepsle (1996) and characterizes the broader theoretical mindset of the ensuing streak of coalition governance research. However, as this paper shows, the empirical reality of portfolio design, that is, the mapping of policy responsibilities to ministries and, thus, incumbent parties, is considerably untidier than usually assumed. Parties in coalitions can exercise joint ministerial responsibility for policy issues, which manifests itself in overlapping portfolios and, hence, often extensive ministerial cooperation. Addressing this idea, this paper echoes previous work on the strategic assignment of neighboring ministerial portfolios (Fernandes et al., 2016); however, it goes considerably further by not just assuming overlapping policy responsibilities between defined pairs of ministerial portfolios, but directly modeling the extent to which ministerial responsibility for policy issues transgresses party boundaries within coalition governments.

The theoretical claim is tested on newly elicited data about ministerial policy responsibilities of five Danish, four German, and seven Dutch governments, covering from 1995 to 2013. Based on charts of the organizational structure of individual ministries, a sequential data collection pipeline consisting of optical character recognition, a supervised classifier, human review, and final automated consistency checks was developed to both extract and content-code about 30,000 policy responsibilities assigned to ministerial working units at the lowest hierarchical level according to the Comparative Agenda Project's codebook (Baumgartner et al., 2019; Bevan, 2019). The resulting dataset shows how policy responsibility is distributed across ministries, and therefore also parties, in coalition governments. The analysis largely supports the theoretical expectation and suggests that incumbent parties in the analyzed coalition governments share responsibility for highly salient, contentious policy issues to curb ministerial discretion.

1. Curbing ministerial discretion through ministerial coordination

Ministerial discretion describes the potential for ministers to deviate from the government's preferences. This can be particularly hurtful to coalition governments, as incumbent parties can (ab)use the ministerial bureaucracy to advocate their own policy ideas, regardless of whether they conform with any sort of policy compromise struck at the onset of a coalition. In the absence of coalition management devices, ministers are powerful gatekeepers with the ability to control the flow of information between their own civil servants and the government as such. Unable to interfere with or monitor the internal procedures of single departments, the cabinet relies on the expertise the minister in charge presents to it. Thus, ministers can rather easily bias information or withhold it from the cabinet to nudge the final policy outcome toward their own or their party's preferences (Timmermans, 1994; Laver and Shepsle, 1996).

The idea of ministerial discretion essentially relies on the assumption that “ministerial jurisdictions are mutually exclusive” (Andeweg, 2000, 378). Although country experts provide some illustrative evidence throughout the numerous case studies in Laver and Shepsle's main contributions to this topic (1994; 1996), showing that governments claim to allocate policy responsibilities tidily among their ministries, this perspective does not pay due credit to the abundant episodes of inter-ministerial coordination reported in many governments. For instance, the Danish Jørgensen III cabinet (1978–1979) and many Austrian governments explicitly team up ministers from different parties and require them to reach an agreement before decisions can be made (Christensen, 1985, 131; Müller, 1997, 135) and Swedish key portfolios are often allocated to different cabinet ministers to keep parties at bay (Bergman, 1997, 274). In the Westminster system, the idea of dividing ministerial responsibility is well documented as a tool of strengthening the position of the prime minister (Dewan and Hortala-Vallve, 2011). Harold Wilson's incoming Labour government of 1964 is a case-in-point. Having won the party leadership, prevailing against his inner-party rivals George Brown and James Callaghan, the prime minister was suspicious about them plotting a coup within his government. Hence, he deliberately recast Brown's and Callaghan's departments—economic affairs and the treasury—to stimulate tensions between the two contenders and thereby deflect their attention from mobilizing against him (Heppell, 2011).

From a theoretical perspective, overlapping responsibility among ministries creates (semi-)formalized venues or incentive structures, which encourage ministers to consider their colleagues’ preferences or even collaborate during the process of designing legislation. Formalized venues explicitly speak to instances of positive coordination, that is, situations where actors seek solutions to policy problems while simultaneously trying to resolve distributive problems (Scharpf and Mohr, 1994, 18). In their study on the German federal bureaucracy, Mayntz and Scharpf (1975) frequently identified this mode of coordination, whenever solutions were to be found for problems that transgressed the jurisdictions of ministerial portfolios (see also: Wegrich and Hammerschmid, 2018). The working groups that were established in response to these problems were tasked to both provide innovative and effective solutions, while at the same time protecting the interests of their home ministries and the responsible party.¹ In Denmark, the government regularly leaves the precise coordination of policy proposals between ministries to two standing committees, chaired by permanent secretaries from the PM's office and the Ministry of Finance (Greve, 2018). These bodies—the coordination committee and the economy committee—seek solutions to important policy problems that crosscut ministerial boundaries and coordinate the day-to-day business between the members of government. The existence of these committees results in frequent inter-ministerial coordination and cooperation that has transformed decision-making into a reconciliatory process, in which contentious issues must be resolved among those ministries upon whose stakes they touch (Bo Smith-udvalget, 2015, 85–86).

Whereas positive coordination was found to be the preferred mode of inter-ministerial cooperation (Mayntz and Scharpf, 1975), for lower-profile cases governments rather opt for a cheaper alternative. Ministries can use negative coordination, which resembles a form of voluntary self-restraint, to avoid negative externalities a decision might induce (Scharpf and Mohr, 1994). Similar to the idea of deferential adjustment (Lindblom, 1965, 45), it describes a situation in which decision-makers seek to avoid inducing a response from other stakeholders (read: other affected ministries) by avoiding impinging on their values and preferences. Cabinets often deliberately amplify this behavior by being unwilling to act upon ministerial initiatives that stir up conflict (Mayntz and Scharpf, 1975, 145–50). This form of cooperation only requires minimal to no active coordination among the involved ministries, as initiatives are pre-emptively designed to be mutually acceptable, which makes it the most prevalent form of inter-ministerial coordination.

Both forms of inter-ministerial coordination lend themselves to curbing ministerial discretion, and, hence, coalitions can use them as a governance device. However, this requires a slight change of perspective. As coordination has been described so far, it speaks to diverging preferences among ministries, while coalition governance is generally concerned with how ministerial discretion can be exploited to implement preferences of incumbent parties (e.g., Thies, 2001; Martin and Vanberg, 2005, 2011). In order to turn inter-ministerial coordination into a coalition governance device, the focus needs to move from interests of single ministries to preferences of parties represented by the ministries they direct. From this perspective, speaking of inter-ministerial coordination refers to employing coordination tools between ministries to sort out disagreements among the coalescing parties.

If coalition governments want to use either form of inter-ministerial coordination to rein in ministers, they must ensure that potentially contentious legislation indeed touches upon the jurisdiction of ministries that belong to different incumbent parties. Put differently, ministerial jurisdictions must overlap in policy areas that are contentious within the coalition. Effectively, this implies that incumbent parties must share responsibility for such policy issues. If this is given, parties can use coordination mechanisms to either jointly cooperate on prospective legislation (positive coordination) or have the ministries in charge voluntarily exercise self-restraint (negative coordination). In both cases, the suggested policy that emanates from the ministerial bureaucracy mirrors the coalition's agreed policy objectives much closer than if it had been drafted unilaterally by one ministry or different ministries belonging to just one party. This governance tool is qualitatively different from ex-post controls that parties can use to ensure the enactment of the coalition compromise (Thies, 2001; Martin and Vanberg, 2004, 2005, 2011; Carroll and Cox, 2012; Fortunato et al., 2017). Instead of actively obtaining policy expertise to level the informational advantage of ministers and thereby facilitating more detailed scrutiny of policy proposals, involved parties can rely on negative or positive coordination to jointly collaborate on contested policies that are designed inside the executive.

Yet, coalitions may not always prefer shared ministerial responsibility to curb ministerial autonomy. First, as the purposeful sharing of ministerial responsibility requires a certain degree of redundancy in the administrative structure, this coalition governance mechanism entails comparatively high fixed costs. Second, given the transaction costs that positive inter-ministerial coordination inflicts upon the involved parties (Mayntz and Scharpf, 1975), it is conceivable that, on balance, amending legislation within parliament (Martin and Vanberg, 2005, 2011) is cheaper and, hence, preferable to having differences being sorted out within the executive in a ping-pong-like procedure. Furthermore, relying on parliamentary review shifts the burden of coalition management from the executive to the legislative branch of government. Since the same party elites who often form a coalition, design its portfolios, and negotiate its policy agreement are regularly rewarded with a ministerial office, it might be that they rather have their fellow partisans in parliament taking care of curbing ministerial dominance. Given these considerations, the benefits of shared responsibility—direct collaboration on policy proposals and the containment of biased proposals before they even see the light of day—may only outweigh the costs for highly salient issues.

In terms of the empirical implications that follow from this line of thought, it is important to note that the argument put forward here only stipulates that a coalition's desire to curb ministerial discretion motivates the allocation of issue responsibility across the ministries of different incumbent parties. More precisely, this implies that governments who deem a policy issue salient advocate curbing ministerial autonomy via ministerial collaboration in the context of policy conflict. Translated into language of effects, the argument thus predicts

1. (1) a positive multiplicative effect of the conflict around and the salience of a policy area within a coalition government; and

2. (2) a positive effect of policy conflict on the extent to which responsibility is shared among incumbent parties for salient policy issues.

2. Data on ministerial responsibility

To investigate the rationale set out before, this paper combines existing, content-coded information on party manifestos with a novel dataset on ministerial jurisdictions. Although already called for at the onset of the new millennium (Thies, 2001), a lacuna of data on the various tasks and functions ministries are charged with has persisted. As a result, researchers often still seek to infer ministerial jurisdictions from the corresponding ministerial name tag. This article takes a different approach and looks at the precise set of tasks national ministries perform as outlined in official publications on the ministerial bureaucracy.

2.1 Sample selection

The theoretical conjectures are tested on a sample consisting of five Danish, seven Dutch, and four German governments, jointly spanning the time from 1995 to 2013. The reason for the country selection is threefold. First, albeit in different formats, these countries’ governments regularly publish information on the structure of their ministries and the tasks each of their working units is commissioned to perform. While Germany and the Netherlands release detailed organization charts of their ministries, the Danish government issues an annual report on the public sector that, analogously to organization charts, describes the structure of ministries and the policy tasks of several divisions and subdivisions therein.²

Second, all three countries have a long steady history of coalition governments, which is a prerequisite for testing the proposed deliberate use of overlapping ministerial responsibilities in multiparty governments. Since 1980, both Germany and the Netherlands have only been governed by coalitions of at least two parties. In contrast, Denmark has seen three single-party governments ever since: Anker Jørgensen's sixth cabinet (1981–1982), and the more recent social-democratic single-party cabinets led by Helle Thorning-Schmidt (2014–2015) as well as Mette Frederiksen's current government, which has been in office since 2019. However, none of those Danish single-party governments falls within the timeframe of this study, which is roughly from the mid-1990s to 2010.

Lastly, Denmark, Germany, and the Netherlands represent three unlikely cases to find deliberate use of shared ministerial responsibilities. All three countries have strong parliaments that are equipped with numerous means to gather relevant information, monitor governments, and amend bills during the legislative process if deemed necessary (Lijphart, 1999; Martin and Vanberg, 2011). Hence, parties in all countries do not have to rely on the executive's responsibility structure for curbing ministerial discretion, but also have plenty of tools available farther down the line of legislation. Thus, finding politically motivated overlapping ministerial jurisdictions in these unlikely cases makes it probable that governments in countries with weaker parliaments may do so, too. At the same time and regardless of this similarity in parliamentary policing power, the three sampled countries are sufficiently distinct in terms of their executive arenas as to exclusively trace back instances of ministerial coordination to bureaucratic peculiarities. For example, Denmark distinguishes itself from both Germany and the Netherlands in the way it structures its ministries: all three countries outsource more mundane and administrative tasks to a set of public agencies attached to each ministry (Meer, 2018; Wegrich and Hammerschmid, 2018); however, only Danish governments have come to actively embrace the idea of an “agency structure” of ministerial organization, which limits the size of the actual ministry to often as few as 150 employees, while outsourcing a considerable portion of the daily business including decision-making authority to semi-independent agencies. For example, in 2016 a total of 18 Danish ministries and 131 agencies resided in the central government (Greve, 2018).

2.2 Coding organization charts

While ministries can also change their organizational structure during a government's term, the overwhelming majority of ministerial re-designs occurs during the formation of a new government (Sieberer et al., 2021). Hence, each ministry was only coded once per government, drawing on the first document published six months after the inauguration of a new government. Besides rendering the data collection manageable, this selection of cases ensures a focus on the political aspects of ministerial re-designs, in contrast to changes later in a government's incumbency, which might rather be driven by administrative concerns. The structural differences between Danish and German ministries feedback into the actual object of study. While it suffices to focus on the actual ministerial department in Germany and the Netherlands, the Danish agency model requires a broader perspective. Thus, the Danish data collection does not just draw on the actual ministerial department but understands a ministry as the larger ministerial conglomerate comprised of the department and its affiliated agencies. The final data span each ministry in five Danish, four German, and seven Dutch governments between the mid-1990s and roughly 2010.

Within the ministry, the units of data elicitation are the individual working units on the lowest hierarchical level as described in the organizational structure.³ These working units are conventionally called “enhed” or “kontor” in Denmark, “Referat” in Germany, and “bureau” or “directie” in the Netherlands. The brief and concise description of policy responsibilities that accompanies each working unit forms the basis for the subsequent quantitative content analysis.

For each observed working unit, all identified policy tasks are mapped to a set of policy content codes as defined by the Comparative Agenda Project's (CAP)⁴ codebook. The CAP studies how political actors deal with policy issues across numerous political systems. The central interest lies in investigating the rise and fall of policy issues on different political agendas (Baumgartner and Jones, 1993; Baumgartner et al., 2019). For this purpose, the project developed a coding scheme of about 230 distinct policy issues, which define mutually exclusive and exhaustive categories that are time-consistent.⁵ The national adaptations of this scheme, which are fully compatible with each other (Bevan, 2019), are used to classify the set of policy tasks each office performs.

The actual coding process was carried out by three student coders and the author, all of whom were trained in and had ample competences with the CAP coding scheme. Moreover, all coders were either native speakers of or completely fluent in the language the organization charts were written in. For each policy department, based on a random sample of ministries from each country, the coders first decided whether the unit was merely in charge of administrative tasks, such as HR or budgeting. These departments were flagged as such and not content coded. For units that were considered in charge of substantial policy content, the coders determined how many distinct policy tasks they perform, usually based on conjunctions or appropriate punctuation. As policy areas are often split up or consolidated in the bureaucracy (Hösl et al., 2016), simply assuming that all departments are solely responsible for one policy task could severely misrepresent the set of responsibilities each ministry bears. On average, a Danish working unit attends to 3.8 distinct policy issues, which is twice the number of issues their German counterparts deal with. Lastly, each identified policy task was coded according to the national version of the CAP coding scheme. The inter-coder agreement amounts to about 87 percent at the level of major policy codes and 80 percent for minor policy codes, that are clustered within the major categories. Based on this hand-coded sample, a supervised classifier was trained and subsequently applied to the remaining ministries. Importantly, the classifier's predictions were meant to help the coders achieve internal consistency during the coding process, meaning that the coders reviewed and potentially adapted the computationally assigned content codes.

To check for internal consistence of the data, several sanity checks were applied to the coded data. First, a computer program used the task descriptions to identify working units that are substantially similar across all ministries and governments within each country. Having identified similar working units, the program checked whether all had been equally flagged as either administrative or substantial and, subsequently, whether their policy description had been subdivided into the same number of codes. Inconsistencies were reported to the author, who decided authoritatively to make corrections where necessary. Thereafter, the program crawled through all individual policy tasks and identified those cases where different CAP content codes had been assigned to linguistically similar policy task descriptions. Again, all inconsistencies were reported to the author, who subsequently made changes where necessary. Given the very specific and brief language of policy descriptions in organization charts, these computational sanity checks ensure the internal consistency of the data within each country sample. The final dataset comprises more than 30,000 coded policy tasks.

2.3 Describing policy responsibility during Angela Merkel's first grand coalition

As the argument put forward focuses on how coalition parties can use the ministerial bureaucracy to keep tabs on their partners in government, it disregards the potential for inter-ministerial coordination of ministries belonging to the same incumbent party. Hence, in cases where multiple ministries of one party attend to the same policy issue, those ministries are collapsed under the party label. To this end, the data have been aggregated to the level of policy issue—party dyads, informing about the ministerial responsibility each governing party assumes for each policy issue. Mathematically, responsibility is defined as the frequency of policy issue codes per party divided by the total frequency of policy issue codes within a coalition, that is,

$$Resp_{i, j} = {n_{i, j}\over \sum_{\,j = 1}^{m}n_{i, j}}$$

where n _i,j denotes the frequency of codes for an issue i and party j, with m being the number of parties per government. These responsibility scores can be used to discriminate between policy issues that are either exclusively addressed by ministries of one political party or jointly dealt with ministries of different parties. Taking Angela Merkel's first coalition government (2005–2009) as an example, Figure 1 plots the partisan location of ministerial responsibility. Since there were only two parties in this government—the Social Democrats (SPD) and the Christian-Conservatives (CDU/CSU)—issue responsibilities for both parties can be plotted on one dimension by calculating the difference between both parties’ issue responsibilities. This score is Resp _CDU/CSU − Resp _SPD = 1 − 0 = 1 if all responsibility rests with the conservatives and Resp _CDU/CSU − Resp _SPD = 0 − 1 = −1 when a policy issue is addressed solely by the social-democrats’ ministries.

Fig. 1. Distribution of responsibility for CAP policy issues between the Conservatives (CDU/CSU) and the Social-Democrats (SPD) in Angela Merkel's first government, 2005–2009. Each panel shows one policy area. Horizontal lines represent detailed policy issues within these policy areas. Definition of policy areas and detailed policy issues follows the CAP codebook. Circles on lines denote partisan location of ministerial responsibility. Horizontal lines without circles symbolize policy issues to which the government's ministries did not attend. Blank spaces denote policy issues not defined by the CAP codebook.

The visualization strategy of Figure 1 closely follows the CAP's definition of policy issues, which groups detailed policy issues within more encompassing policy areas. Each panel shows one of these policy areas—ranging from Macroeconomy, over the Environment to Government Operations—that the CAP has defined for Germany. The horizontal lines within each panel symbolize detailed policy issues within policy areas. For example, within the area of Civil Rights, 01 represents “minority discrimination,” 02 stands for “gender discrimination,” and 06 shows the policy issue of “voting rights.” Note that, due to the iterative process by which the CAP codebook was developed, most minor policy codes are not being used in many policy areas, which is why there are many blank spaces. The circles on the horizontal lines inform about the partisan location of ministerial responsibility for a given policy issue.⁶ For example, the CDU/CSU—holding the defense ministry in the Merkel I government—controlled almost the entire area of defense policies. Except for defense alliance policies (02), nuclear disarmament (05), and military aid (06), responsibility for all remaining defense issues rest with the conservatives and does not transgress party boundaries. In contrast, the Social-Democrats were clearly in charge of health and transportation policies. In between those two poles, there are many policy issues for which parties exercise shared responsibility. Examples of such policy can be found in the areas of environment, energy, and foreign trade.

Especially for broader issue categories, the operationalization of responsibility can induce measurement error. It is possible that parties only appear to share responsibility prima facie, although the clusters are vague enough to allow for a neat ministerial allocation of responsibilities within them. However, the detailed nature of the CAP coding scheme, which comprises about 230 distinct policy issues, counterbalances these tendencies and ensures that the measure should generally pick up on real overlaps within narrow categories. A German piece of consumer protection legislation intended to limit unwanted advertising and sales calls serves as a case in point of responsibility overlap and interparty cooperation.

During the first half of Angela Merkel's first Grand Coalition, Cold calls developed into a real nuisance in Germany. Within one year, the number of such calls rose by about 30 percent (Graw and Ehrenstein, 2007), which, according to pollsters, annoyed about 90 percent of the German population (Leins, 2007; Süddeutsche, 2007). Even though the German government at that time had a CDU/CSU-led ministry dedicated to matters of consumer protection, it was the social-democratic Ministry of Justice that took the lead and in 2007 announced legislative action. Yet, due to the high salience of the issue and diverging objectives regarding the scope of the amendment, it was intensively discussed among parties (Die Welt, 2008; Sigmund, 2009) and, hence, among the initiating Ministry of Justice and the two conservative ministries of consumer protection and economy, postponing its adoption by one year from mid-2008 to mid-2009 (Stuttgarter, 2008). When the bill eventually was signed into effect by the Federal President, it also bore the signatures of the three ministers that had been involved with its drafting. This example of positive cooperation resulting from overlapping responsibility is clearly reflected in the data, which lists 13 mentions of the corresponding issue category (15–25) in the entire government. Of this number, six belonged to social-democratic and seven to Christian-democratic ministries, which yields policy responsibility scores of 0.46 and 0.54, respectively.

3. Methodological matters

The simple measure of responsibility distance introduced earlier to visualize the distribution of ministerial responsibilities in Angela Merkel's first government can be converted to a more general measure of the extent to which two incumbent parties “share” responsibility for a policy issue. Taking the absolute value of the calculated difference yields a figure ranging on the closed unit interval, where low values indicate that both parties possess equal amounts of ministerial responsibility for a policy issue, whereas higher values hint at situations in which there is one party dominating policymaking within the executive via its ministries. To facilitate interpretability, this score is subtracted from 1, effectively swapping the extreme points of the unit interval and, hence, creating a measure of responsibility sharedness, where larger values denote a higher degree of responsibility sharing between coalescing parties.

This procedure is fine for coalition governments composed of two parties, where it yields exactly one measurement of sharedness per policy issue. Yet, it fails to capture the numerous interparty relations of governments comprised of more than just two parties, as it is often the case in the Netherlands. Therefore, the measurement is generalized to capture the extent of responsibility sharedness per policy-issue on the level of party-dyads, not entire governments. While this does not effectively change the procedure for two-party coalitions—there is still only one measurement per government—the generalized procedure yields one measurement for each two-combination of incumbent parties in situations where more than two parties form a government. Hence, the unit of analysis is the party dyad within each government-issue. Mathematically, sharedness is defined as

$$Sharedness_{a, ij} = 1 - \vert Resp_{a, i} - Resp_{a, j}\vert$$

with a representing a policy and i,  j denoting the observed party-dyad within a coalition government. A brief example from the Netherlands showcases how this measure works in practice. In Wim Kok's second coalition government (1998–2002) two of three parties shared responsibility for industrial policy. About one-sixth of the government's resources addressing this issue were located in the Ministry for Spatial Planning, controlled by the Labour Party (PvdA), and the remaining five-sixths were bound in the Ministry for Spacial Planning, led by a minister from the People's Part for Freedom and Democracy (VVD). The third incumbent party, the Democrats 66 (D66), did not bear any responsibility for industrial policy. Consequently, the responsibility sharedness scores of the three party-dyads calculate as 1 − |0.167 − 0.833| = 0.333 between the PvdA and the VVD, 1 − |0.167 − 0| = 0.833 between the PvdA and the D66, and 1 − |0.833 − 0| = 0.167 between the VVD and the D66.

This measurement approach yields a total of 3919 observations of responsibility sharedness across the three sampled countries, which are summarized in Figure 2. It reveals that particularly Danish and German governments regularly create or accept ministerial jurisdictions sharing responsibility for policy issues. However, in all three countries, the mode and the majority of observations clusters around zero, indicating that responsibility for most policy issues is not spread out, but rather concentrated within ministries belonging to a single party. Hence, if interparty responsibility sharing is used strategically, it does not happen on a large scale, but governments generally strive to consolidate responsibility within ministries of one incumbent party.

Fig. 2. Distribution of responsibility sharedness within Denmark, Germany, and the Netherlands.

3.1 Explanatory variables

Presented as operating at the level of single policy issues, a strong test of the effect of policy conflict on responsibility sharedness requires positional or at least directional data on the level of detailed policy issues, as defined by the Comparative Agendas Project. Unfortunately, to the best of my knowledge, data on such a low level is not directly available, but conventional sources like the Chapel Hill Expert Survey (Polk et al., 2017) or the Manifesto Project Dataset (Volkens et al., 2019) provide sufficiently long time-series data only for a few dimensions. Confronted with a similar measurement problem, Klüver and Zubek (2018) develop a mapping of 13 policy areas defined by the Manifesto Project onto the broader issue clusters as defined by the CAP. Within each of these areas, positive and negative mentions by parties can be used to compute an index that yields a party's position on each dimension, following the recommendations by Lowe et al. (2011). From here on, in each of the 13 areas, policy conflict is operationalized as the absolute difference between two parties’ policy positions. The procedure, especially the grouping of Manifesto Project issues to 13 policy areas (Table A6 in the online Appendix) and the mapping of these areas onto the CAP policy codes (Table A7 in the online Appendix), is described in detail in the online Appendix. While still not ideal, using data generated by this process allows studying the effect of ideological conflict within coalition governments on the extent of responsibility sharedness within 13 distinct policy clusters, containing about two-thirds of all policy issues defined by the CAP.

As it is argued that coalitions are more prone to use shared responsibility for highly salient policy issues, there is a need for a measure of issue salience. Drawing on previous work and data collected by the CAP, party manifestos drafted before general elections are used to capture the extent to which policy issues are salient to incumbent parties.⁷ A party's issue salience is operationalized as relative frequencies, that is, the number a policy issue occurs in a manifesto divided by the manifesto's length. The total issue salience per observed party dyad is simply the sum of the individual parties’ salience scores.

The analysis controls for the quantitative distribution of ministries within a coalition. Crucially, the absolute difference between the relative amount of ministries parties in a dyad control should decrease a dyad's propensity to share responsibility for a policy issue. The reason for this is less political, but rather a statistical artifact. More ministries tend to go hand in hand with a broadening of a party's jurisdiction, which makes overlap both more likely to occur and more simply to create. Hence, two parties commanding a similar number of ministries, translating to a small difference, are more likely to observe shared responsibility. Vice versa, a larger difference in the number of ministries per party should decrease the extent to which responsibility is shared within a party dyad. Yet, this reasoning is only true unconditionally for governments composed of two parties. Whenever more than two parties coalesce—as often the case in the Netherlands—a small difference might also result from two similarly small parties being pitted against each other. Accounting for this, the analysis models the effect of ministry difference conditional on the number of ministries a dyad commands as a whole (ministries dyad). Given that governments often change the number of ministries (Verzichelli, 2008), all figures are expressed relative to the total number of ministries of which a government is composed. The necessary data on the composition of national governments was retrieved from the “Party Systems & Governments Observatory” (Bértoa, 2016).⁸

3.2 Estimation strategy

As already apparent from Figure 2, the data on responsibility sharedness both only takes on values in the closed unit interval and is markedly skewed to the right ($\widetilde {\mu _3} = 1.26$). There are two widely used strategies to model such data. The first strategy is to use a beta regression model (Ferrari and Cribari-Neto, 2004; Cribari-Neto and Zeileis, 2010). However, as the beta distribution is not defined for the polar values of zero and one, a “fudge factor” must be applied to squeeze the data into the open unit-interval, which can be described by a beta distribution (Smithson and Verkuilen, 2006). The second approach is known as fractional logistic regression. It makes use of the commonly known logistic regression analysis but applies it to model proportions instead of binary outcomes. As both strategies produce substantially similar results, the article presents the results from the more familiar logistic model, leaving the discussion and presentation of the beta regression model to the online Appendix as a robustness check. Like most models, fractional regression models the mean of the dependent variable y on a set of covariates X. Since y—responsibility sharedness—only takes on values between and including zero and one, it must be ensured that the linear predictor βX is mapped onto the closed unit interval. This is done by using a logit model for the mean of the dependent variable, which can be written as

$$\eqalign{& E\left(y\middle\vert X\right) = \ {{\rm e}^{X\beta}\over 1 + {\rm e}^{X\beta}}\cr & X\beta = \beta_0 + \beta_1 \; \ast \; Conflict + \beta_2 \; \ast \; Salience + \beta_{12} \; \ast \; Conflict \; \ast \; Salience + \gamma Z}$$

where Z represents the model matrix containing information on control variables as well as country dummies. Hence, the estimation strategy generates fixed effects within countries. Moreover, following Abadie et al. (2017), the ensuing analysis treats countries as the primary sampling unit and, hence, reports standard errors that account for potential correlation within them.

4. Results

The results of the fractional logistic regression analysis are reported in Table 1. The first model only includes the control variables plus country fixed effects. Thereto, the second model adds a coefficient capturing the effect of policy conflict, while the third model also estimates the interaction effect between conflict conditional and issue salience. Beginning with a brief discussion of the control variables, a larger absolute difference in allocated offices between parties within a dyad decreases the degree to which responsibility for policy issues is shared. However, as expected, this relation does not hold true unconditionally, but only dyads comprised of parties which jointly command most ministries of a government witness such a reduction in sharedness. In contrast, if an observation pits two smaller or medium-sized parties against each other—for instance, the “The People's Party for Freedom and Democracy” and “The Pim Fortuyn List” in Jan Peter Balkenende's second cabinet that jointly held 53 percent of ministries—the negative effect of ministry difference on sharedness is markedly smaller.

Table 1. Fractional logistic regression models

Cluster-robust standard errors shown in parentheses. Drop in number of observations due to lack of data on policy conflict for some policy areas. Refer to Table A7 in the online Appendix for further details.

Note: *p < 0.1; ^**p < 0.05; ^***p < 0.01.

Centrally, the article claims that parties use the ministerial responsibility structure to counteract diverging policy preferences between the incumbent parties. Yet, as explained in the theoretical argument, this effect cannot be expected to hold unconditional of the salience incumbent parties attach to a policy issue, since the purposeful sharing of ministerial responsibility imposes comparatively high costs on the incumbent parties. Hence, coalitions likely reserve it for policy matters where stakes are high and policy deviations, thus, particularly undesirable. In line with this reasoning, the effect of policy conflict on sharedness estimated in model 2 remains insignificant (β = 0.01; SE = 0.13). Hence, dyads of incumbent parties that disagree on the handling of policy issues are not unconditionally more likely to share ministerial responsibility to rein in ministerial drift. A similar conclusion can be drawn from model 3. The estimated effect of policy conflict in this model is −0.03, but again, is statistically insignificant (SE = 0.14). This means that for policy issues that are not salient at all to any party of an observed dyad, policy conflict per se does not stimulate shared ministerial responsibility.

The positive effect of issue salience shows that at some point, governments must have decided to spread responsibility for salient policy issues across different ministries. Hence, sharing of issue responsibility would also happen “mechanically,” if portfolios were allocated randomly among the incumbent parties, simply because more salient policy issues are more spread out. However, the analysis reveals that this mechanical part is not the entire story, but there is indeed a political twist to it. Model 3 shows that parties are more prone to share ministerial responsibility for contentious policy issues once these become politically salient. The multiplicative effect of policy conflict and issue salience is positive and statistically significant at the 5 percent level (β = 2.85; SE = 1.12). Moreover, while consistently positive and significant, the effect of issue salience shrinks by 50 percent once it is made conditional on policy conflict. This halved effect, conditional on policy consensus, can be interpreted as the mechanical component, that is, the degree of sharedness brought about by issue salience without considering policy conflict. However, the multiplicative effect is indicative of a second, “political” component explaining responsibility sharedness. This second component is driven by policy conflict and shows that incumbent parties consider the risk for policy drift when allocating policy responsibilities among themselves. Hence, incumbent parties in coalition governments tend to share ministerial responsibility for contentious policy issues that are comparatively salient to them, which empirically supports the theoretical conjecture discussed previously.

The contour plot in Figure 3 illustrates this central finding. On the x-axis, it shows the extent to which a policy issue is contentious, which is pitted against the issue's salience (y-axis). The solid contour lines denote the extent to which parties share ministerial responsibility as predicted by model 3, given a Danish government with all non-plotted variables hold constant at their median. If there was no interaction effect, the contours’ inclination would be constant across both dimensions, showing that neither effect changes in response to the other. The rugs along both axes visualize the distribution of both explanatory variables.⁹ Lastly, the two gray bands at the bottom visualize both the 10 percent (dark gray) and 5 percent (light gray) level of statistical insignificance based on a Johnson–Neyman interval (Esarey and Sumner, 2018). Hence, in the area outside the respective bands, the conditional effect of policy conflict on sharedness is statistically significant. In combination with the rugs along the vertical axis, these bands also underline a point made earlier in the theoretical discussion: only highly salient observations fall within the area of significance, which means that parties indeed only react to policy conflict by sharing ministerial responsibility for policy issues that are vital to them.

Fig. 3. Predicted extent of responsibility sharedness conditional on policy conflict and issue salience. Calculated for a Danish government with all non-plotted variables fixed at their median. Based on model 3. Gray bands at the bottom show the area outside which the conditional effect of policy conflict is significant (dark: 10 percent, light: 5 percent).

Reading the plot vertically shows that parties become more willing to share ministerial responsibility for salient policy issues—a tendency that is indeed unconditional, in the sense that the effect remains significant for all observed value of policy conflict. For example, keeping policy conflict fixed at 1, a party dyad is predicted to exhibit 0.27 units of sharedness for policy issues that are fairly salient at 0.05 units. Doubling salience increases sharedness by about 20 percent to 0.33 units. In contrast, reading the same plot horizontally reveals the growing conditional effect of policy conflict. For lower values of issue salience, the contour lines are almost horizontal, which is equivalent to saying that there is no effect of policy conflict on sharedness in this area. Yet, as Figure 3 demonstrates, the left part of the contour lines is being bent upward as issue salience rises, which is the graphical manifestation of a growing conditional effect of policy conflict on the extent to which incumbent parties share ministerial responsibility.

The magnitude of this effect can be more palpably expressed as fitted values, which the contour plot directly visualizes. In terms of responsibility sharing, given a policy issues that is not salient at all, a party dyad that perfectly agrees about the proper handling of the matter (y = 0.23) is essentially indistinguishable from parties with a gulf of 2 conflict units between them (y = 0.22). Yet, the situation changes as policy issues become more salient. Comparing the same two-party dyad in a context of 0.1 units of issue salience shows how policy conflict raises the sharing of ministerial responsibility by almost one half from 0.27 units to 0.39 units. This roughly corresponds to a 0.5 standard deviation change in the response variable that can be traced back to a three standard deviation change in policy conflict.

Since responsibility sharedness’ operationalization is directly related to parties’ individual responsibility scores and, hence, the distribution of ministerial resources regarding policy issues, the results are also illustrative of how resources are allocated to individual parties. For instance, 0.27 units of sharedness—the predicted value in the high salience, low conflict scenario, mean that there is a 1 − 0.27 = 0.63 distance between two parties’ relative responsibility for a policy issue. In a two-party coalition, where such scores must sum up to 1, the underlying ministerial bureaucracy must, therefore, be organized in a way where one party receives 82 percent of all ministerial resources related to a policy issue, while the other is left with the remaining 18 percent. In contrast, if policy conflict was as high as 2 units, these responsibility shares would approach each other to about 70 percent versus 30 percent, which is a substantially more even distribution of ministerial responsibility within a party dyad.

5. Conclusion

Based on novel data of ministerial responsibility for individual policy issues, this paper shows that responsibility for policy issues regularly transgresses partisan boundaries. Only for very specific issues like defense, ministerial responsibility is concentrated within one incumbent party's ministries. Besides making the case that ministerial responsibility for policy issues cannot be assumed to be concentrated and, thus, ministerial portfolios can overlap, this paper asserts that parties share responsibility for salient and contentious policy issues to reduce the potential for ministers to bias policy according to their preferences. As a coalition management tool, this mechanism is conceptually different from those discussed in most previous studies on coalition governance (a notable exception is Fernandes et al., 2016). Whereas institutions like junior ministers (Thies, 2001), legislative review (Martin and Vanberg, 2004, 2005, 2011), and parliamentary committees (Carroll and Cox, 2012; Fortunato et al., 2017) suggest that information is gathered to ex-post amend a ministerial policy proposal, shared ministerial responsibility creates the opportunity for parties to cooperate already during the process of drafting legislation and, hence, ex-ante limit the need for later policing. Using a series of fractional logistic regressions, this reasoning is largely corroborated. The analysis shows that incumbent parties are willing to create the administrative redundancies that underpin shared responsibility only for policy issues that are vital to them.

Importantly, the argument put forward here only stipulates that the observed allocation of issue responsibility across ministries of different parties is conducive of curbing ministerial discretion. In this light, it welcomes the push toward portfolio malleability advocated by the growing body of studies of portfolio design, that is, how and why governments recast their portfolio structures, and acknowledges that portfolios are, empirically speaking, much more malleable than commonly assumed (Fleischer et al., 2018; Sieberer et al., 2021). While generally studied within single countries, like Denmark (Mortensen and Green-Pedersen, 2015), Germany (Sieberer, 2015), or the UK (White and Dunleavy, 2010; Heppell, 2011), Sieberer et al. (2021) employ a comparative cross-national focus and find that frequently changing portfolios are a recurring phenomenon in most European countries. These re-designs suggest that coalitions are not restricted to allocate fixed ministerial portfolios whose jurisdictions may happen to overlap in crucial policy areas but are free to tweak the precise allocation of policy responsibilities to ministerial portfolios. However, the extent to which governments indeed deliberately overhaul ministerial structures to create overlapping portfolios is a different and inherently longitudinal question, which is outside the scope of this paper. The reason is that the data on which the findings draw consist of snapshots of the distribution of ministerial responsibility taken at the onset of coalition governments. Hence, the analysis is blind to potential changes in the ministerial responsibility structure that happen after a coalition government started operations.

Moving forward, it is an open question whether and how the sharing of ministerial responsibility integrates with other, more established tools of coalition management and seek to study how different devices can complement or substitute each other. Given the high costs that positive inter-ministerial coordination inflicts upon the involved parties (Mayntz and Scharpf, 1975), it is conceivable that coalitions in some countries may prefer to resort to different coalition governance mechanisms. For example, at least for the executive, having legislation amended in parliament (Martin and Vanberg, 2005, 2011) may be cheaper and, hence, preferable to having differences being sorted out between ministries in a ping-pong-like procedure. Such reasoning is in line with both the comparatively low number of policy issues in the present sample for which incumbent parties properly share ministerial responsibility and the finding that policy conflict only drives responsibility-sharing if the stakes are high. Furthermore, coalitions might be generally less reliant on shared ministerial responsibility, the more governance tools they have at their disposal. The Netherlands are the textbook example of how coalitions employ hostile junior ministers to control whether “real” ministers remain faithful to the coalition's policy goals (Thies, 2001). This frequent deployment of junior ministers may explain why shared responsibilities are markedly less frequent in the Netherlands as compared to Denmark and Germany.

As mentioned before, all three countries constituting the present sample feature rather powerful parliaments, which is why they are a hard test for the stipulated theoretical expectations. In contrast, countries with fewer available management tools, such as Ireland or Spain, should be more prone to use inter-ministerial coordination and have conflicts resolved directly within the government, as parties cannot rely on ex-post devices to curb ministerial discretion. In the same vein, it is telling that the mechanism of shared ministerial responsibility was first discussed in the context of the British Cameron-Clegg government and the—in comparative perspective—weak House of Commons. In parliamentary systems featuring majoritarian representation, such as the UK, single-party governments may equally use overlapping policy briefs to stifle not inter-, but interparty competition (Dewan and Hortala-Vallve, 2011; Heppell, 2011). However, probing into these arguments, both about the interaction of different control mechanisms and the question of how well the mechanism of shared responsibility travels, can only be achieved by expanding the data collection as to increase the institutional variation beyond the present sample.