Mapping Cabinet Conflicts and Conflict Features

• 1 Introduction

  Conflict is inherent to coalition politics and both descriptive and explanatory analyses of cabinet conflicts are on the rise (Andeweg & Timmermans, 2008; Blondel & Müller-Rommel, 1993; Marangoni & Vercesi, 2015; Moury & Timmermans, 2013; Nousiainen, 1993; Timmermans & Moury, 2006). However, this emerging field is marked by several lacunae. First, existing (operational) definitions of cabinet conflict fail to discern them from mere disagreements. Indeed, this very distinction is often ignored and some studies use indicators of disagreement (e.g. the range of party positions) to operationalise cabinet conflicts (e.g. Klüver & Bäck, 2019). But in coalition politics, disagreement between partners is abundant. Such different views and stances can but need not lead to clashes. Conflict does not refer to disagreement as such, but to the way disagreement is handled (i.e. the behaviour of those who disagree). Second, empirical analyses are scarce and coding choices are often unclear or suboptimal: What about cases of doubt? What search strategy is used to select and code sources, for example what keywords are used to search digital archives? Also, many studies only focus on major clashes. One example is the set of 44 conflicts in Dutch and Belgian cabinets (1989-2003) discussed by Timmermans and Moury (2006) (see also Moury & Timmermans, 2013). But when is a conflict ‘major’ enough to be included? And why exclude smaller clashes? Others, such as Marangoni and Vercesi (2015), search digital news archives and succeed in exposing more conflicts (851 Italian cabinet conflicts between 1996 and 2011). However, by using keywords that refer to clashes directly (e.g. ‘conflict’, ‘struggle’), such existing analyses are blind to the many conflicts that are not explicitly labelled as such. Third, the intensity of conflict is usually grasped by using ‘risk factors’ as proxies. For instance, divergent preferences are riskier than tangential preferences and interparty conflict outweighs interdepartmental conflict (Andeweg & Timmermans, 2008; Marangoni & Vercesi, 2015; Nousiainen, 1993). But proxies are no direct indicators, and using them comes with avoidable distortion (e.g. not all interparty conflicts are more intense than interdepartmental conflicts).
  Building on and adding to these previous works, this research note wishes to address these lacunae. It presents a novel approach to defining and measuring cabinet conflicts. These are defined as explicit and antagonistic disagreements between cabinet members and/or relevant coalition party actors. This analysis uses a detailed codebook and provides an intercoder reliability test (cf. infra). Conflict intensities are discerned directly, and several variables keep track of doubtful cases. The approach presented here is suitable for comparative applications on a wide range of sources (newspapers, digital news archives, TV or radio shows, etc.). I also discuss one such application, the result of which is the most fine-grained dataset on the frequency and features of Belgian cabinet conflicts to date (N = 1,090; 1995-2018).¹x Following an embargo, the data will be available on 1 November 2023 in Mendeley Data (Vandenberghe, M. (2022). Cabinet conflicts in Belgium (1979-2018) - Extended version. In: Mendeley Data. doi: 10.17632/zvyt86jffd.1). Meanwhile, data can be made available by the corresponding author, upon reasonable request. Specific attention goes to ethno-territorial conflicts. They shaped the course of Belgium’s history, and empirical studies on the subject are on the rise (e.g. Caluwaerts & Reuchamps, 2015; De Winter & Baudewyns, 2009; Deschouwer, 2006; Hooghe, 2004). For instance, the duration of cabinet negotiations and the number of failed cabinet formation attempts have been shown to increase when ethno-territorial issues are on the table (De Winter & Dumont, 2014). Similarly, we know that most fatal cabinet conflicts between 1946 and 1999 were ethno-territorial in nature (Dumont et al., 2001). These insights are valuable, but only expose the tip of the iceberg. What lacks is a profound and systematic view of the prevalence and evolution of such conflicts. In addressing this gap, small clashes and minor tensions deserve our attention too. They are indicative of the general political atmosphere and the cooperative or confrontational nature of everyday politics. As consociational theory posits (Lijphart, 1969, 1977, 2002), it is exactly this ‘mundane’ kind of day-to-day elite cooperation that is key to understanding the stability of divided states.
  I proceed as follows. The first section addresses cabinet conflict’s definition and measurement, while conflict variables are discussed in the second section. Illustrating the potential of the new approach described here, the third section presents some first results on conflicts in Belgium (1995-2018).

• 2 Defining and Mapping Cabinet Conflict

  2.1 Defining Cabinet Conflict

  Cabinet conflict has been defined as “any situation in which cabinet members differ in opinion, preference, interest or activity” (Appendix II in Blondel & Müller-Rommel, 1993, p. 316) or as “any quarrel or explicit disagreement between two or more executive members and/or coalition (individual or collective) party actors” (Marangoni & Vercesi, 2015, p. 21). In practice, disagreement must of course be externally visible for conflicts to be mapped (e.g. reported in media sources). But not all disagreements are conflicts. Coalition partners disagree constantly, but in many cases, different views never turn ugly. A third criterion is needed. Disagreement must be ‘antagonistic’ or ‘hostile’, that is the pathway of accommodative politics must be left. This shift from constructive to confrontational politics can manifest itself in many ways: forms of swearing, blocking, threatening and so forth (cf. infra). Also, not all coalition party actors are relevant. For instance, is an individual backbencher’s frustrated rant really a ‘cabinet conflict’? A ‘relevance criterion’ excludes such ballast: those involved must be cabinet members or represent a coalition party at large (e.g. president, PPG leader). When this is certainly not the case, situations are ignored. Cases of doubt are included and coded as such (cf. infra). In sum, cabinet conflicts are defined as any explicit and antagonistic disagreement between two or more cabinet members and/or relevant coalition party actors.

  2.2 Mapping Cabinet Conflict

  A fine-grained codebook (available on demand) was used to identify new conflicts and code conflict variables. It includes both general rules and detailed guidelines (e.g. ‘code according to the nub of the conflict’, ‘when hesitating, opt for the least extreme category’, ‘code as a conflict when term X is used’). To guarantee coding consistency and transparency, a logbook was used. The main coding rationales are discussed here. To begin with, I inductively constructed six conflict indicators to decide whether disagreements are ‘antagonistic’ (Table 1). Note that combinations are possible (e.g. threatening with resignation).
  To find and code cabinet conflicts, most studies use (amongst others) quantitative content analyses of printed newspapers or keyword searches in digital newspaper archives (e.g. Blondel & Müller-Rommel, 1993; Marangoni & Vercesi, 2015; Moury & Timmermans, 2013; Timmermans & Moury, 2006). The codebook used here can easily be applied to a wide range of sources, including TV or radio archives, printed newspapers, magazines and political yearbooks. My own exploration of sources taught me some noteworthy lessons. Importantly, unless a very large team of coders is used, it is not feasible to read and code decades of newspapers when the focus is on all articles on all pages of every newspaper of every single day. Longitudinal analyses of printed newspapers only seem to be feasible when coders focus on samples of certain journals, issues (days in the week), pages and perhaps even articles (based on their titles). My own attempts to find Belgian cabinet conflicts this way only exposed the major conflicts but provided no consistent view of smaller tensions (grumpy sneers, short-lived criticism, etc.). Indeed, reading full journals (as a test) showed that the sample approach only exposed some of these small tensions, while many clashes of equal magnitude remained hidden. Numerous other sources were also prone to the feasibility problem, or were manageably short but failed to report most small clashes.²x I explored magazines like Knack, parliamentary records, EJPR Political Data Yearbooks and political or news shows on TV and the radio (using the archives of the Flemish public broadcaster, VRT).

  Table 1 Conflict indicators and intensity levels

  Indicator	Description	Intensity level
  Verbal	Pejorative criticism, swearing, scapegoating, anger, etc.	0
  Block	Manifest blocking (attempts) (restrictive coding): empty chair tactics, sharp veto against X despite urgency, etc.	1
  Impose	Imposing or trying to impose X on partner (restrictive coding): faits accomplis, sharp ultimatum for X, etc.	1
  Threat	Threatening partner (whatever the threat). Any reference to future retaliations against the behaviour of partner(s) (if they do X, we will Y)	1 (2 if survival)
  Branding	Situation is labelled a conflict (or crisis, tensions, escalation, etc.) by coalition partner(s) or third actor(s)	1 (2 if ‘crisis’)
  Survival	Resignation of the cabinet, a coalition party or a cabinet member is in the balance, openly questioned (incl. threats), or effectively occurring	2
  Other	Residual category (logbook)	case-specific

  Details: codebook (available on demand)

  The first results described below are based on a coding effort of digital news archives (GoPress). Similar archives are available in most countries and typically cover many years if not decades. Hence, the usage of digital news archives is a fruitful avenue for comparative and longitudinal reflections. It also avoids the feasibility problem that characterises for example printed newspaper analysis (where no prior keyword selection can downsize the sample of articles). I searched the full Dutch selection of Belga, the largest and most authoritative Belgian news agency.³x Except for the period between 15 May 1997 and 31 December 2000 (French selection, as no Dutch Belga titles were available). Therefore, the selection includes 1,213 French articles. For several reasons, news agency articles are preferred over journal or magazine articles. First, most incidents are covered by various journals or magazines, resulting in a large selection of similar articles without added value. Second, using digital news archives avoids the issue of having to choose a selection of specific journals or magazines (to keep coding feasible). Third, news agency articles include all articles and news releases of all days (which is a major asset). Finally, as compared to journals and magazines, news agencies such as Belga also use more straightforward headlines, which further facilitates data collection.
  Media sources are well suited for mapping and coding conflicts. It is widely known that the media focuses on conflict in political news (cf. Lengauer et al., 2012) and uses strategy and game-frames (cf. Aalberg et al., 2012). Using media sources thus ensures that few conflicts slip through the net. Rather, a significant bycatch was thrown overboard after checking for conflict indicators myself. I am aware of the debates on the potentially growing media focus on conflict (e.g. Vliegenthart et al., 2011), which could distort the data. To some extent, this problem is unavoidable. All sources can be prone to such trends – including statements by politicians themselves (perhaps, their focus on conflict increases too). The only view we have is an indirect one. There is no ‘direct’ source exposing cabinet clashes. We are not flies on the cabinet’s walls. Although it serves as a relevant nuance, the potential presence of blind spots should not withhold us from studying what is visible. Also, in this respect, the first findings presented below are rather comforting (cf. infra). They show no general increase in conflict levels since 1995, but strong fluctuations. Roughly speaking, conflict was on the rise for years, but this trend did not persist. The most recent years covered by the data (2017 and 2018) even show the lowest conflict levels in more than a decade (except for 2011). If anything, a growing media focus on conflicts would imply that this apparent decline is even sharper in reality.
  Having discussed the issue of sources, I now turn to the equally important issue of sample selection (using keywords). Previous studies use keywords referring to cabinet conflicts directly, such as ‘contrast’, ‘conflict’, ‘disagreement’, ‘struggle’ or ‘against’ (Marangoni & Vercesi, 2015, n. 11). One problem with this approach is that it is hard to establish an exhaustive list of relevant concepts. A second and related problem is that it leaves us blind to the many clashes that are not explicitly labelled as such. Not all articles on conflicts actually mention such terms. To avoid missing conflicts, a more thorough approach was used. First, a keyword search with Boolean operators finds all articles on a given coalition (Table 2). To exclude irrelevant articles, this search had to apply to articles’ texts and titles. In a second phase, I read all headlines of the resulting selection of tens of thousands of articles. Articles were selected when their headline included conflict indicators, pleas against something (e.g. “Proposal X faces criticism”) or potential rebuttals against sneers (e.g. “Proposal X is not unrealistic, says PM”). The only excluded articles are the ones whose titles fail to indicate any kind of disagreement and articles that are certainly not about the coalition partners. In this phase, and to avoid missing conflicts, being overly inclusive was preferred to being overly restrictive. The gold was sieved from the resulting selection in a third phase, when the 9,547 resulting articles were fully read to find and code conflicts.

• 3 Conflict Variables

  Conflict variables are listed in Table 3. Each conflict’s starting date, cabinet and source information is provided, next to a detailed description (355 words on average). Using the conflict indicators, three levels of intensity are discerned (cf. Table 1). Dummy variables grasp whether conflicts are between actors of the same party and identify cases of doubt.

  Table 2 GoPress keyword search

  Boolean operators
  Seven distinct references to (Belgian) cabinets (separated by OR), all cabinet members (ministers, secretaries of state, prime minister) (separated by OR), all names of coalition parties (separated by OR) and all names of coalition party presidents (separated by OR), excluding (NOT) nine irrelevant recurring article titles (separated by OR). For a cabinet with n parties and m cabinet members (ministers and secretaries of state): ("reference to cabinet 1"OR"reference to cabinet 2"OR(…)"reference to cabinet 7"OR"name party 1"OR"name party 2"OR(…)"name party n"OR"name PM"OR"name cabinet member 1"OR"name cabinet member 2"OR(…)"name cabinet member m")NOT("recurring article title 1"OR"recurring article title 2"OR(…)"recurring article title 9")
  Example: Leterme I
  ("meerderheid"OR"regering"OR"federale regering"OR"Belgische regering"OR"federale coalitie"OR"Belgische coalitie"OR"wetstraat"OR"PS"OR"MR"OR"VLD"OR"Open Vld"OR"CD&V"OR"CDH"OR"FDF"OR"Leterme"OR"Reynders"OR"Onkelinx"OR"Dewael"OR"Vandeurzen"OR"Milquet"OR"Vervotte"OR"De Gucht"OR"Magnette"OR"Laruelle"OR"De Crem"OR"Arena"OR"Turtelboom"OR"Van Quickenborne"OR"Michel"OR"Laloux"OR"Delizee"OR"Delizée"OR"Wathelet"OR"Chastel"OR"Devlies"OR"Schouppe"OR"Fernandez"OR"Clerfayt"OR"Maingain"OR"Reynders"OR"Di Rupo"OR"Beke"OR"Thyssen"OR"Somers")NOT("persselectie"OR"krantentitels"OR"program of the day"OR"BELGA CORRECTION"OR"1LEAD"OR"Bilan de l’actualité"OR"titres journeaux"OR"Revue sélective de la presse"OR"Actualité internationale pour la semaine")

  Table 3 Cabinet conflict in Belgium: variables (N = 1,090; 1995-2018)

  Variable	N	%	Value(s)
  ID			Identification number
  Cabinet			Cabinet name
  Year			Starting year
  mm/yy start			Starting month and year
  Topic+summary			Topic and description
  Article first			Date of first article on conflict
  Intensity	855	78.4	0 (low)
  	149	13.7	1 (intermediate)
  	85	7.8	2 (high)
  	0	0	T (doubt/missing)
  Copa	1,013	92.9	0 (no doubt: conflict is between coalition partners)
  	77	7.1	T (doubt/missing)
  Copa.doubt+non-pol	1,043	95.7	0 (no doubt: conflict is between political actors)
  	47	4.3	T (doubt/missing)
  Same_PP	1,028	94.3	0 (not solely between actors of same party)
  	13	1.2	1 (solely between actors of same party)
  	49	4.5	T (doubt/missing)
  Solo_doubt	893	81.9	0 (no doubt: conflict meets relevance criterion)
  	197	18.1	T (doubt/missing)
  ET_issue	919	84.3	0 (no ethno-territorial issue)
  	169	15.5	1 (ethno-territorial issue)
  	2	0.2	T (doubt/missing)
  ET_issue_ detail	919	84.3	0 (no ethno-territorial issue)
  	45	4.1	1 (linguistic struggle; e.g. language law)
  	39	3.6	2 (state reform; not fiscal/financial aspects)
  	3	0.3	3 (fiscal/financial aspects of state reform)
  	44	4.0	4 (distribution of assets and liabilities; e.g. allocation key for doctor contingents, distribution of EU funds)
  	11	1.0	5 (combinations)
  	27	2.5	6 (other)
  	2	0.2	T (doubt/missing)
  SGS2.0	1,013	92.9	0 (no full segmental sides/no info)
  	57	5.2	1 (one full segmental side)
  	15	1.4	2 (two full segmental sides)
  	5	0.5	6 (segmental frame, unverifiable)
  	0	0	T (doubt/missing)
  SGS_detail	245	22.5	0 (mixed sides)
  	123	11.3	1 (intra-Francophone)
  	260	23.9	2 (intra-Flemish)
  	295	27.1	3 (partial segmental sides; e.g. one of the Flemish against one of the Francophone parties)
  	43	3.9	4 (full Francophone vs. partial Flemish side; e.g. all Francophone parties against one Flemish minister)
  	14	1.3	5 (partial Francophone vs. full Flemish side)
  	15	1.4	6 (two full segmental sides)
  	95	8.7	T (doubt/missing)

  Four variables grasp conflicts’ ethno-territorial nature. Building on the concept of segmental cleavages presented by Eckstein (1966, p. 34), ethno-territorial conflicts are defined here as conflicts between segmental sides (Flemings vs. Francophones) and/or on ethno-territorial issues (e.g. language policy, state reform). To identify such issues, I use the codebook of the Belgian Agendas Project (BAP) which is customised to fit the Belgian context (cf. Walgrave et al., 2019).⁴x Original data collection: Walgrave, Joly, Hardy, Zicha, Sevenans and Van Assche. Funding: European Science Foundation (07-ECRP-008), Flemish National Science Foundation (G.0117.11N), Belgian Federal Science Policy (IUAP P7/46). The BAP is part of the Comparative Agendas Project (CAP) (Baumgartner et al., 2019), which collects data on the ‘issue attention’ of different institutions and actors (newspapers, parties, etc.). It is not preoccupied with conflicts itself, but the categorisation of issues it presents is useful for operationalising ethno-territorial issues. Specifically, three codes are combined, two of which are amended to fit this study: intergovernmental relations (e.g. transfers, decentralisation; code 2001), state reform and constitution (2033) and the promotion and defence of national culture (e.g. language law; 2311). When clearly linked to these codes, conflicts are also coded as clashes on ethno-territorial issues (e.g. clashes on the appointment of a minister due to his/her legacy of language law violations). Ethno-territorial issues are mapped on two levels: with a dummy variable (ET_issue: 0/1) and with a detailed variable that distinguishes four subsets of issues: language struggle, state reform, fiscal/financial issues and the distribution of assets and liabilities (ET_issue_detail).
  Segmental sides are linguistically homogeneous and encompassing. So, all Flemish and/or all Francophone coalition partners must stand oppose: e.g. when all Francophone parties attack one Flemish minister (one segmental side) or when all Flemish parties clash with all Francophone parties (two segmental sides). A general variable (SGS2.0) discerns 0, 1 or 2 segmental sides and grasps when conflict was described by the news agency in segmental terms that could not be verified (e.g. “proposal X fiercely criticised in Flanders”). A second variable (SGS_detail) grasps intra-Flemish and intra-Francophone conflicts as well as variations of intersegmental conflicts. To be considered an ethno-territorial conflict, “ET_issue” must be 1 and/or “SGS2.0” must be 1 or 2.⁵x I acknowledge the media’s tendency to depict segments as homogeneous, conflicting players, and their tendency to present actions of individuals/parties as those of ‘Flanders’ or ‘Francophone Belgium’ as a whole (Sinardet, 2008, 2012). But these trends do not really distort the data since, with very few exceptions, I didn’t have to rely on media frames. The articles usually provided sufficient info on the specific parties or actors involved. As its exceptional composition would cause distortion, an exception was made for the Michel I cabinet (which included three Flemish parties but only one Francophone party: MR, Mouvement Réformateur). Here, two segmental sides are needed to be considered a conflict along segmental lines (one side does not suffice). This avoids any clash between the MR and a coalition partner from being considered an ethno-territorial conflict.
  An intercoder reliability test measures the degree of resemblance (Cohen’s κ) between the original coding and that of an external coder with layman’s knowledge of Belgian politics (Appendix 1). After an extensive briefing on the codebook and a day of practice, all variables of 32 randomly selected conflicts were coded. All κ values indicate excellent (κ > 0.8) or good (κ > 0.7) intercoder reliability ratings (p < 0.001), except for the ethno-territorial issue variable (fair agreement; κ = 0.434; p < 0.05).

• 4 First Results: Conflicts in Belgium (1995-2018)

  Using the method described above allowed me to expose 1,090 conflicts between 1995 and 2018.⁶x The reported results include cases of doubt. Most conflicts are not intense (78.4%). Intermediate intensity characterises 13.7% of cases, while the cabinet’s survival or composition is at stake in 7.8% of clashes. Conflict between same party actors is scarce (1.2%).
  Of all conflicts, 16.5% is ethno-territorial in nature (N = 180). Ethno-territorial issues triggered 169 cabinet conflicts (15.5%). Most of these concerned linguistic issues (4.1%), state reform (3.6%) or the distribution of assets/liabilities (4.0%). Strikingly, at the cabinet level, fiscal/financial issues such as interregional transfers are hardly explosive. Of the 1,090 conflicts I found, only 3 revolved around such debates (0.3%). Conflicts on ethno-territorial issues are almost two and a half times as frequent as conflicts along segmental lines (when the Flemings and/or the Francophones are mobilised as a block; 6.6%). In most of these clashes, only one segmental side is involved (5.2%). Recall that for the Michel I cabinet, such clashes are not considered as ethno-territorial conflicts. Often, some of the Flemish partners (e.g. one party) clash with some of the Francophones (e.g. one minister) (27.1%). Intra-segmental conflict is also frequent, although more amongst Flemings (23.9%) than amongst Francophones (11.3%).
  What about the evolution of conflicts? Strong fluctuations can be noted. Conflict levels were generally on the rise for years and peaked around 2008. This is clear both at the yearly (Figure 1) and cabinet level (Figure 2). What follows is a blurry picture. Figure 1 also depicts the absolute number of ethno-territorial conflicts, which shows no linear trend. Furthermore, assessing intensity levels shows that intense conflicts were clearly more frequent in the past. Also, and confirming its reputation as a ‘bickering cabinet’, Michel I was subject to an unprecedented number of small sneers and minor clashes (peaking in 2016). This coalition ended with a fatal clash (in December 2018) but in terms of conflict frequency, its final year is the calmest year of coalition governance since the late 1990s (except for 2011, when the caretaker cabinet could work in the shadows of the formation negotiations). This indicates that raw conflict frequency need not predict a cabinet’s survival chances.
  Figure 2 shows the yearly equivalent of cabinet conflicts for each cabinet (number of conflicts divided by cabinet duration in days, multiplied by 365).⁷x Including the periods in which the coalitions were caretaker cabinets. Dehaene I not included (only its last months are covered by the data). Michel I cabinet: end date equals resignation date, as this marks the end of the coding effort. This allows for a solid comparison of cabinets with highly divergent lifespans (ranging from three months to over four years). Here, the consecutive rise and decline of conflict levels is even clearer.

  Yearly Cabinet Conflict Frequency by Intensity and Ethno-territorial Nature (1995-2018) (N=1,090)
  
  Cabinet Conflict Frequency: Yearly Equivalent by Cabinet (1995-2018)
  

• 5 Conclusion

  This research note presents a novel approach to defining, mapping and coding cabinet conflicts. Illustrating its potential, it also presents some first results of an application to Belgium (N = 1,090; 1995-2018). Doing so might add to the field in several ways. First and importantly, the underlying methodology is easily applicable to other sources, periods, policy levels and countries. Second, assessing the relation between dataset variables and/or external variables opens many doors, including that of re-examinations of explanatory theories of cabinet conflict. What is the role of cabinet composition, coalition agreements, external factors, and so forth (for factors, cf. Bergman et al., 2008)? Why are some periods characterised by intense conflicts while others are not? Similarly, the data provide an impetus for debates on ethno-territorial tensions. Are such conflicts more intense than other clashes? What factors explain their prevalence? Do the Flemish and the Francophones stand increasingly opposed? Apart from pleasing academics, answering such questions would provide a more solid factual basis for public debates on the functioning and future of Belgium and other such cases. Perhaps, this is its greatest contribution.

• Funding

  This work was supported by the FWO (Fonds Wetenschappelijk Onderzoek – Vlaanderen) under project number 3F009819.

References

• Aalberg, T., Strömbäck, J. & de Vreese, C.H. (2012). The Framing of Politics as Strategy and Game: A Review of Concepts, Operationalizations and key Findings. Journalism, 13(2), 162-178. doi: 10.1177/1464884911427799.

• Andeweg, R.B. & Timmermans, A. (2008). Conflict Management in Coalition Government. In K. Strøm, W. C. Müller & T. Bergman (Eds.), Cabinets and Coalition Bargaining: The Democratic Life Cycle in Western Europe (pp. 269-300). Oxford University Press.

• Baumgartner, F.R., Breunig, C. & Grossman, E. (Eds.) (2019). Comparative Policy Agendas: Theory, Tools, Data. Oxford University Press.

• Bergman, T., Gerber, E.R., Kastner, S. & Nyblade, B. (2008). The Empirical Study of Cabinet Governance. In K. Strøm, W. C. Müller & T. Bergman (Eds.), Cabinets and Coalition Bargaining: The Democratic Life Cycle in Western Europe (pp. 85-122). Oxford University Press.

• Blondel, J. & Müller-Rommel, F. (Eds.) (1993). Governing Together: The Extent and Limits of Joint Decision-Making in Western European Cabinets. Palgrave Macmillan.

• Caluwaerts, D. & Reuchamps, M. (2015). Combining Federalism with Consociationalism: Is Belgian Consociational Federalism Digging its Own Grave?. Ethnopolitics, 14(3), 277-295. doi: 10.1080/17449057.2014.986866.

• De Winter, L. & Baudewyns, P. (2009). Belgium: Towards the Breakdown of a Nation-State in the Heart of Europe?. Nationalism and Ethnic Politics, 15(3-4), 280-304. doi: 10.1080/13537110903358689.

• De Winter, L. & Dumont, P. (2014). Regeringsvorming in België: Mission Impossible?. In C. Devos (Ed.), België #2014: Een politieke geschiedenis van morgen (pp. 241-271). Borgerhoff & Lamberigts.

• Deschouwer, K. (2006). And the Peace Goes On? Consociational Democracy and Belgian Politics in the Twenty-First Century. West European Politics, 29(5), 895-911. doi: 10.1080/01402380600968760.

• Dumont, P., De Winter, L. & Dandoy, R. (2001). Démissions gouvernementales et performances électorales des majorités sortantes (1946-1999). Courrier hebdomadaire du CRISP, 1722(17), 5-51. doi: 10.3917/cris.1722.0005.

• Eckstein, H. (1966). Division and Cohesion in Democracy: A Study of Norway. Princeton University Press.

• Hooghe, L. (2004). Belgium: Hollowing the Center. In U. M. Amoretti & N. Bermeo (Eds.), Federalism and Territorial Cleavages (pp. 55-92). Johns Hopkins University Press.

• Klüver, H. & Bäck, H. (2019). Coalition Agreements, Issue Attention, and Cabinet Governance. Comparative Political Studies, 52(13-14), 1995-2031. doi: 10.1177/0010414019830726.

• Lengauer, G., Esser, F. & Berganza, R. (2012). Negativity in Political News: A Review of Concepts, Operationalizations and Key Findings. Journalism, 13(2), 179-202. doi: 10.1177/1464884911427800.

• Lijphart, A. (1969). Consociational Democracy. World Politics, 21(2), 207-225. doi: 10.2307/2009820.

• Lijphart, A. (1977). Democracy in Plural Societies: A Comparative Exploration. Yale University Press.

• Lijphart, A. (2002). The Wave of Power-Sharing Democracy. In A. Reynolds (Ed.), The Architecture of Democracy: Constitutional Design, Conflict Management, and Democracy (pp. 37-54). Oxford University Press.

• Marangoni, F. & Vercesi, M. (2015). The Government and Its Hard Decisions: How Conflict is Managed Within the Coalition. In N. Conti & F. Marangoni (Eds.), The Challenge of Coalition Government: The Italian Case (pp. 17-35). Routledge.

• Moury, C. & Timmermans, A. (2013). Inter-Party Conflict Management in Coalition Governments: Analyzing the Role of Coalition Agreements in Belgium, Germany, Italy and the Netherlands. Politics and Governance, 1(2), 117-131. doi: 10.17645/pag.v1i2.94.

• Nousiainen, J. (1993). Decision-Making, Policy Content and Conflict Resolution in Western European Cabinets. In J. Blondel & F. Müller-Rommel (Eds.), Governing Together: The Extent and Limits of Joint Decision-Making in Western European Cabinets (pp. 259-282). MacMillan.

• Sinardet, D. (2008). Vlaamse en Franstalige media over Franstaligen en Vlamingen Wederzijdse representaties van de ‘andere’ in politieke debatprogramma’s. In H. Van den Bulck & A. Dhoest (Eds.), Media, Cultuur, Identiteit: Actueel onderzoek naar media en maatschappij (pp. 51-69). Academia Press.

• Sinardet, D. (2012). Is There a Belgian Public Sphere? What the Case of a Federal Multilingual Country Can Contribute to the Debate on Transnational Public Spheres, and Vice Versa. In M. Seymour & A.-G. Gagnon (Eds.), Multinational Federalism: Problems and Prospects (pp. 172-202). Palgrave Macmillan.

• Timmermans, A. & Moury, C. (2006). Coalition Governance in Belgium and The Netherlands: Rising Government Stability Against All Electoral Odds. Acta Politica, 41, pp. 389-407. doi: 10.1057/palgrave.ap.5500139.

• Vliegenthart, R., Boomgaarden, H.G. & Boumans, J.W. (2011). Changes in Political News Coverage: Personalization, Conflict and Negativity in British and Dutch Newspapers. In K. Brants & K. Voltmer (Eds.), Political Communication in Postmodern Democracy (pp. 92-110). Palgrave Macmillan.

• Walgrave, S., Joly, J. & Sevenans, J. (2019). The Belgian Agendas Project. In F. R. Baumgartner, C. Breunig & E. Grossman (Eds.), Comparative Policy Agendas: Theory, Tools, Data (pp. 57-66) Oxford University Press.

Appendix 1 Intercoder reliability: Cohen’s κ by variable (N = 32)

Variable	Cohen’s κa
Intensity	0.770
Coalition parties (doubt)	1.000b
Same party conflict	1.000
Solo doubt (relevance criterion)	0.834
Ethno-territorial issue (binary)	0.434c
Ethno-territorial issue (detailed)	0.887d
Segmental sides (general: SGS2.0)	0.724
Segmental sides (detailed: SGS_detail)	0.824

^a p < 0.001 (all variables except^c)
^b No results (variable was a constant in the selection of cases for this test).
^c p < 0.05
^d Original coding done by the person who was ‘coder 2’ for the other variables (and vice versa).

Noten

• 1 Following an embargo, the data will be available on 1 November 2023 in Mendeley Data (Vandenberghe, M. (2022). Cabinet conflicts in Belgium (1979-2018) - Extended version. In: Mendeley Data. doi: 10.17632/zvyt86jffd.1). Meanwhile, data can be made available by the corresponding author, upon reasonable request.

• 2 I explored magazines like Knack, parliamentary records, EJPR Political Data Yearbooks and political or news shows on TV and the radio (using the archives of the Flemish public broadcaster, VRT).

• 3 Except for the period between 15 May 1997 and 31 December 2000 (French selection, as no Dutch Belga titles were available). Therefore, the selection includes 1,213 French articles.

• 4 Original data collection: Walgrave, Joly, Hardy, Zicha, Sevenans and Van Assche. Funding: European Science Foundation (07-ECRP-008), Flemish National Science Foundation (G.0117.11N), Belgian Federal Science Policy (IUAP P7/46).

• 5 I acknowledge the media’s tendency to depict segments as homogeneous, conflicting players, and their tendency to present actions of individuals/parties as those of ‘Flanders’ or ‘Francophone Belgium’ as a whole (Sinardet, 2008, 2012). But these trends do not really distort the data since, with very few exceptions, I didn’t have to rely on media frames. The articles usually provided sufficient info on the specific parties or actors involved.

• 6 The reported results include cases of doubt.

• 7 Including the periods in which the coalitions were caretaker cabinets. Dehaene I not included (only its last months are covered by the data). Michel I cabinet: end date equals resignation date, as this marks the end of the coding effort.