What Do We Know about Power Sharing after 50 Years?

The power-sharing literature lacks a review that synthesizes its findings, despite spanning over 50 years since Arend Lijphart published his seminal 1969 article ‘ Consociational Democracy ’ . This review article contributes to the literature by introducing and analysing an original dataset, the Power Sharing Articles Dataset, which extracts data on 23 variables from 373 academic articles published between 1969 and 2018. The power-sharing literature, our analysis shows, has witnessed a boom in publications in the last two decades, more than the average publication rate in the social sciences. This review offers a synthesis of how power sharing is theorized, operationalized and studied. We demonstrate that power sharing has generally positive effects, regardless of institutional set-up, post-conflict transitional character and world region. Furthermore, we highlight structural factors that are mostly associated with the success of power sharing. Finally, the review develops a research agenda to guide future scholarly work on power sharing.

Consociational democracy, which Lijphart later termed power sharing (Lijphart 1985), has turned into 'one of the strongest, widely discussed, and influential research programmes in the field of comparative politics' (Taylor 2009: 1). 1 At first, the term 'power sharing' was used empirically to account for democratic stability in Austria, Belgium, the Netherlands and Switzerland (Lijphart 1977). Later it was prescribed normatively as a solution for all deeply divided societies (Andeweg 2015). Lijphart's work represented a paradigm shift by demonstrating the compatibility between democracy and societal cleavages contrary to the claims that justified authoritarian rule as the only way to ensure stability in divided societies (Bogaards 2006: 119). 2 Over time, power sharing has become a 'dominant conflict-solving approach' (Binningsbø 2013: 89) by going beyond its strict focus on consociational cases. Both the consociational and the broader power-sharing strands of literature, however, have been tightly entangled, particularly in cross-national research (Bochsler and Juon 2021;Hartzell and Hoddie 2015;Kelly 2019;Norris 2008;Strøm et al. 2017). 3 Power sharing has been heavily criticized on normative, theoretical and empirical grounds (Barry 1975;Bogaards 2000;Horowitz 2014;Lustick 1997;Rothchild and Roeder 2005;Spears 2002;van Schendelen 1985). In response, advocates of power sharing further developed the theory to address some of its theoretical weaknesses (Lijphart 1981(Lijphart , 2000O'Leary 2006a, 2006b;Noel 2005). Advocates argue that power sharing is the most viable solution for peace and democracy in divided societies (Bochsler and Juon 2021;Cederman et al. 2017;Hartzell and Hoddie 2020;Keil and McCulloch 2021;McCulloch and McGarry 2017;McEvoy and O'Leary 2013).
The literature lacks a study that synthesizes its findings. Previous attempts at synthesizing the literature, while valuable, are either traditional literature reviews (Andeweg 2000) or based on a small group of studies (Binningsbø 2013). The most ambitious attempt to map the literature (Bogaards et al. 2019) groups 346 articles identified under specific topics over time without synthesizing the findings of the literature. The puzzling record of power sharing particularly calls for such analysis. According to John McGarry (2020: 100), power sharing 'has enjoyed success in some places but not in others, in some places at particular times but not at other times, and with respect to some issues but not all … inconsistent with the views of overly-enthusiastic consociationalists and of their critics, many of whom appear to believe that consociations have nothing to offer'. This review article bridges this gap by synthesizing the findings of the power-sharing literature. For this purpose, we introduce an original dataset, the Power Sharing Articles Dataset (PSAD), which extracts data from 373 academic articles published between 1969 and 2018. 4 We next introduce the dataset. The subsequent section maps the theoretical and empirical foundations of the power-sharing literature. After that, we synthesize the positive and negative effects of power sharing as reported by the articles analysed here. The next section outlines the factors favourable to the success of power sharing. A final section concludes by developing a research agenda for future research.

Introducing the Power Sharing Articles Dataset
In building the PSAD, this review article uses Hilary Arksey and Lisa O'Malley's (2005) widely cited scoping review methodology (Cacchione 2016: 116). Our review is guided by this question: what do we know about power sharing from the academic literature in English published between 1969 and 2018? The research question limits our review both temporally and linguistically. The year 1969 is chosen as the starting year because it corresponds to Lijphart's (1969) article, considered the 'classic statement of consociational theory' (Lijphart 2008: 3). 5 We decided to focus on academic articles and exclude books for practical considerations. 6 There is a trade-off since leaving out some classic texts on power sharing (Lijphart 1977;McGarry and O'Leary 1995) could potentially bias the findings of this review.
Relevant academic articles were identified by searching for consociationalism (and its variants, consociation and consociational) and power sharing (and its variants power-sharing and powersharing) in articles' titles, abstracts and keywords. To ensure maximum coverage of articles, an identical search was done in three major social science databases: Scopus, the International Bibliography of the Social Sciences and Political Science Complete. The search returned 1,133, 1,076 and 962 articles, respectively, giving a total of 3,171 articles. Online Appendix 1 includes the full search terms used.
We shortlisted articles based on five exclusion criteria. First, we excluded 1,439 duplicate articles. Second, we excluded articles with incomplete bibliographic information and book reviews. Third, we excluded articles that are off topic, such as those on authoritarian power sharing or those articles that do not meet our definition of power sharing. We follow O'Leary (2013: 3) in defining power sharing as 'any set of arrangements that prevents one agent, or organized collective agency, from being the winner who holds all critical power, whether temporarily or permanently'. This definition is useful because it goes beyond defining power sharing in terms of its institutional dimensions: grand coalition, proportional representation, cultural autonomy and minority veto (Lijphart 1977). This resonates with the recent literature on power sharing that went beyond its strict consociational focus (Binningsbø 2013). Speaking about power sharing in Africa, Nic Cheeseman argues that, Where political settlements are highly inclusive and permanent, they conform to Lijphart's influential model of consociational democracy in which ethnic diversity is managed by building measures that protect the interests of each community into the foundations of the political system. However, because power-sharing deals in Africa are usually forged amidst insurgency and political crisis, they have typically focused on a more modest agenda: securing a ceasefire, forging agreement on distribution of senior political positions, and scheduling a timetable for fresh elections. (Cheeseman 2011: 339-340) This definition also grounds our understanding of power sharing as an arrangement between social groups within a given society, as originally studied by Lijphart (1969). This warrants our fourth exclusion criterion to exclude articles that are transnational, such as articles that take the European Union as the unit of analysis. 7 Fifth, given our focus on the empirical power-sharing literature, we excluded articles that are theoretical in nature and lack an empirically oriented research question. Applying these five exclusion criteria, the PSAD thus includes a final list of 373 articles.
A team of four researchers and a research team leader carried out the data extraction and analysis. At the project's inception, the research team leader trained the research team as part of a larger team on both the substantive power-sharing literature and on the scoping review methodology. The training included practical exercises on data extraction where extensive feedback was given by the team leader. Data from each article was extracted manually. Included articles in this review were read in full by a designated researcher. Researchers then extracted data from the articles by filling out an online form designed specifically for the data extraction. Two levels of quality control were implemented. First, members of the research team highlighted in a special section of the form when they were unsure about a particular question during data extraction. This was then reviewed by another team member. If the issue remained unresolved, the research team leader took the final decision. Second, the team leader looked at random samples of the extracted data to ensure coding reliability and consistency. While an interannotator record was not kept, we estimate the reliability and consistency of the data collection at around 90-95%. For example, 391 articles were initially shortlisted after applying the exclusion criteria. Of these, 18 articles were later excluded during the data-extraction phase for having been found to meet some exclusion criterion. If we generalize the margin of error from this example to the data-extraction phase, the coding reliability and consistency lie at 95.39%.
For each of the 373 articles, we collected data on 23 variables, including bibliographic information (e.g. article title, author, date and journal), theory and concepts (e.g. research question, terms used to describe power sharing, the definition of power sharing used), research design, measurement and contextual information (e.g. names of countries or cases studied, research strategy, time period and method of analysis, how power sharing is measured and power-sharing institutions studied) and empirical findings (e.g. positive or negative effects of power sharing and favourable factors). The next section maps the theoretical and empirical boundaries of the literature.

The theoretical and empirical dimensions of power sharing
The empirical power-sharing literature has been growing steadily over time, with an exponential increase in the last decade. In the first three decadesbetween 1969 and 1998only 42 (out of 373 articles) were published, which more than doubled in the following decade with 99 articles. The fifth decade (2009-18) witnessed an unprecedented increase, with the publication of 232 articles. Certainly, there has been a parallel increase in the publication of social science articles overall, but the exponential increase in articles on power sharing is much higher, as illustrated in Figure 1. The number of articles published in social sciences in the first decade  stood at 51,982. This number increased to 154,944 for articles published in the third decade between 1989 and 1998 before reaching an all-time high with 437,992 articles published in the fifth decade . In other words, there is a more than fivefold increase in the publication of power-sharing articles between the first and fifth decades compared to only a little less than a threefold increase in social science articles published during that period.
Scholars have been interested in studying the origins of power sharing (107 out of 373 articles) along with its effects on stability and peace, moderation of societal/ ethnic cleavages and democracy (188 articles). Table A2 in the Online Appendix offers an overview of the top five independent and dependent variables in the power-sharing literature. In addition, there have been calls to distinguish between consociationalism and power sharing (Bogaards 2000). However, the literature has been converging on the interchangeable use of 'consociationalism' and 'power sharing' in around half of the studies examined, with the other half evenly distributed between the exclusive use of one of those terms. Table A3 in the Online Appendix includes the exact distribution of terms used to describe power sharing.
To date, Lijphart's definitions are the most cited in the literature on power sharing; 158 out of 373 articles adopt his definitions to describe power sharing. The most common definition of power sharing features the four institutional dimensions: namely, grand coalition, minority veto, proportional representation and segmental autonomy (Lijphart 1977: 25). The second common definition focuses on the role of elites in power sharing, viewed as 'government by elite cartel' (Lijphart 1969: 216). Despite the former 'institutional turn' in the power-sharing literature by focusing more on the role of institutions, the use of this latter definition reflects the belief of scholars that the agency of political elites is important. Table A4 in the Online Appendix presents the frequency of definitions of power sharing in the literature.
The power-sharing literature is certainly more diverse now through the study of cases across the globe (Bogaards et al. 2019). Nevertheless, it is still dominated by studies of cases in Europe and Central Asia (204 articles). Figure 2 presents the most studied cases in the literature. Within Europe, Northern Ireland comes first, followed by Bosnia and Herzegovina (81 and 47 articles, respectively). There is variation in the study of the classic consociational casesnamely Belgium, Switzerland and the Netherlandswhich are covered in 25, 16 and 11 articles, respectively. This mapping of the cases studied in the literature is in and of itself a contribution, given the absence of very basic agreement on the countries studied in the power-sharing literature (Dixon 2018). In the words of Helga Binningsbø (2013: 97), 'it is not quite clear which countries and which time periods are categorized as power sharing cases'.
using qualitative methods and 50 articles adopting quantitative methods Graham et al. 2017;Hartzell and Hoddie 2003). The covered articles also include a few studies using game theory and formal modelling (Tangerås and Lagerlöf 2009;Tridimas 2011;Wantchekon 2000) and a notable study using computer simulations (Lustick et al. 2004). For more information on research strategies, methods, units of analysis and time periods, see Online Appendix 6.
When it comes to operationalizing power sharing, the PSAD documents only 99 articles providing an empirical measure of power sharing. In fact, Lijphart (2002) admitted 20 years ago that operationalizing and measuring power sharing is inherently a daunting task due to the different forms it can take. Our review confirms Lijphart's point. Scholars have not so far been able to operationalize power sharing exactly along the four institutional dimensions: grand coalition, proportional representation, minority veto and segmental autonomy. The most ambitious attempts include operationalizations of the three dimensions (Cammett and Malesky 2012) with other operationalizations along two dimensions such as grand coalition and regional autonomy (Cederman et al. 2017(Cederman et al. , 2018 or grand coalitions and minority veto (Gates et al. 2016;Graham et al. 2017;Strøm et al. 2017). 8 Grand coalition and proportionality are the most studied institutions of power sharing (217 and 216 articles, respectively). Minority veto and cultural autonomy are studied each in 136 and 131 articles, respectively. However, the fact that only one-third of the articles studies cultural autonomy is surprising since it is one of the primary features of power sharing alongside grand coalition (Lijphart 2008: 38-39). Cultural autonomy allows ethnic groups self-rule over issues of 'exclusive concern' (Lijphart 1977: 41). It takes several forms, including federal arrangements and certain rights for religious and linguistic groups as in Belgium or Macedonia (McCulloch 2014). Our findings thus support recent calls that more emphasis should be put on the study of cultural autonomy (McGarry 2017).
To overcome the difficulty of operationalizing the classic institutional features of power sharing, studies show an increasing diversity of approaches. One prominent example is the distinction by Caroline Hartzell and Matthew Hoddie (2003) between political, military, territorial and economic power sharing. It has been adopted widely in cross-national analysis (Jarstad and Nilsson 2018;Joshi and Mason 2011;Mattes and Savun 2009). Other operationalizations of power sharing include the proportional ethnic distribution of party members (Shastri 2005), recognition of minority language (Liu 2011), the representativeness of government or party coalitions (Conley and Dahan 2013;Strasheim and Fjelde 2014), control over fiscal or political decision-making (Brancati and Snyder 2011) and ethnofederalism (Charron 2009). Table A9 in the Online Appendix presents the detailed operationalizations of 40 quantitative studies. The next section examines the positive and negative effects of power sharing.
The positive and negative effects of power sharing According to our analysis, 229 of the 373 articles report positive effects of power sharing while only 45 report negative effects. Nineteen report mixed positive and negative effects and 78 report no effects. As outlined in Table 1, the articles reviewed found power sharing to have a positive effect on stability and peace, democracy or democratization, pacification of cleavages and moderation of social groups. 9 While this generally indicates that power sharing works, it raises an important question: why does it work in some cases but not others?
One argument could be that the positive effects are a function of liberal, not corporate, power-sharing systems. According to Allison McCulloch (2014: 503), 'a corporate consociation entails the constitutional entrenchment of group representation', whereas 'liberal consociationalism avoids constitutionally entrenching group representation by leaving the question of who shares power in the hands of voters'. Liberal power sharing, also referred to as self-determination as compared to corporate pre-determination, allows for flexibility and adaptability to changes in the salience of ethnicity or to fluctuating strengths of ethnic groups (Lijphart 1995). In contrast, corporate power sharing can 'entrench and institutionalize pre-existing, and often conflict-hardened, ethnic identities, thus decreasing the incentives for elites to moderate' (Wolff 2011(Wolff : 1783 Belgium is a classic corporate case given its fixed legislative representation for Flemish and Walloons (60% and 40% respectively) and their equal representation in the cabinet. In contrast, Iraq's liberal power-sharing system has no such fixed representational quotas either in the legislature or in the government. However, there is an unwritten, informal agreement in Iraq according to which the presidency is awarded to a Kurd, the premiership to a Shiite Muslim and the parliament speakership to a Sunni Muslim. In Northern Ireland, a hybrid power-sharing case, Protestant Unionists and Catholic Nationalists do not have fixed shares in government but have mutual veto powers.
In Table 2, we disaggregate the positive and negative effects of power sharing by the corporate, liberal and hybrid systems. The analysis shows that the claimed advantage of liberal power sharing in the literature is to some extent overstated. In other words, power sharing has both positive and negative effects in countries belonging to the various systems. 10 The articles reviewed demonstrate that the corporate power-sharing systems in Burundi, Belgium and South Tyrol are shown to have predominantly positive effects on stability and peace, moderation of ethnic groups and democratization. This is the case particularly during the earlier powersharing stages in Burundi (Vandeginste 2011) and throughout in Belgium and South Tyrol (Wolff 2004). The record, however, is mixed in Bosnia and Herzegovina and in Lebanon. On the one hand, corporate power sharing was reported to have led to peace and stability in both countries (Rice 2017; Rosiny 2015). On the other hand, corporate power sharing was found to lead to political stalemate, exacerbation of ethnic divisions and clientelism along with a low quality of democracy (Hulsey 2010;Salamey and Tabar 2012). The same is also said about liberal power-sharing systems. The articles reviewed suggest that in Malaysia, for example, power sharing has contributed to stability and peace and moderation of social groups (Jarrett 2016) while leading at the same time to instability and institutionalization of ethnic cleavages (Segawa 2015). The picture is not very different under hybrid power sharing. In the articles studying Switzerland, Macedonia and Northern Ireland, power sharing was found to have positive effects on stability and peace, moderation of ethnic groups and democracy (Bochsler and Bousbah 2015;Lyon 2012;McGarry and O'Leary 2016). In Kenya, however, according to the analysed articles, power sharing has a mixed record: stability and peace at the expense of democracy (Cheeseman and Tendi 2010;Koko 2013). This analysis suggests that there might be factors beyond the institutional set-up, be it corporate, liberal or hybrid, that drive the positive and negative effects of power sharing. It might be argued that the positive and negative effects of power sharing are rooted in the distinction between power sharing narrowly defined as consociationalism and the wider post-conflict, transitional notion of power sharing (Binningsbø 2013;Cheeseman 2011). Power sharing, this argument goes, takes many forms and consociationalism is only one of them (Bogaards 2000). The PSAD allows for such disaggregation by analysing the positive and negative effects based on the terms used in the article: power sharing, consociationalism, the interchangeable use of both terms, and consociational power sharing. As illustrated in Table 3, the terms used to describe power sharing do not influence the generally positive effects of power sharing prevalent in the articles reviewed: 41 articles that exclusively use 'consociationalism' report positive effects, compared to 12 articles reporting negative effects and an even lower number (4 articles) reporting mixed effects. A similar pattern is observed when looking at articles that employ power sharing or those where both terms are used interchangeably along with the small number of articles using consociational power sharing.
This finding challenges the argument rooting Iraq's instability, for instance, in its liberal 'consociational light' system (Bogaards 2021b). At the same time, the analysed articles suggest that institutionalizing ethnicity through corporate consociational systems as in Bosnia and Herzegovina is not without its own problems (Bahtić-Kunrath 2011). In sum, the articles covered in this review highlight that transitional power sharing contributes to stopping violence, but cannot rule out its onset or recurrenceas the case of Colombia or generally in Africa shows (Daly 2014;Spears 2013). It is also true that post-conflict, transitional power sharing can in some cases drive corruption: 'since elites within the power sharing arrangement cannot be certain that they will be represented in the post-transitional political order, they have strong incentives to capture as many state resources as possible' (Haass and Ottmann 2017: 69). In sum, the evidence gathered from the reviewed articles suggests that post-conflict, transitional power sharing ensures peace in the short term but has more negative effects in the long run. This should not be interpreted as the triumph of consociationalism. Consociational institutions did not save Lebanon from its deadly civil war between 1975and 1990(Dekmejian 1978Hudson 1997). According to Imad Salamey and Paul Tabar (2012: 510), consociationalism can even lead to 'greater national fragmentation, and, at best, establish authoritarian enclaves whose populist leaders take the state and sectarian constituencies hostage for their opportunistic interests'. In short, consociational arrangements are also not a panacea.
We now move to examine the geographical dimensions of power sharing. Table 4 illustrates the varying positive and negative effects of power sharing across world regions. In North America, the articles reviewed demonstrate that power sharing has exclusively positive effects in Canada (Kennedy 2004). 11 In Europe and Central Asia, power sharing has the second-highest frequency of positive effects in 113 (out of 130) articles, negative effects in 13 articles, mixed effects in 4 articles and no effects in 32 articles. In Africa, argues Andreas Mehler (2009: 470), power sharing 'offer[s] no miraculous solutions to complex crisis situations'. Nevertheless, the analysed articles demonstrate that power sharing has a positive effect in 48 out of 70 articles, mixed effects in 9 articles, negative effects in 6 articles and no effects in 7 articles.
The remaining world regions have a mixed record combining nearly equal positive and negative effects. This, as a result, sheds doubt on any regional dimension of power sharing. That power sharing can succeed across the globe, regardless of country or world region, challenges essentialist connotations of power sharing that Lijphart (1977) himself pushed against. While a political culture of accommodation between social groups is important for power sharing (Lijphart 1968a;Steiner 2009), such culture is not restricted to a particular geographic region. The next section directs attention to the effect of structural factors on the success of power sharing. Source: The authors. Note: Some articles report more than one positive or negative effect and thus are double-counted.

Favourable factors and the success of power sharing
The development of the power-sharing theory has resulted in refining the favourable, structural factors over time. Between 1968 and1985, Lijphart used 14 different favourable factors with only 4 in common (Bogaards 1998: 476-477). But after the mid-1980s, Lijphart (2008: 5) settled on nine favourable factors, with two considered as the most important: the lack of an ethnic or religious majority and limited socioeconomic inequalities among the different groups. The other seven factors include the number of societal groups, same size of groups, small population size, external threats, overarching loyalties (such as nationalism), geographical concentration of groups, and traditions of compromise and accommodation. There is, nevertheless, a lack of theoretical clarity on the empirical relevance of the favourable factors. Lijphart (2008: 5) argued that the 'nine conditions should not be regarded as either necessary or sufficient conditions: an attempt at consociationalism can fail even if all the background conditions are positive, and it is not impossible for it to succeed even if all of these conditions are negative'. As a result, van Schendelen (1985: 160) criticized Lijphart's position, arguing that the favourable factors 'may be present and absent, necessary and unnecessary, in short conditions or not conditions at all'. In the following, we shed some light on the relationship between the favourable factors and power sharing. Table 5 presents the frequency of favourable factors compared to their contribution to the success of power sharing. Three primary observations are in order. First, the two most important factors in Lijphart's viewnamely no majority group and socioeconomic equalityare not among the top factors enabling the success of power sharing, as appears from the analysed articles. Both are reported respectively by 28 and 14 of the articles covered in this review. Majority groups, Lijphart (1969: 217) theorized, tend to dominate rather than cooperate with minorities, while low socioeconomic inequality lowers the grievances of poorer groups and reduces the magnitude of risks perceived by the well-off ones (Lijphart 1985: 124-125). While the data at hand do not entirely negate the relevance of both conditions, it seems plausible to question treating them as superior to the other factors.
Second, it is significant that 79 articles point towards the geographical concentration of groups as the favourable factor associated most with the success of power sharing. Its significance lies in supporting Lijphart's notion that 'good fences make good neighbors' (Lijphart 1977: 140). Such territorial isolation, the argument goes, limits tensions and hostility arising from mutual contact between different groups (Lijphart 1977: 88). This type of geographical concentration has enabled to a large extent the maintenance of stability and peace between the Flemish community (Flanders) and the French-speaking community (Wallonia) in Belgium and between Muslim-Croats and Serbs in Bosnia and Herzegovina (Schneckener 2002: 212).
This challenges the 'centripetal' school, which advocates for 'vote pooling' (Horowitz 1993). 12 Benjamin Reilly (2002: 157) summarizes this logic, saying that the aim is 'not to encourage the formation of ethnic parties, thereby replicating existing ethnic divisions in the legislature, but rather to utilize electoral systems that encourage cooperation and accommodation among rival groups, and therefore work to reduce the salience of ethnicity'. This review clearly shows that the success of power sharing, at least in part, depends on isolation, not integration.
Third, the favourable factors based on the covered articles in this review seem to have a combined effect. In other words, rarely does a study ascribe the success of power sharing to one factor. There are exceptions such as attributing the success of power sharing in Malaysia to the traditions of elite accommodation (Mohd Sani 2009;Morris 1999). However, this is not the norm. The survival of the Lebanese power-sharing system, John Nagle (2018) contends, is a function of the reinforcing effects of the following factors: the lack of a majority group, the equal size of groups, a strong sense of national identity and the role of external dangers or threats. In Switzerland a number of factors are reported to have shaped the success of power sharing: a medium number of linguistic groups along with a degree of geographical concentration (Bochsler and Bousbah 2015;Mueller and Rohner 2018). In Burundi the interaction between a small population, an evolving tradition of accommodation, external dangers or threats and the mediating role of South Africa are all claimed to have had a positive effect on power sharing (Vandeginste 2009).
To summarize, while many studies covered in this review evoke the importance of favourable factors, only 20 out of 373 articles exclusively focus on how favourable factors shape power sharing. Without exception, they are all either single case studies such as on Pakistan (Mushtaq et al. 2011) and Rwanda (Njoku 2005) or small-N comparisons (Lemarchand 2007;Schneckener 2002). Unpacking the structural roots of power sharing would increase the theory's predictive power (Bogaards 1998). Total 438 Source: The authors. Note: The absolute number exceeds the total number of articles (N = 373) because some articles report more than one favourable factor. Thirteen articles do not report any favourable factors.

Conclusion: a research agenda
This review article has attempted to analyse the published articles on power sharing over 50 years. The power-sharing literature has been growing since its inception in 1969. The literature witnessed a boom of publications in the last two decades at a rate not matched at the level of the social sciences. Power sharing, therefore, 'is more alive than ever' (Bogaards et al. 2019: 351). Despite the critique of power sharing as becoming 'increasingly vague, ambiguous, and even contradictory' (Dixon 2011: 309), this review demonstrates that power sharing generally has positive effects, regardless of its institutional set-up, post-conflict transitional character and world region. Nevertheless, the findings should be treated with caution for two reasons. First, the review has focused on articles, to the exclusion of books, including some classic texts on power sharing. Second, the findings might be affected by general biases characterizing academic research such as the positive publication bias. The power-sharing literature might thus benefit from a systematic review (Dacombe 2018) that evaluates the evidence base of all the quantitative studies included in the PSAD. We use this conclusion to develop a research agenda for future research. The research agenda is mainly based on insights from the reviewed articles. However, the potential areas of research outlined below have neither been fully addressed in classic books on power sharing (Lijphart 1977;McGarry and O'Leary 1995;Norris 2008) nor in recent volumes (Hartzell and Hoddie 2020;Jakala et al. 2018;Keil and McCulloch 2021;McCulloch and McGarry 2017). First, more research is needed to identify how the agency of political elites is shaped by and also shapes power-sharing institutions and social structures (Bogaards 2021a). In his early theorization of power sharing, Lijphart (1968a) highlighted the role of elites and even defined power sharing as 'government by elite cartel' (Lijphart 1969: 216). Later on, the literature took an institutional turn by focusing on the four institutional dimensions of power sharing: grand coalition, proportional representational, cultural autonomy and minority veto (Lijphart 1977). From this institutional perspective, structural conditions are treated as favourable factors while the agency of elites is considered a requirement or a built-in assumption. Treating agency as a requirement, argues Matthijs Bogaards, presupposes the dominance of elites, as strategic actors, over structural conditions such as societal cleavages (Bogaards 1998: 485). To date, the agency dilemma in power-sharing theory has not been resolved.
Answering the following questions might help bridge this gap. The first question is: why do elites behave similarly under different institutional arrangements or behave differently under similar ones? For example, in Iraq and Lebanon, despite having liberal and corporate power-sharing institutions, respectively, competition between elites has often resulted in gridlock and poor governance outcomes (Abu Ltaif 2015; Younis 2011). Another question is: which institutional and structural configurations shape the decision of elites to cooperate and which ones do not? 13 In 1990, for example, a major opposition party in Fiji declined to take up its constitutional quota of cabinet posts after elections (Premdas 2002: 34-35). In contrast, the Kenyan elites' commitment to stability prevented the country from disintegration (Kagwanja and Southall 2009). Power sharing incentivizes not only cooperation but also outbidding (McCulloch 2021). Understanding which structural and institutional factors shape elites' actions is of paramount importance.
Second, the articles analysed in this review have primarily focused on the role of elites, to the neglect of citizens. The ability of elites to manage their group members and to ensure their support has been forwarded as a 'requirement' for power sharing (Lijphart 1969: 216). According to McCulloch (2014: 512), 'for consociationalism to work, elites must be able to collectively implement the terms of the peace settlement while still retaining the support of their respective constituents'. This, however, overlooks that elites are sometimes constrained by the demands of their groups (Barry 1975). Thus, a useful question to ask is: when can political elites shape the actions of their group members and when do the latter constrain the way their leaders act? Clarifying the various citizen-elite linkages (Kitschelt 2000) under power sharing is at the heart of this agenda.
Third, power sharing is based on the premise that divided societies have 'sharp cleavages with no or very few overlapping memberships and loyalties' (Lijphart 1969: 208). The assumption is that such societies are composed of mutually exclusive groups based on attributes such as ethnicity, religion or race. In other words, the power-sharing theory overlooks 'others' under power-sharing systems (Agarin et al. 2018). Those others include marginalized non-dominant minorities (Agarin and McCulloch 2020;Stojanović 2018) or women who are usually sidelined in power-sharing agreements (Sriram 2013).
Most importantly, the theory did not predict the emergence of ethnic parties that would appeal to voters beyond their own ethnic group (Murtagh and McCulloch 2021). Nor did the theory allow for the possibility of the rise of crossethnic civic parties and social movements as is the case in Bosnia and Herzegovina, Lebanon and Northern Ireland (Agarin and Jarrett 2021;Deets 2018;Milan 2019;Murtagh 2020;Nagle 2016). This begs the following question: why do cross-ethnic social movements emerge under some power-sharing systems but not others? In answering the question, the literature would greatly benefit from applying social movement theory to divided societies, something that has for a long time been neglected (Bosi and Fazio 2017). Generally, crossing the ethnic divide is a significant move. The future will only tell if power sharing will be a 'victim of its own success' by diffusing the cleavages it was initially designed to manage (Andeweg 2000: 515) or if citizens who feel marginalized under power-sharing systems will use contentious action to challenge the system from below (Agarin 2021 Notes 1 Lijphart (2008: 3) notes that he borrowed the term 'consociational' from David Apter's 1961 study on Uganda but the term dates back to Johannes Althusius, who used its Latin equivalent 'consociatio'. Lijphart (2008: 6) argues that the shift towards using power sharing, instead of consociational democracy, was to facilitate communication with policymakers and non-academics. This review article, therefore, uses power sharing and consociationalism interchangeably.
2 For an overview of other strategies to manage ethnicity, see Hartmann (2019). 3 However, a few consociational cases such as Belgium, Lebanon, Northern Ireland and Switzerland are still mostly studied from the narrow consociational lens. 4 The Power Sharing Articles Dataset, the Online Appendix and other supplementary materials are available on Harvard Dataverse: https://doi.org/10.7910/DVN/CJN3SG. 5 In fact, consociational democracy appeared a year earlier in Lijphart (1968b) but was part of a wider discussion on a typology of political systems. 6 The decision to exclude books was based on two reasons. The first is the issue of accessibility as not all books written on power sharing are available at the university libraries where the research team members are based. Second, edited volumes are challenging when it comes to data extraction, given that each chapter has different methods (quantitative versus qualitative) and sometimes reaches contradictory findings. This is also in line with recent reviews such as Pelke and Friesen's (2019) review that only focuses on articles. 7 Whether the European Union is actually a consociational system is disputed (Bogaards and Crepaz 2002). 8 Juon's (2020) Constitutional Power-Sharing Dataset is a welcome contribution despite not covering cultural autonomy. Another limitation with the dataset is that it depends on de jure constitutional guarantees and not de facto power sharing. Writing on Ethiopia, for instance, Fiseha (2012) argues that despite the constitutional guarantees of autonomy for the different regions, in reality they are dependent on central authorities. 9 Power sharing is also reported to have negative effects, albeit with less frequency, on the inclusion of women (Kennedy et al. 2016), non-responsiveness to citizens (Sejfija and Fink-Hafner 2016), clientelism (Deets 2018), dysfunctional state institutions (Alkadiri 2011), exclusion of non-dominant minority groups (Agarin et al. 2018), impunity and lack of accountability (Tull and Mehler 2005), and corruption (Haass and Ottmann 2017). 10 Given the lower number of articles covering liberal power-sharing cases, however, the results have to be interpreted cautiously. 11 Lijphart (1977) describes Canada as a semi-power-sharing system but it has been analysed in the power-sharing literature, albeit with less frequency than other cases, particularly for its territorial power sharing between the English-and French-speaking communities (Wimmer 2007). 12 This school is also dubbed as 'integration' in contrast to consociational 'accommodation' (Reilly 2012). 13 Answering such a question is best done by using configurational methods such as qualitative comparative analysis (Schneider and Wagemann 2012) or explanatory typology (Møller and Skaaning 2017).