Democratic Institutions Under Autocracy Erin A. York Submitted In

Democratic Institutions Under Autocracy

Erin A. York

Submitted in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy under the Executive Committee of the Graduate School of Arts and Sciences

COLUMBIA UNIVERSITY

Erin A. York

This dissertation is about the function of borrowed democratic institutions in autocratic contexts, theorized to provide an arena of limited political competition between the regime and political outsiders. Despite existing explanations for how such institutions benefit the autocrat, there is much that remains unknown about their function in practice. I explore how the regime and opposition manipulate institutional features to their benefit in three papers. In the first, I show that systems of executive oversight create opportunities for the opposition to serve constituents and increase their support base. In the second, I find evidence that regime control over executive appointments is used to limit effectiveness of that opposition activity. In the third, I find that the regime’s authorities over the legislature create distributional distortions in its favor – but that other coalition members can also benefit. I address these topics using empirical analysis of novel data sources gathered during extensive fieldwork in Morocco, as well as theoretical modeling of institutional characteristics. Autocratic regimes are notorious for their opacity, and previous research has been limited by data accessibility; for the research presented here, I collect and analyze a vast database of legislative actions using techniques in webscraping and text analysis in order to obtain a more systematic understanding of legislative behavior and executive response. The results provide insight into how autocratic institutions – superficially similar to democratic analogues – operate differently in practice. Table of Contents

List of Tables ...... v

List of Figures ...... vii

Acknowledgments ...... viii

Introduction ...... 1

Chapter 1: Constituency Service and Opposition Support under Autocracy ...... 12

1.1 Introduction ...... 12

1.2 Vote Choice under Autocracy ...... 15

1.3 Institutional Authorities and Opposition Vote Mobilization ...... 17

1.3.1 Electoral competition in Morocco ...... 19

1.4 Institutional Avenues for Constituent Service ...... 23

1.4.1 Pressuring ministers via oral queries ...... 24

1.4.2 Addressing constituency issues with written questions ...... 25

1.4.3 Questions submitted during the 2011-2016 legislative term ...... 27

1.4.4 Deputy activity ...... 31

1.5 Eﬀort and Electoral Accountability ...... 34

1.5.1 Opposition-speciﬁc electoral beneﬁts ...... 39

i 1.6 Localized Support for Local Eﬀort ...... 41

1.7 Regime Strategic Considerations ...... 45

1.8 Conclusion ...... 46

Chapter 2: Horizontal Accountability under Autocracy ...... 49

2.1 Introduction ...... 49

2.2 Moroccan Institutionalism ...... 52

2.2.1 The elected legislature and executive queries ...... 54

2.3 Executive Oversight with Weak Institutions ...... 57

2.4 Data and Approach ...... 61

2.4.1 2011 elections and cabinet control ...... 62

2.4.2 Independent and dependent variables ...... 63

2.4.3 Additional covariates ...... 65

2.4.4 Approach ...... 66

2.5 Results ...... 70

2.5.1 Diﬀerence-in-diﬀerences ...... 75

2.5.2 Heterogeneous eﬀects – Royalist versus nonroyalist ...... 79

2.6 Discussion ...... 81

2.7 Conclusion ...... 84

Chapter 3: Distribution under Autocracy ...... 85

3.1 Introduction ...... 85

3.2 Autocratic Dissolution Power ...... 88

3.2.1 Legislative procedures in practice ...... 92

ii 3.3 The Model ...... 94

3.4 Results ...... 98

3.4.1 Policy Periods ...... 98

3.4.2 Coalition Formation ...... 100

3.5 Coupled Decision-making ...... 103

3.6 Conclusion ...... 108

Conclusion ...... 113

References ...... 113

Appendix A: Appendices to Chapter 1 ...... 126

A.1 Moroccan Political Parties ...... 126

A.2 Robustness Check: Negative Binomial Regression ...... 128

A.3 Eﬀort and Electoral Outcomes: Deputy Level Analysis ...... 130

Appendix B: Appendices to Chapter 2 ...... 133

B.1 Ministry Aﬃliations during the 2011-2016 Term ...... 133

B.2 Summary Statistics: Full Data and Diﬀerence in Diﬀerences Subsamples . . 135

B.3 Observational Analyses ...... 136

B.4 Matching: Robustness Using Exact Matching ...... 137

B.5 Diﬀerence in Diﬀerences: Identifying Assumptions ...... 140

B.6 Diﬀerence-in-Diﬀerences: Sample Robustness ...... 143

B.7 Robustness: Technocrats and Non-Royalist Deputies ...... 145

B.8 Non-Partisan Treatment ...... 146

iii Appendix C: Appendices to Chapter 3 ...... 150

C.1 Proofs ...... 150

C.1.1 Minimum Winning Coalition ...... 150

C.1.2 Proposition 1 ...... 151

C.1.3 Remark 1 ...... 154

C.1.4 Proposition 2 ...... 154

C.1.5 Remark 2 ...... 159

C.1.6 Remark 3 ...... 161

C.2 Empirical Speciﬁcations ...... 163

iv List of Tables

1.1 Summary statistics – Deputies in the 2011 parliament ...... 32

1.2 Predictors of written query submission ...... 33

1.3 Legislative activity and electoral outcomes – Party-district level ...... 38

1.4 Legislative activity and electoral outcomes by party type ...... 40

1.5 Localized activity and municipal vote share ...... 44

2.1 Summary statistics – Nearest neighbor matched data ...... 69

2.2 Responsiveness – Matched data ...... 71

2.3 Cox proportional hazards model – Matched data ...... 74

2.4 Difference-in-differences specification – Technocrat ministries ...... 78

2.5 Heterogeneous eﬀects – Royalist deputies ...... 80

A.1 Key political parties in the 2011 Moroccan parliament ...... 127

A.2 Predictors of written query submission – Negative binomial speciﬁcation . . 129

A.3 Legislative activity and electoral outcomes – Deputy level ...... 131

B.1 Partisan aﬃliations of cabinet ministers ...... 134

B.2 Summary statistics ...... 135

B.3 Observational analysis ...... 136

v B.4 Responsiveness – Matched data ...... 138

B.5 Cox proportional hazards model – Matched data ...... 139

B.6 Difference in differences specification – Restricted sample ...... 144

B.7 Non-royalist regression – Matched Data ...... 145

B.8 Average response rates by minister type ...... 147

B.9 Summary statistics – Matched data, non-partisan treatment ...... 148

B.10 Non-partisan regression – Matched data ...... 149

C.1 Data on dissolution power and military expenditures...... 163

C.2 Military expenditures as a function of head of state power of dissolution . . . 164

vi List of Figures

1 Elected legislatures in autocratic states ...... 2

1.1 Queries submitted by date ...... 29

1.2 Oral and written query content ...... 30

2.1 Kaplan-Meier curves – Matched data ...... 73

2.2 Kaplan-Meier curves – Diﬀerence-in-diﬀerences ...... 79

3.1 Use of elected legislatures within autocratic states, 1900-2017 ...... 89

3.2 Executive powers in nonautocratic versus autocratic states ...... 90

3.3 Military expenditures as a function of legislative dissolution power ...... 92

3.4 Game structure ...... 96

B.1 Technocrat diﬀerence-in-diﬀerences – Parallel trends ...... 141

B.2 Technocrat diﬀerence-in-diﬀerences – Patterns in query submission ...... 142

vii Acknowledgements

I owe a great deal to a great many people for their support of this project. I have benefited immensely from the faculty at Columbia. Dan Corstange and Mike Ting provided generous mentorship and demonstrated incredible patience as I worked through iterations of each chapter. For their instruction, intellectual standard-setting, and feedback, I especially thank: Alessandra Casella, Tim Frye, Nikhar Gaikwad, Donald Green, Shigeo Hirano, Macartan Humphreys, and Kimuli Kasara. And I would be remiss in not mentioning my cohort, surely the greatest in Columbia history. I have depended on your inspiration and support since we arrived in New York, and I look forward to the persistence of this network of friendship and scholarship for the duration of my academic career. I have also profited from connections across the broader academic community. Individual chapters were improved immensely by thoughtful feedback from conference participants at several points. Fotini Christia has been a terrific mentor and friend. Most recently I have benefited from the generous support and fantastic intellectual environment at Harvard’s Middle East Initiative. I owe my thanks to the excellent faculty affiliates, my cohort of fellows, and the staff for their helpful advice. I owe a debt to the countless individuals and organizations that supported the fieldwork involved in this project and others, especially Ismail Ilsouk, my research assistants Khaoula Erraoui, Imane Lahrich, and Ismail Khejjou, as well as the staff at Simsim and the Moroccan political parties, NGOs, and elected officials that gave unselfishly of their time and insights.

viii Additionally, I want to acknowledge the generous ﬁnancial support from Columbia, the National Science Foundation, the US Department of Education, and the Project on Middle East Political Science that made all of this possible. Finally, I thank my family and friends for keeping me sane during this process. I am lucky to have a network of support that is both broad and deep. I want to thank especially: my parents, who raised me (literally) in the PhD tradition and who believed in me unequivocally from day one; Michelle, who constantly inspires me with her hard work and unguarded empathy; and Michael, for being my best friend and stabilizing force while this whole thing got done.

ix Introduction

Democratic institutions are increasingly common under autocracy. These include the use of competitive national elections to select political representatives at the local and national level, the presence of both regime-linked and opposition political parties, and the establishment of legislatures modeled off of presidential or parliamentary systems and granted law-making authority and executive oversight. Globally, a majority of authoritarian states today contain legislatures that are elected through constrained or fully multi-party competition (Figure 3.1). A great deal of scholarly work addresses the rationale for this trend, primarily with attention to the implications for regime survival (Geddes, 1999; Magaloni, 2006), though other studies explore related outcomes such as how citizens decide to participate (Blaydes, 2006; Lust-Okar, 2006) and how incumbents incentivize support and maintain their advantage in electoral contests (Magaloni, 2006; Pepinsky, 2007). Yet there are still major gaps in our understanding of these institutions and the way they affect governance outcomes beyond regime survival. How do these institutions affect within-state redistribution and other sub- national outcomes? And under what conditions do elections ensure citizen representation in government? Both the theoretical and empirical literature on this topic involve a bit of scholarly hand-waving with respect to what happens within these elected institutions. Under one explanation, the legislature offers a ‘forum’ for negotiating policy concessions (Gandhi and Przeworski, 2006) – but how does the opposition extract these concessions in practice, and how does the structure of the forum facilitate this behavior? In contrast to this theorized role, a number of legislatures globally have been deemed generally irrelevant to policy-making (Malesky and Schuler, 2009; Blaydes, 2010; Truex, 2014). But it is challenging to identify

1 1.00

● ● ● ● ● ●● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ● 0.75 ●● ● ● ●●●●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ●● ●● ● 0.50 ●

●

●● ● ●●● ●● ●● ●● ●● ● ●● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● 0.25

0.00

1900 1925 1950 1975 2000 Year

Figure 1: Use of elected legislatures within autocratic states, 1900-2017. Figure shows the proportion of autocratic states by year with elected legislatures. Plot constructed using data from the Varieties of Democracy project, using the “Regimes in the World” measure to identify autocratic states. policy impact even in the most transparent, democratic settings (Clinton, 2017); the diﬃculty is heightened in autocracies where legislative deliberations are not made public. Reconciling these two ﬁndings requires deeper study.

Do Autocratic Institutions Matter for Governance?

There is a tendency to take autocratic institutions less seriously than their analogues in more democratic polities. The inﬂuence of the autocrat may be deemed to be so substantial that he can ignore the constitution – merely a paper document – when it inconveniences him. There is a kind of circular logic to such a perspective: the autocrat is so powerful that he can break rules at will, yet he is deemed powerful because of all of the constitutional authorities he has allotted to himself. If we accept that his power comes at least in part from the institutions he has created, we must acknowledge that those institutions have some signiﬁcance in regulating government behavior and outcomes.

2 Alternately, there may be a belief that the true source of autocratic power is not the paper documents, but rather the support of the informal network of powerful actors that installed him. Yet a key explanation for the existence of autocratic institutionalism is to facilitate the distribution of rents to this selectorate (Haber, 2008) – an exchange of benefits for continued support. Autocracies may indeed lack an impartial enforcement body such as an independent judiciary (Svolik, 2013), but without some means of enforcing agreements, they would fail to provide the required credible commitment to supporters (Magaloni, 2008; Boix and Svolik, 2013). Institutions cannot perform their theorized function of serving as the autocrat’s credible commitment to the selectorate or would-be political challengers while simultaneously operating as meaningless rubber stamp bodies that can be wiped away with a single dictatorial decree. So there are important theoretical reasons to expect autocratic institutions to matter for governance outcomes. And yet, while scholars of autocratic states are careful to contextualize their findings in light of the cultural and historical environments in which they take place (Gandhi and Lust-Okar, 2009), the institutional specifics at work – including, for example, the authorities granted to the legislature and the procedures by which policy-making or oversight are carried out – have often been absent in previous scholarship.1 A robust literature in democratic governance examines the impact of various institutional features – electoral systems and voting rules, legislative procedures, veto powers and the influence of the executive – on a wide range of governance outcomes, including inequality, representation, redistribution, and polarization.2 Yet the subset of studies that take on autocratic rules is comparatively small. What is it about the choice of institutional rules that facilitates different outcomes? Do some governing structures allow the opposition influence over policy, or better equip them to provide goods to their supporters? Under what circumstances is the autocrat able to retain more goods for himself?

1One exception to this is the study of electoral rules. A limited literature examines the impact of diﬀering systems on the regime’s electoral advantage (Lust-Okar and Jamal, 2002; Pripstein Posusney, 2002). 2See, for example, Ordeshook and Shvetsova (1994); Huber (1996); McCarty (2000); Austen-Smith and Feddersen (2006) as a very small sample of the vast literature on institutional rules and political outcomes.

3 This relative blind spot in our understanding also follows from the general opacity of autocratic polities. These states often impose restrictions to government transparency and the media that result in severe limitations in what is shared about their internal workings in practice. While in most democratic contexts, there is a standard of transparency that requires public disclosure of all but the most security-oriented topics, this is absent under autocracy (Hollyer, Rosendorff and Vreeland, 2011). Even information on very public political action like national elections is often limited in these contexts. Whether due to capacity issues or paranoia about the effects of information on the public, few autocratic regimes have chosen to make their internal proceedings available. So there is a practical rationale for the limited research on this topic: data availability, including systematic information on behavior in office, the policy preferences of both the public and their elected representatives, and the distributive policy implemented by the regime, remains a major challenge in many states. In the research that follows, I seek to take the institutional rules of autocracy seriously as a subject of academic study, while recognizing the existing imbalance between the autocrat and his political competition. And this work is rooted in scholarship borrowed from more democratic settings. Dictators establishing institutional autocracy do not invent these institutions from scratch: they modify constitutions borrowed from democracies to suit their needs. As a result, autocratic institutions carry the rough outlines of these democratic systems – but with important alterations that require closer examination.

Legislative Function under Autocracy

Democratic institutions help the autocrat coopt the political opposition (Gandhi, 2008), address the needs of the electorate (Blaydes, 2010), and establish domestic and international legitimacy (Schedler et al., 2006). Yet they also create an arena of competition in which both the regime and outsiders grapple to increase their respective political stakes. The opposition might seek to further its inﬂuence within the conﬁnes of the system by attracting greater

4 electoral support, while the regime seeks to prevent outsiders from growing too powerful. Understanding how these institutions work in practice is important in order to understand the stability of these arrangements and how they achieve an uneasy equilibrium despite the power imbalance at their core. This dissertation contributes to a growing literature on the function of legislative institutions under autocracy (Malesky and Schuler, 2010; Truex, 2014; Reuter and Robertson, 2015). I seek to show that the rules and procedures in such settings are not wholly superficial but may have meaningful consequences for governance – who participates, how citizen representation works, and the kinds of policies that emerge. Why does the opposition participate in a system that disadvantages it, and how does it succeed in this context? How do institutional differences relative to democracy impact political behavior and policy outcomes? I explore the impact of autocratic institutional rules on a variety of outcomes, including opposition behavior in office and strategies for attracting electoral support, discretionary patterns of responsiveness from regime-linked versus partisan ministers, and distributive policy. My approach in addressing these questions involves examining autocratic institutions as they are constitutionally characterized as well as how they operate in practice. I argue that by doing so, we can learn more about how these benefit the autocrat – and whether the political opposition matters in such contexts. From a theoretical standpoint, I take advantage of the fact that aspects of legislative procedure in these regimes are modified from existing democratic structures and adapt a model of policy bargaining under democracy to reflect common features of autocratic states. Empirically, I use novel data that systematically capture legislative activity in the Moroccan context to examine behavior in office and how this is related to electoral outcomes. The focus of the three dissertation papers that follow is on borrowed institutions and how they shape behavior and governance outcomes in an autocratic setting. In the first, I take on the question of how political outsiders gain popular support, and find that the institutional authorities of the elected legislature over the executive serve as an important mechanism

5 for providing the constituency service that voters want. This demonstrates that legislative authorities help to level the playing ﬁeld between regime-linked and opposition actors. In the second, I show that the weak checks and balances in this setting mean that the regime is in some cases able to undermine this mechanism of horizontal accountability by exploiting its inﬂuence over executive appointments. In the third paper, I examine institutional rules from a formal theoretical perspective, using a model of legislative policy bargaining to show that a common autocratic power over the legislature is a source of policy distortions in his favor.

A note about case selection

As with much other scholarship on autocracy (Gandhi and Lust-Okar, 2009), a great deal of the inference in this dissertation is drawn from observation and data collection conducted in a single case: Morocco, a constitutional monarchy since achieving independence from France in the mid-20th century. This is a case with a set of unique, inherited political features,3 and it is certainly true that some aspects of governance in Morocco will be specific to the context. For the purposes of scholarly inference, Morocco is not intended to serve as a median case and the research here is not being presented as ‘representative’ of autocracies globally. Yet by focusing on institutional features – and, more particularly, institutional features that are utilized in other, less similar contexts – I hope that the findings here have broader applicability. Indeed, the institutional features under study are relatively prevalent. The focus of the first paper is on executive queries, a common lever of executive oversight granted to the legislature in most settings with parliamentary institutions, including both democratic and autocratic states.4 The second addresses regime control of key ministries and the appoint-

3These include, for example, a royal family with a great deal of historical and religious legitimacy as well as a class of political notables that continue to exert outside inﬂuence on political competition (Waterbury, 1970). 4Martin (2011) documents a broad class of research addressing the use and importance of this authority in a range of settings.

6 ment of technocrats to these posts – also a typical lever of authority in autocratic states globally. Finally, the third paper abstracts from the Moroccan institutional setting to examine dissolution power over the legislature; this authority is present in around half of autocracies today (see Figure 3.2).

Constituency service and opposition support under autocracy

In the first paper, I examine legislative behavior and electoral competition, addressing the question of how opposition parties mobilize turnout when they lack the regime connections ostensibly required to deliver private goods to its supporters. Autocratic regimes possess an extensive set of strategies to incentivize electoral support for its preferred candidates, yet political outsiders often achieve significant electoral successes. I argue that institutional authorities allocated to the legislature, created through the same process and contemporaneously with elected institutions themselves, help to level the playing field between regime and opposition candidates in developing autocracies. Though regime candidates may have the ear of those in power through personal connections and patronage networks, opposition politicians can use ‘by the book’ politics – the legitimate authorities that accompany their office – to ensure the supply of constituency goods. I draw on a vast database of activity from recent legislative terms in Morocco, including more than 38000 unique written and oral queries submitted by elected members of parliament, and identify a strong linkage between activity in office and subsequent electoral outcomes. Voters reward parties and deputies that engage in more constituency service via institutionalized action. But the relationship between activity and vote share is exclusive to opposition parties: regime-linked parties do not see electoral gains from increased legislative activity, but neither are they punished for shirking in office.

7 Horizontal accountability under autocracy

Autocrats use institutions to coopt members of the opposition and neutralize their threat as outsiders. But given the extremely minor policy concessions typically offered through such mechanisms and the threat to their credibility posed by alliances with the regime, why do opposition actors allow themselves to be coopted? In Morocco, a constitutional monarchy, the constitution grants elected members of parliament the authority to make requests of government ministers, but it does not require that ministers respond. This is consistent with general patterns of autocratic structure, in which competitively elected bodies are subsidiary to appointed positions. As a result, executive responsiveness – and, correspondingly, legislative effectiveness – is dependent on the incentives of the ministers themselves to answer a given query. I expect that regime-selected technocrats are disincentivized to respond to queries, given their importance for attracting electoral support. I test this hypothesis using data on ministry responsiveness in the Moroccan parliament, using both matching and difference-in-differences approaches to show that ministers with partisan affiliations are more likely to answer requests posed by legislative deputies than technocrats selected by the regime. The results demonstrate that (partisan) participation in government helps to facilitate institutions of horizontal accountability and provide suggestive evidence that the presence of regime agents may undermine legislative effectiveness.

Distribution under autocracy

How does the presence of an autocrat with unique authorities over the legislature aﬀect distributional politics in what are otherwise democratic institutions? Under institutional autocracy, the autocrat often exerts considerable control over government formation via manipulation of elections or executive privilege. And yet he relinquishes other powers – namely, the ability to enact laws – to elected legislatures that might include representatives from outside the regime. How should we expect this to aﬀect the distribution of goods? And how

8 might different institutional restrictions alter this distribution? I consider the implications of two common autocratic authorities: the ability to dismiss an elected legislature, necessitating new elections, and the ability to manipulate electoral results in his own favor. I construct a formal model of agenda control and policy bargaining in an autocratic context, drawing on modeling approaches used in democratic settings. I find that the ability to dismiss the legislature alone creates an incentive for other players to include the autocrat in governing coalitions, since it is only by including him that they can ensure coalitional stability and retain the benefits of rule with certainty. This, in turn, creates a distributional distortion in the autocrat’s favor, as he is assured of access to proposal power. If the autocrat creates a more binding institution, in which his decision to dismiss the legislature is linked to his support for policy proposals, he can further incentivize members of the ruling coalition to include him in their policy proposals and, under some conditions, increase his own expected payoff relative to a game in which he votes separately on policy and coalition dissolution. In both cases, payoffs are asymmetrically distributed relative to a democratic setting. While we might expect patterns of distribution that favor the autocrat, the model here demonstrates one institutional mechanism that contributes to this asymmetry. And this has practical implications for regime behavior: I find that the ability to dissolve the legislature produces the same distributional outcome as the ability to completely control electoral outcomes. But extreme electoral manipulation may be costly behavior for an autocrat to engage in, leading to reduced domestic and international legitimacy. As a result, it may be preferable for a regime to invest in these less visible authorities that do not need to be regularly deployed to be effective in influencing distribution.

Implications for Further Study

These articles demonstrate the importance of taking a closer look at the speciﬁcs of autocratic institutions in light of a broader range of outcomes than regime survival. This

9 is not the first or only set of papers to do this, but with this research I make the case that the rules in such contexts are not totally ignorable, but can have important implications for sub-national outcomes such as representation and redistribution. While it may seem that authoritarian institutions are subject to dictator ‘discretion’ (Brancati, 2014), they have similar underpinnings to democratic settings – they are constitutionally established and rooted in agreements with existing political stakeholders. Thus, even if they are more malleable than democratic constitutions, we should still expect that there are costs to revising or overriding them. The work here offers a rare look into elite political behavior, using systematic data from within an autocratic legislature. Much of the micrologic of autocratic legislatures remains under-theorized, including how these institutions serve to benefit the regime as well as the potential risks that they pose by allowing political competition to flourish. Though many seem to lack strong policy influence in practice, I propose that this may be an endogenous outcome of a system filled with regime manipulation, not an institutional feature – and as such, we should study the circumstances under which citizens elect representatives that will contravene regime objectives while in office. And I expect that these institutions will continue to provide fertile ground for study. In particular, autocratic legislatures often engage in greater transparency than more sensitive aspects of regime administration, and there are signs that this is improving globally.5 Par- ticipants in such bodies have an incentive to make their work visible to voters and combat the assumption that they are ineffective in representing constituent interests. As a result, the data limitations that hamper research activity in autocratic regimes are lessened here. Ultimately, I intend for this research agenda to inform our understanding of autocratic governance. The research here does not claim to address all of the questions posed above or to overcome all of the challenges documented here and elsewhere regarding the study of autocratic regimes at a global level. Yet by focusing on the function of institutions, I hope

5The Moroccan parliament recently restructured its public access website in order to make it more accessible to citizens, in consultation with civil society actors.

10 to demonstrate both the importance of taking these rules seriously and studying them as if they are, at the least, costly to ignore.

11 Chapter 1: Constituency Service and Opposition Support under

Autocracy

1.1 Introduction

How does the opposition attract voter support under autocracy? I answer this question with an argument about the importance of autocratic institutional rules; in doing so, I also uncover evidence of electoral accountability in an autocratic setting. Briefly, I will argue that institutional channels – created contemporaneously with elected institutions themselves – help to level the playing field between regime and opposition candidates. Though regime candidates may have the ear of those in power through backdoor channels (historical rela- tionships, family connections, etc.), opposition politicians provide the goods voters expect using the authorities of their office. Other scholars have argued that regime-affiliated candidates satisfy voter demands by exploiting their close personal connections to the government to supply patronage goods to supporters, but little work has been done to explain how opposition parties generate electoral support in the face of structural disadvantages. And yet, globally, opposition groups such as the Egyptian Muslim Brotherhood and the Mexican PAN have achieved notable electoral successes. In Morocco, the case studied here, opposition parties – groups lacking linkages to the regime1 – have won pluralities in the national legislature since 1997: how have such parties moved beyond niche appeal? To attract a broader swath of the electorate, opposition parties must be able to compete directly with regime-affiliated parties for voter support. In Morocco, as in other autocratic

1In Morocco, a constitutional monarchy, I refer to regime-aﬃliated parties as ‘royalist’ and those lacking such linkages as ‘non-royalist’ or opposition.

12 settings, voters primarily expect politicians to deliver constituency focused services.2 I argue that though they lack the personal access that gives regime-backed candidates influence over state goods, opposition outsiders can instead take advantage of institutional features that offer them some influence over government ministries and the bureaucracy. Theories of cooptation hold that autocrats use minor political concessions to incentivize the cooperation of the opposition. As a result, though the balance of power remains heavily weighted in favor of the executive, there are nevertheless a set of authorities that give legislative actors oversight over policy and its implementation. These institutional features are in many cases constitutionally enshrined and therefore difficult for the regime to remove – but due to a dearth of empirical evidence, we know little about how the opposition actually behaves in office. I argue that these concessions to the legislature, however slight, give the opposition an opportunity to compete for the same set of voters as regime-affiliated candidates – those motivated by the promise of local goods – but through different means: while regime insiders trade on their connections for favors and local patronage, opposition deputies use the legitimate powers of office to deliver the same goods. Voters then evaluate incumbents based on the constituency outputs they received, regardless of the mechanism through which they were acquired. In Morocco, as in many parliamentary systems, bureaucratic oversight occurs through querying of cabinet ministers, which allows elected representatives to formally request information or explanation directly from the executive branch. Drawing on interviews with politicians and content analysis of written and oral queries, I show that these legislative authorities – particularly written queries – facilitate constituency service in crucial ways in Morocco. They allow deputies to access information relevant to their districts, to pressure government agencies on local issues, and to convey this information to the constituents concerned. Notably, though deputies from both royalist and non-royalist parties agree that written questions are an important component of their activities and duties as elected politi-

2Both royalist and opposition party members interviewed for this project consistently characterized constituency service as a priority for acquiring electoral support.

13 cians, I find evidence that this authority is exercised far more frequently by the latter group, consistent with the claim that this ‘by the book’ action can substitute for other avenues of service provision such as patronage activity accessible only to those with personal connections to the regime. I further link electoral outcomes to this behavior and find that party-level vote share within a district is strongly linked to the legislative performance of its deputies. Voters reward parties and deputies that engage in more constituency service via institutionalized action; they do not, however, increase their support for deputies who submit more nationally- focused oral queries, suggesting that they see this behavior as a distraction from the more important district-level concerns. Finally, I find that the relationship between institutionalized constituency service and vote share is exclusive to opposition parties. Royalist parties do not see electoral gains from increased legislative activity, but neither are they punished for shirking in office. On the other hand, opposition parties experienced meaningful electoral gains in districts where their deputies were more active. The results here provide a novel look inside an autocratic legislature and challenge the assumption that such institutions are mere window dressing. I find systematic evidence that elected deputies work for constituents in office. And far from being trivial exercises in regime sycophancy, autocratic elections have consequences: better performance in office is associated with electoral support. The institutional rules in this setting matter and can create space for opposition contestation, helping to level the playing field with regime actors. Yet the differential response of voters to behavior by actors with closer ties to the regime implies that substantial politicking occurs off the books. Royalist parties seem to satisfy voter expectations without going through official channels, suggesting that institutional actions and those that draw on personal connections may be substitutes in the eyes of voters. While voters may have the same ultimate expectations for both types of politicians, the ways in which regime-affiliated and opposition actors fulfill these expectations differ substantially.

14 1.2 Vote Choice under Autocracy

What motivates voters in autocratic settings? Other scholars have addressed this question, largely in the context of explaining voter support for regime representatives (Magaloni, 2006; Greene, 2007; Pepinsky, 2007). The general academic consensus is that vote choice in autocratic settings is primarily the result of competition over localized or private goods: voters are motivated to support those politicians that they expect are best able to deliver to their constituency (Gandhi and Lust-Okar, 2009). Elections are thus seen as a contest not over policy, but rather over patronage and how and to whom it is distributed (Lust, 2009). This explanation of voter demands makes it clear why, in such a context, citizens are often motivated to support regime-backed candidates. Individuals with close connections to the regime are expected to have insider access (Magaloni, 2006; Lust-Okar, 2006) or personal wealth (Blaydes, 2010) that enables them to deliver desired goods directly to their supporters. Citizens anxious to avoid exclusion from state resources will therefore seek to ensure that their district is represented by politicians with the capacity to deliver. But if patronage is the normal mechanism of service delivery, opposition candidates – those that are not aﬃliated with the regime – should be at a disadvantage. A corresponding implication of the argument in favor of patronage-based voting is that the electorate is expected to be largely unconcerned with policy promises. That is to say, voter utility is based primarily on expected private or group-level redistribution, and individuals ascribe comparatively low weight to candidate ideology. This argument is particularly plausible in light of the challenges in delivering on policy-based campaign promises in autocratic contexts. Many such legislatures are seen as ‘rubber stamp’ institutions with little agency over the policy that is produced (Malesky and Schuler, 2009; Blaydes, 2010; Truex, 2014). Part of this may be due to the manipulations the regime takes to ensure that the legislature is fragmented and the electoral deck is stacked in favor of regime candidates (Prip- stein Posusney, 2002; Schedler, 2002). Even if legislatures are able to independently produce

15 policy changes on a national level, though, such progress must necessarily be incremental and requires broad electoral support beyond that provided by a single district. In other words, practical voters should not expect policy-focused campaign promises to directly translate into new laws, even if the candidate in question wins office. This does not imply that opposition parties lack ideological objectives; in fact, that is often what distinguishes them from regime-affiliated politicians.3 Greene (2007) posits that opposition parties are created and supported by a specific subset of the population: those with particularly strong policy convictions. But this overemphasis on policy limits such parties to occupying ‘niche’ segments of the political spectrum due to their narrow appeal. It is only when the regime loses its advantage that the opposition can compete, citing a ‘leaner bureaucracy’ that reduced the regime’s ability to provide patronage benefits to voters as the downfall of the Mexican PRI. There may be more complexity to voter interests, despite the broad scholarly consensus from an array of regimes.4 But there is substantial evidence in support of the claim that regime-affiliated candidates seek to attract votes through the promise of patronage. To compete within the broader electorate and avoid ‘niche’ status, then, the opposition must also be able to deliver local or even personal benefits. And yet the existing literature suggests few avenues for them to do so, given their status as regime outsiders. Lindsay Benstead explores the citizen-deputy relationship in Morocco and Algeria, identifying ‘constituency responsiveness’ as a key expectation of citizens and arguing that the need to satisfy such demands leads to regime cooptation of the opposition (Benstead, 2008). Masoud (2014) offers an alternative explanation in addressing how the Muslim Brotherhood rose to political prominence in Egypt. Like others, he argues that they did so by providing voters with material rather than ideological benefits, though he finds that they acquired the necessary

3Speciﬁcally, opposition parties typically have ideological objectives that diﬀer from the policies embraced by the regime (Lust-Okar, 2005). 4For example, the high rate of invalid ballot submission in Morocco, as in other autocracies, may constitute evidence of protest voting, indicating voter dissatisfaction with the status quo (Blaydes, 2006; Domínguez, Galvis and Superti, 2017).

16 resources via the strong social and fiscal networks they created within communities rather than through increasing dependence on the regime. These few examples notwithstanding, the literature on opposition support is relatively limited, focusing on the demographic characteristics of those voters hardest for the regime to buy off (the middle class), but finding in general that turnout is typically low for this bloc of the electorate (Magaloni, 2006; Malesky and Schuler, 2009). And yet opposition parties are represented in a large number of autocratic legislatures globally, and in some cases achieve substantial successes.5 In Morocco, the case examined here, successive opposition parties have won a plurality (though never close to a majority) of seats in every legislative election since 1997, surpassing parties with royalist linkages. How have they achieved such success despite their comparative disadvantage in regime connections? In the next section, I develop an explanation that draws on the relevance of institutions for satisfying voter demands.

1.3 Institutional Authorities and Opposition Vote Mobilization

Elected legislatures in autocratic settings often appear to be rubber stamp bodies that simply enact the regime’s preferred policies and lack real authority relative to the executive. And yet, there is reason to believe that they are not entirely without importance. Com- petition for legislative seats is often fierce (Levitsky and Way, 2002), suggesting that there are, at the least, personal returns to office and perhaps more at stake (Lust-Okar, 2006; Blaydes, 2010; Reuter and Robertson, 2015). And, importantly, such bodies are typically granted a set of constitutional authorities. Though their authority relative to the executive is limited compared to their counterparts in democratic polities, members of the legislature nevertheless have certain privileges within the governing apparatus. Once having granted such privileges through constitutional commitment, it is difficult for autocrats to remove them entirely.6 Indeed, the extensive literature on why such bodies

5The Muslim Brotherhood in Egypt and the Mexican PAN are prominent examples of electorally successful former opposition parties. 6Though constitutional amendment is generally challenging even in such unbalanced regimes, autocrats

17 exist stresses the importance of institutions in tying the hands of the autocrat in order to induce cooperation from other members of the political elite (Gandhi and Przeworski, 2006). Institutions serve as the credible commitment required for such a bargain (Boix and Svolik, 2013) and should therefore be costly to alter. So what is the role of legislative authority in governance, and should we expect this to matter for electoral competition? The aforementioned literature on rubber stamp parliaments argues that legislatures play an extremely minor role in establishing policy; though they ostensibly vote on laws, most bills are drafted by executive ministries and transmitted to parliament simply for ratification (Blaydes, 2010; Thomas and Sissokho, 2005). Simi- larly, in most cases the implementation of policy is also controlled by the executive via the bureaucracy (Jensen, Malesky and Weymouth, 2014). But in many cases, the constitutionally established legislative authorities include provisions for interaction with and oversight of this bureaucracy. Even limited institutionalization can produce some horizontal accountability within government structures (Diamond, 1999). Elected office comes with perks: this helps to explain the competitiveness of these elections (Magaloni, Diaz-Cayeros and Estévez, 2007; Shehata, 2008). The enhanced status acquired with office can then be used to expropriate state resources for legislators themselves (Truex, 2014), but it might also be used to force bureaucrats to do their jobs (in other words, to oversee policy implementation). This is a related argument to that put forth by Lust (2009), in which she argues that elected politicians in Jordan use their office to direct goods to supporters. But instead of doing so through exploitation of their position (that is, by using their influence in a corrupt or illegitimate manner), I suggest that they may achieve similar effects through regularized, by the book activities. Legislators may not be able to immediately and directly translate electoral wins into policy advancements that favor their constituents, but they can instead influence the implementation of existing policy to ensure that their district is not overlooked by the regime. In that sense, then, this type of action blurs the may in some cases be able to manipulate or discard other rules of the game, such as through modifying voting systems or gerrymandering electoral districts (Pripstein Posusney, 2002).

18 line between non-programmatic and programmatic politics (Stokes, Dunning and Nazareno, 2013): the reality in these settings is that implementation of policy can occur selectively but through institutionalized channels. The argument is thus as follows: in order to buy cooperation, the autocrat must make political concessions in the form of institutional authorities granted to the elected legislature, including oversight of the executive-controlled bureaucracy. These institutions create opportunities, especially for members of the political opposition: while regime insiders may trade on their personal connections in order to convince voters to support them, opposition legislators use the power of their oﬃce in legitimate ways to serve their constituents. Voters then evaluate incumbent parties on the basis of the constituency service they received during the preceding term, though that service is provided through diﬀering mechanisms (cronyism or by the book politics) depending on each party’s proximity to the regime.

1.3.1 Electoral competition in Morocco

How do opposition actors use these legislative authorities in practice? I explore legislative activity and electoral accountability in the context of the Moroccan parliament, an institution created in the years following independence from France and first elected in 1963. Parliamentary elections are held every five years, and a majority of seats (305 out of 395) are filled via closed list proportional representation within districts of between two and six seats. The remaining 90 seats are filled via national lists reserved for female and youth candidates.7 Electoral competition is party-based and highly competitive; in many cases, a vote share in the single digits is sufficient to earn a seat.8 Though in the years post- independence, elections were occasionally marred by allegations of fraudulent behavior, the past few decades have seen voting conducted in a generally free and fair manner. Indepen- dent election observers assessed the 2011 parliamentary elections positively, describing them

7National list candidates are selected by party leadership and are typically ﬁrst-time deputies; seats are allocated proportionally based on the votes a party receives nationally. 8In the 2016 national legislative elections, there were an average of ﬁfteen lists deposited per district, though there are only two to six seats available per district.

19 as “procedurally sound and transparent” and generally free from concerns about “tampering or systematic procedural violations” (National Democratic Institute, 2011). Morocco is a monarchy, and no party represents the palace in an official capacity in electoral competition. Instead, within the crowded political scene today9 are a number of parties seen as having close ties to the monarchy through their connections to the royal family, often through leaders’ membership in the historical class of notables (makhzen) first established during the French protectorate and further cultivated by the palace during the early stages of independence (Waterbury, 1970; Liddell, 2010). Though party discipline is relatively loose, with occasional party switching by prominent individuals, other scholars have generally sought to categorize party organizations on the very basic criterion of whether or not they have ties to the monarchy (Lust-Okar, 2005; Pellicer and Wegner, 2013). Parties for which the party leadership has direct, personal links to the palace or the makhzen are characterized as ‘royalist’ (Willis, 2002a).10 I refer to those formed without such ties as ‘non-royalist’ or part of the opposition.11 Opposition parties can typically be characterized according to their ideological stance, while it is often difficult to identify a cohesive policy orientation for royalist groups. The Moroccan opposition is quite fragmented, comprising a variety of parties with a diverse set of ideological objectives including both secular leftist and Islamist ideals that typically compete independently in elections (Willis, 2002b; Szmolka, 2010).12 They are distinguished primarily by what they apparently lack: proximity to existing power structures and a resulting ability to deliver non-programmatic patronage benefits to their supporters.

9There are more than 30 registered political parties, and more than 27 participated in the 2016 national elections, though only 12 won seats in parliament. 10An example of one such party is the Party of Authenticity and Modernity, or PAM, created in 2008 by a close friend of King Mohammed VI. 11In settings with parliamentary institutions, the word opposition may also refer to parties outside the governing coalition. In this paper, I use ‘opposition’ exclusively to refer to political outsiders lacking regime connections and advocating for policies that deviate from the status quo. I use ‘government’ and ‘non- government’ where necessary to indicate whether or not a party is a participant in the governing coalition. 12Opposition parties very occasionally join forces in electoral contests (or agree to compete in separate districts), but these coalitions are generally short-lived (Buehler, 2018) and have failed to transfer to collective behavior within parliament.

20 Among the eight parties that are most successful in the current political environment, four have close regime linkages.13 Though the literature on vote choice under autocracy would suggest that parties proximate to the regime should be the most electorally successful, non- royalist parties in Morocco have regularly won over large swathes of the electorate. In the last five parliamentary elections, an opposition party won a plurality of total vote share, and the 2011 national elections resulted in opposition parties winning roughly 60% of seats in parliament. Some scholars have argued that the royalist/non-royalist categories overlap to an extent with clientelist/programmatic campaign styles (Pellicer and Wegner, 2013). Monarchy- affiliated parties are seen as taking a clientelistic approach to campaigns, relying on the personal appeal of businessmen and local notables to provide jobs and/or financial favors in exchange for votes, while the opposition gains support by advocating for political reforms. But Pellicer and Wegner (2013) might oversell the extent to which opposition parties promote social justice in exchange for votes; as discussed in the preceding sections, it is not clear that parties can deliver on policy promises in the immediate term due to legislative fragmentation and the direct and indirect influence of the monarchy over political outcomes. Also, they identify only one party (the Islamist PJD) as fully programmatic, despite the existence of several other opposition parties. While the PJD may indeed be more structured and disciplined than other parties in this setting, they must still compete with parties that can deliver localized goods in an environment in which voters have been conditioned to expect such benefits in exchange for their support (Lust-Okar, 2006). Both types of parties implement similar electoral campaigns – and are competitive in the same districts.14 The campaign period lasts only two weeks according to the electoral law, so parties are limited in the complexity of the platform they seek to communicate to voters, a further discouragement to programmatic appeals. Instead, campaign materials

13See Appendix Table A.1 for information on party aﬃliations. 14In the 2011 elections, the median and modal outcome across the 92 parliamentary districts was a 50-50 split of seats between royalist and non-royalist parties. Only 14% of districts elected exclusively royalist or non-royalist deputies.

21 emphasize the individuals running for office and party symbols rather than the often am- bitious but unrealistic policy agendas developed by party leadership. Even when explicitly addressing policy considerations, parties tend to offer goals (‘reduce illiteracy by 10%’15) rather than concrete proposals for achieving them. During the 2016 campaign period, in an effort to increase emphasis on programmatic considerations, an NGO arranged for the first time a televised debate between national leaders of prominent parties. Other NGOs facilitated radio-based Q&A sessions with candidates for office. And yet these efforts to refocus elections on policy issues are thus far quite limited. Instead, conventional scholarship on vote mobilization in Morocco emphasizes candidates’ use of clientelistic appeals and vote-buying.16 This is perhaps unsurprising given the brevity of the campaign period and intense level of competition for representation. Two weeks offers little time for changing the mind of the electorate; instead, incumbent parties must rely on voter evaluations of their performance in office and try to maximize turnout among supporters to contend with the many challengers. For the purposes of this study, I restrict attention to the eight largest parties, which together won all but fifteen seats (96% of parliament) in the 2011 legislative elections. Table A.1 in the Appendix identifies these parties and provides some notes on their ideology and affiliations. In the analyses that follow, ‘royalist’ parties include members of the Party of Authenticity and Modernity (PAM), the National Rally of Independents (RNI), the Consti- tutional Union (UC), and the Popular Movement (MP). The opposition includes both leftists (the Socialist Union of Popular Forces, or USFP, and the Party of Progress and Socialism, or PPS), centrists (the Istiqlal party), and Islamists (the Party of Justice and Development, or PJD).17

15This was an objective included in campaign materials for the non-royalist Istiqlal Party distributed in October 2016. 16Though current scholars largely agree that this takes place, evidence of such activities is primarily anecdotal (see for example Sater (2009)). 17It is important to clarify that, like the set of opposition parties, the set of royalist parties do not operate as a bloc in parliament; though all have connections to the regime, the personalist nature of such connections do not necessarily generate common ground.

22 1.4 Institutional Avenues for Constituent Service

Existing literature and empirical explorations in similar contexts have demonstrated that voters seek representatives who can deliver localized benefits. In Morocco, this constituency service may comprise a variety of different activities addressing district level needs, including development of local infrastructure, service provision at the local level, job creation, or even directly assisting individual citizens in their interactions with the state (Benstead, 2008). In other words, the expectation is that voters are less concerned with the establishment of national level policy but instead on ensuring that government agencies distributing services are adequately addressing the needs of the district. And deputies are certainly aware of this: one deputy noted that ‘interaction with citizens’ was the most important part of her job.18 While connected individuals may be able to provide these benefits through under the table dealings, the Moroccan constitution also establishes an institutionalized way for legislators to hold government agencies accountable. Among other forms of executive oversight, members of parliament have the authority to query government ministries both orally, in regular televised sessions, and in written form.19 Other scholars have argued that legislative questions constitute a meaningful mechanism of horizontal accountability between branches of government in mature and developing democracies (Herron and Boyko, 2015). Though the questions do not themselves create policy, legislators see them as a necessary – and usually successful – interface with government ministries, allowing them to gather information, extract public promises, and pressure ministers into action.20 In that sense, then, they offer elected representatives a lever for influencing policy implementation. Most deputies interviewed for this project stated that oral and written questions constitute an important part of their responsibilities as MPs.

18Deputy interview, February 1, 2018. 19This is a common feature of parliamentary institutions in both democratic and autocratic settings; see, for example Raunio (1996) and Malesky and Schuler (2010). Other legislative authorities relative to the executive include the power of censure – the ability to force the resignation of the government through majority vote. 20Deputy interview, January 23, 2018.

23 1.4.1 Pressuring ministers via oral queries

Though in theory, both oral and written questions might be used to satisfy constituent concerns, structural diﬀerences in how queries are submitted mean that lawmakers see written questions as more useful in addressing district-level issues. In particular, opportunities for oral questioning are limited in a way that written questions are not: oral sessions occur once every few weeks, so a given minister might appear only a few times in the course of a legislative session. Additionally, the sessions are time delimited to a few hours per minister, with each party receiving a set number of minutes proportional to the amount of seats they have in parliament. As a result, party leadership usually exerts substantial control over the allotted time, choosing which deputies get to ask questions and which topics are addressed.21 This time constraint has implications for the types of questions that get asked: in general, the emphasis is on the broader policy considerations that the party wishes to address. As a deputy from one of the smaller parties represented put it, “We have only one minute and forty seconds per week, so we try to raise the most relevant issues.”22 As a result, it is rare that deputies choose to use these sessions to address local considerations; instead, party leaders seek to spend their time pressuring government ministries to concede on policy priorities associated with their legislative agenda.23 On the other hand, the fact that these sessions are conducted in real-time and televised make them an unusually public form of legislative activity. Often, parties use them to try to put ministers on the spot in the hopes of extracting public commitments.24 One deputy described oral query sessions as “a weekly opportunity to observe and control the government’s work and its progress and to make ministers engage into promises.”25 Somewhat optimistically, some MPs also point to the fact that these sessions are televised to argue that

21Deputy interview, October 17, 2016. 22Deputy interview, January 23, 2018. 23This is common in parliamentary settings with this institutional construct (Martin, 2011) and similar to how these queries are used elsewhere, including in democratic polities such as the UK (Rogers and Walters, 2013). 24Deputy interviews, January 23 and January 30, 2018. 25Deputy interview, February 12, 2018.

24 they might have an inﬂuence on electoral support: one deputy stressed the fact that this is the only form of legislative activity that is visible to citizens, as voting sessions are not televised or published.26 Another pointed out that the sessions oﬀer an accountability mechanism for citizens, as they can directly observe their representatives raising their concerns with ministers.27 Of course, it is likely that only the most politically engaged citizens – a fraction of the electorate – follow the televised sessions.28 More importantly, these query sessions provide an opportunity for signaling to other deputies and their own party, given that they take place in front of the entire legislative assembly. Deputies who wish to prove their value to party leadership may thus seek to take a more active role in proposing questions that resonate with the party’s issue agenda.

1.4.2 Addressing constituency issues with written questions

In contrast, written questions offer a regular means for addressing constituency level issues. They can be submitted to ministries at any time rather than simply once or twice a term, and there is no cap on the number of queries that may be posed by an individual deputy. In fact, some deputies produce thousands of questions in a single term. These constitute a more quotidian, impromptu method of interaction with government agencies. Also, although both written and oral questions are recorded in a publically accessible database on the parliamentary website, written questions clearly represent a less visible activity given that they are not broadcast by default to other politicians or the general public. They offer comparatively little utility from a performative standpoint, and deputies who submit them are thus more likely to do so as honest effort in office rather than in an attempt to impress party leadership. Though written questions do not provide the same opportunity to directly confront min-

26Deputy interview, February 5, 2018. 27Deputy interview, February 1, 2018. 28While television statistics are not available, the session videos are also published on YouTube and, in general, receive no more than a few thousand views (the most viewed video published on the channel had been watched 16,480 times as of May 2018).

25 istry heads, they are intended to elicit a written response from the ministry in question, generally expected to be sent within twenty days.29 This measured response platform allows deputies to obtain “detailed answers” relative to the oral query sessions, making these questions more useful as an information-gathering tool.30 These queries also allow deputies to channel constituent concerns directly to ministries: in a scenario analogous to traditional conceptions of wasta (mediation through cronyism) elsewhere in the Arab world (Lust-Okar, 2006; Kilani and Sakijha, 2002), Moroccan deputies serve as intermediaries between citizens and the government. By using written queries, however, they act via a formalized institutional mechanism rather than their personal connections. Additionally, though they are far less public and confrontational than oral queries and lack an explicit enforcement mechanism, written questions still allow deputies to pressure government ministries toward action. They may do this by repeatedly raising an issue, bombarding a given ministry with demands, and by advertising any non-response. Near the end of the 2011-2016 legislative term, USFP, a socialist opposition party, publicized the fact that it had asked more than 4000 questions that went unanswered.31 Some deputies also publicly share information about their questions and the responses received on their social media accounts. In practice, written questions are an important and eﬀective tool for deputies to provide constituency service. In interviews, deputies described using written questions to address “local rather than national” issues.32 MPs often focus their queries on issues related directly to their district or region, or even on the concerns of an individual citizen. One deputy explained the role these play in supporting a sense of government accountability among citizens as follows: “[written queries] allow the deputy to answer her constituents with a

29In practice, most questions are not answered within this time frame - see Section 1.4.3. Deputy interview, January 30, 2018. 30Deputy interview, February 5, 2018. 31See Reda Zaireg, “Plus de 8.500 questions de parlementaires au gouvernement restés sans réponse,” Huﬃngton Post, July 12, 2016 (https://www.huffpostmaghreb.com/2016/07/12/ questions-maroc-parlement_n_10943522.html, accessed May 10, 2018). 32Deputy interview, January 30, 2018.

26 written answer from the ministry concerned.”33 Another offered a succinct summary of their utility: “it’s a successful method for solving local problems.”34 Ultimately, then, written questions provide deputies with an institutional channel through which to serve constituent interests. As one of the legislative authorities specified in the constitution, it is not easy for the regime to remove this option, meaning that opposition parties that achieve seats in parliament have an established institutional channel through which to address district level concerns. Though there is no direct enforcement mechanism for this action, in practice, deputies find it to be one of the most effective tools available for working on behalf of their district. Additionally, despite the fact that the queries do not have direct legislative or redistributive implications, savvy deputies can deploy them strategically to induce more pressure on the government. If a ministry provides a slow or inadequate response, a deputy might follow up with a barrage of additional queries or publicize the nonresponse to his constituents, shaming the regime for its inaction.35 Though on paper, the additional visibility of oral questions appears more electorally useful for vote seeking deputies, the structural limitations on this type of query mean that written questions are much more practical as a means of providing the localized constituency service expected by voters. Perhaps the best evidence of written questions’ utility is the frequency with which deputies employ them: on average, deputies submitted more than 5000 unique written queries in total each year during the 2011-2016 term.

1.4.3 Questions submitted during the 2011-2016 legislative term

In this section, I use data on all oral and written questions submitted during the 2011-2016 parliamentary term to examine the practical use of this activity in Morocco and validate the diﬀerences in question type discussed in the preceding sections. The data were gathered via a database hosted on the oﬃcial parliamentary website, and include information on the type

33Deputy interview, February 13, 2018. 34Deputy interview, January 23, 2018. 35In a separate paper, I explore the factors that contribute to low response rates among ministries (York, 2020).

27 and content of questions asked, as well as to which ministry they were directed and whether or not they received a response.36 The focus here is on how opposition parties are able to provide constituency level assistance, and the qualitative descriptions from politicians make it clear that written questions are more suitable for this activity. In this section, I validate this characterization using descriptive information for both types of questions and coding of question content. Over the course of the 2011-2016 parliamentary term, MPs asked an impressive number of questions, submitting a total of 11,237 unique queries during oral sessions and 27,196 unique queries in written form.37 The distribution across deputies was highly dispersed, ranging from zero questions asked to thousands submitted by an individual. The sheer number of questions submitted is evidence of the importance of this activity, corroborating deputy assessments: if these actions were entirely ineffectual, it is not clear why legislators would produce so many. There is little evidence that the questions asked follow expectations related to the electoral cycle (Nordhaus, 1975; Healy and Lenz, 2014). While this would lead us to expect an uptick in political activity in the last years of the parliamentary term, the number of oral questions asked instead declines, and the number of written questions is far off the peak in 2014 (Figure 1.1).38 It is also noteworthy that the rate of reply to written questions dropped off steadily over the course of the term, so last minute queries were especially unlikely to receive a response.39 In total, 67% of written questions are recorded

36Though these data relies on self-reporting from an autocratic government, there is reason to expect that the record of queries is complete. Data accessed from this site match summary statistics reported elsewhere by the government. Transcripts from oral query sessions are public, meaning question details are available via secondary sources. Finally, there is little incentive for deputies to undercount their own legislative activity, and parliament itself is the source of this data. I therefore expect that the data collected represent the complete record of written and oral queries, excepting minor errors in record-keeping associated with a database of this size. 37Note that multiple deputies, always from the same party with a single exception, might endorse a given query. Most questions were submitted by solo deputies (92% of oral and 95% of written queries). 38The observable peak in questions submitted in 2014 is largely driven by questions from members of two opposition parties: the Islamist PJD and leftist USFP. 3988% of written questions submitted in 2011 were answered; the response rate fell in each subsequent year. In 2016 (the election year), only 26% of questions received a response. Because oral queries typically receive a real-time response from the minister in question, I do not track response rates for this type of query.

28 Oral Written Most 40 (23%), both units 2016 41 N/A 2015 Yes No 29 Date of Query 2014 Received Reply Received 2013 2012 0 0 Morocco uses the French system of administrative units. Communes are roughly equivalent to munic- Questions were coded using the quanteda text analysis package in R. To identify all georeferences, I Consistent with deputy characterizations, written questions are often used to address

100 400 300 200 500

2000 1000 4000 3000 Number of Queries of Number 41 40 of these references are to provinces (60% of all questions) or communes first constructed a dictionarythen of coded all all administrativequestions questions place that names that explicitly in contained address Morocco, oneand the including a concerns or alternate of dictionary more individual spellings. of placenames constituents common I – as honorifics. was including coded a using georeference. a similar Casework method – localized issues. I undertakeadministrative content units analysis in of Morocco the as query wellof as subjects, these placenames. coding references Figure for references 1.2 the to compares questionmajority the types, of frequency written revealing questions substantial – differences more in than 75% content. – include The a georeference of some kind. Figure 1.1: Number of Questionsand Submitted written by (bottom) Date. queries Figure submittedvertical depicts during the scales the number differ of 2011-2016 across oral parliamentary (top) thebased term; two note on panels. that whether Written or questions not are they also received distinguished a by recorded color as response. receiving a response from the government. 1.00

0.77 0.75

0.6

0.50

0.25 0.23

0.13 0.09

Proportion of oral/written queries with specified content reference 0.07 0.05 0.03 0.03 0.02 0.02 0 0.00

Nation Region Province Commune Georeference Casework

Oral (n = 11237) Written (n = 27196)

Figure 1.2: Oral and Written Question Content. Plot shows the proportion of unique oral and written questions with references to the speciﬁed category. Administrative units are listed in descending order of size.

at the district or subdistrict level.42 More rarely, written questions address larger regional units. Interestingly, though georeferences are more common among written than oral queries for all categories, the reverse is true for national references (speciﬁcally, questions that refer to ‘Morocco’ or ‘nation’), validating comments from politician interviews that these are used to address more universal considerations.43 For the most part, questions reference locations within a politician’s purview: among written questions with georeferences, 59% were local to a politician’s district.44 Additionally, it is noteworthy that roughly 3% of written questions clearly represent

ipalities. The ‘province’ designation used here includes references to both provinces and prefectures (the urban equivalent). 42Most legislative districts comprise a single province; in more densely populated urban areas, a province may be subdivided into more than one constituency. 43Deputy interview, January 30, 2018. 44For this statistic, I exclude questions asked by deputies elected as part of a national list – national list deputies lack linkages to administrative districtes and as such do not have a sub-national constituency. As a result, none of their queries are eligible to address ‘local’ issues. Queries submitted by these deputies were included, however, in other statistics related to georeferences.

30 casework on behalf of constituents, speciﬁcally referencing an individual issue (e.g. “Pension adjustment for Mrs. X” or “Suspension of monthly salary for Mr. Y”). Though these represent a relatively small percentage of the total questions asked, they provide some visibility into how, in practice, legislators perform what is an important aspect of their responsibilities in this context (Benstead, 2008).

1.4.4 Deputy activity

As discussed previously, there are no restrictions on the submission of written questions, and thus this action is equally accessible to all members of parliament. Yet the exercise of this authority varies hugely among individuals. During the 2011-2016 term, some deputies submitted thousands of queries over the course of the term, while approximately 15% did not submit a single written question. Clarifying the nature of this disparity and identifying which politicians submit more questions sheds light on the purpose and political utility of written questions. In this section, I combine data on query submission with information on individual deputies to examine patterns in deputy-level activity. I use the number of written questions submitted by each MP to further explore the use of this activity in practice, combining this with information on politician characteristics gathered via the parliamentary website as well as a local NGO focused on citizen-deputy relations. These data include binary identiﬁers for whether or not a deputy was a committee leader, elected as part of the youth or women’s national list, or a member of the preceding parliament elected in 2007. I also categorize deputies based on whether they belong to a royalist or opposition party45 and whether or not their party formed a part of the governing coalition from 2013-2016.46 Finally, I include an indicator for civic engagement, which I measure based on whether or not a deputy chose to activate a proﬁle on a web-based platform run by a local NGO to facilitate more direct communication between politicians and their

45See Table 1.4 in the appendix for categorization of parties. 46This coalition, formed after Istiqlal left the government to join the opposition, included deputies from the PJD (the largest party), RNI, MP, and PPS, as well as a few deputies from smaller parties.

31 Statistic N Mean St. Dev. Min Max Royalist 413 0.407 0.492 0 1 Governing Coalition 413 0.535 0.499 0 1 Committee Leader 413 0.061 0.239 0 1 National List 413 0.228 0.420 0 1 Female 413 0.167 0.373 0 1 Youth 413 0.082 0.275 0 1 2007 Parliament 413 0.247 0.432 0 1 Civic Engagement 413 0.131 0.338 0 1 Written Questions 413 82.985 252.858 0 3,884 Log Written 413 2.746 1.936 0.000 8.265

Table 1.1: Summary statistics for data on deputies in the 2011-2016 Moroccan parliament. Data were gathered from the official parliamentary website and www.nouabook.ma and include information on all individuals that served during the 2011 term. constituents.47 Summary statistics are shown in Table 1.1. Because the outcome variable (number of written questions) is overdispersed, I use OLS regression on a log transformation of this variable to estimate models of politician characteristics.48 Note that as a result, coefficients should be interpreted in terms of log-units of the outcome variable. Because party characteristics may predict deputy behavior, I also estimate standard errors using block bootstrapping. I first regress the number of questions on the royalist indicator alone, then estimate models including controls for other politician characteristics. Model output is shown in Table 1.2. Most striking is the fact that royalist deputies submit far fewer questions on average – more than two log units, or roughly an order of magnitude, fewer – than politicians from non-royalist parties. The coefficient on the royalist indicator changes very little with the inclusion of controls and is significant to the 0.01 level with errors are clustered at the party

47The website allows Moroccan citizens to submit questions directly to their elected representatives and allows deputies to write a public response. The currently active deputies and questions submitted may be viewed at http://nouabook.ma/ar/. 48To account for the politicians who asked no questions, I use ln(y + 1), where y is the number of written questions submitted by a given politician. Though there are limitations to the use of OLS with count data (O’Hara and Kotze, 2010), this modeling approach reduces the probability of Type I errors (Ives, 2015). For robustness, I include a negative binomial specification in Appendix Table A.2; the coefficient on the royalist indicator from this specification is similar in both magnitude and significance to the results from OLS regression on the log transformed variable.

32 Dependent variable: Log Written Questions (1) (2) (3) Royalist −2.131∗∗ −2.123∗∗ −2.134∗∗ (0.777) (0.765) (0.755)

Governing Coalition 1.117

Committee Leader 0.853∗∗ 0.896∗∗ (0.324) (0.337)

National List −0.357 (0.190)

Female −0.326∗∗ (0.114)

Youth −0.529 (0.303)

2007 Parliament −0.601∗ −0.479∗∗ (0.235) (0.163)

Civic Engagement 0.730∗∗ 0.777∗∗∗ (0.263) (0.233)

Observations 413 413 412 R2 0.293 0.335 0.419 Adjusted R2 0.291 0.327 0.408 Note: ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001 Table 1.2: Predictors of Question Submission in the 2011 Parliament. Models report coeﬃ- cients from OLS regression of a natural log transformation of the number of written questions submitted on politician characteristics. The unit of analysis is the deputy. Standard errors are clustered at the party level and calculated by block bootstrap to account for the fact that key attributes (e.g. being from a royalist party) are assigned at the party level.

33 level, providing strong evidence that opposition parties make substantially greater use of this authority. This is consistent with the argument that this institutional procedure might substitute for other means of providing constituency service: if deputies from royalist parties already have alternative avenues through which to satisfy voter demands, they do not need to avail themselves of this legislative privilege in order to be effective. Politicians from parties without connections to the regime and the related patronage networks, however, find this to be an activity worth investing in. The predictors for question submission also provide additional suggestive evidence that written questions are generally used to address constituency concerns. Individuals from the national list – in other words, deputies without obvious constituencies, geographic or otherwise – submit fewer questions (significant at the 0.1 level), while those deputies who have displayed an express interest in engaging with constituents via the Nouabook website submit significantly more questions on average. Other predictors include parliamentary leadership, which is associated with more questions, and incumbency, which associated with fewer.

1.5 Eﬀort and Electoral Accountability

The preceding sections present evidence that elected politicians make ample use of constitutionally granted authorities to address local, district-level issues, and that this is most used by individuals from parties lacking close links to the regime. But how do voters evaluate these eﬀorts? If the theoretical predictions outlined in Section 1.3 are correct, and these institutional avenues for serving constituents can serve as a substitute for patronage provided through personal links to the regime, then voters should reward politicians that engage in more constituent service through this mechanism. I expect constituency eﬀort to translate into electoral support: all else equal, voters in a district with a hard-working and responsive incumbent should prefer to reelect that candidate or the party he or she represents. I have argued that written questions in particular represent a form of institutionalized

34 constituency service: a majority address local concerns, and deputies themselves report that they offer a useful way to influence issues within their districts. To that end, then, the written questions submitted constitute an important indicator of one particular quality valued by voters: deputy effort on behalf of the district.49 Voters may not observe these queries directly, since they represent interactions between the legislature and the executive bureaucracy. Their assessment of deputy quality should nevertheless be derived from the elements of constituency service that involve interactions between the legislature and the electorate, such as deputy outreach via local offices within their district; accessibility via social media or other platforms; and responsiveness in addressing the concerns raised through these interactions. In other words, voters may not observe the totality of queries themselves, but they have opportunities to observe this effort and the resulting outputs. In some instances, this might occur through personal interaction, such as when a deputy query directly addresses an individual concern; in others, voters might learn of deputy efforts and results through political credit-claiming. Some deputies share the results of their effort via social media: for example, one PJD deputy currently serving his second term as a representative for Tinghir, regularly posts the responses received to his own Facebook page.50 Per deputy interviews, written questions are a critical tool for responding to voter demands and performing constituency service. Deputies who submit more of these questions should do better electorally, because the number of queries submitted proxies for constituency-focused effort. There are some contextual features that may disrupt this relationship: while in many electoral settings, retrospective voters can directly reward incumbents that performed at or above expectations and vote out shirkers, closed list PR gives parties significant control over incumbency. In Morocco, parties determine the content and order of candidate lists. Voters

49As such, written queries stand in contrast to other legislative activities undertaken within parliament, such as oral questions or proposed legislation, which are more likely to address national considerations and party priorities. Deputy interview, October 17, 2016. 50Given that this individual has more than 16,000 followers on the social media platform, it is likely that this is an eﬀective means by which to advertise his eﬀorts and the resulting government action.

35 then cast ballots in support of lists rather than individual candidates, and in some cases may be unable to reelect a hard-working deputy because the party has neglected to renominate that individual.51 The overall incumbency rate for the 2011 parliament was quite low – around 30% – though this can largely be ascribed to low rates of renomination by parties, particularly for deputies elected as part of the national lists.52 Party strategic considerations may impede voters’ ability to retrospectively evaluate individual politicians, and incumbents cannot independently decide to run again on the same party list. I therefore examine electoral outcomes for incumbent parties, analyzing party- district electoral results within a district as a function of the performance of the party’s deputies from that district during the 2011-2016 legislative term. This approach relies on the assumption that the electoral benefits of deputy performance are not exclusive to the individual and can also accrue to the party they represent. However, it allows for identification of effects across the entire population of incumbents, given that parties in 2016 ran lists in all districts where they held seats.53 I combine data on legislative activity with electoral outcomes from the 2016 elections, obtained via the official elections website.54 To assess activity at the party level, I generate a continuous variable for the number of questions submitted, pooling queries submitted by all representatives of the same party within the given district. This is again log transformed to

51In Section 1.2, I argue that constituency service is the most important consideration for voters deciding how to cast a ballot. Parties, on the other hand, must balance a variety of strategic considerations in determining which candidates to nominate. In addition to electability (the candidate most likely to appeal to voters), these may include candidate ideology and independent financing (in Morocco, campaign financing is extremely limited, so many candidates are required to support their own reelection effort). As a result, party incentives may not always align with those of voters. 52National level data on candidate lists is not published by the Moroccan ministry of the interior. In the run-up to the 2016 elections, I gathered information on candidate lists for seven of the largest parties and matched them to the 2011-2016 parliament. Lists were not available for the Constitutional Union party, and some parties were only able to provide information on list heads. Only 40% of deputies from the 2011 parliament were renominated by their party as head of a list in the 2016 elections; this figure is driven in part by extremely low (10%) rates of renomination amongst deputies from the youth or women’s national lists. 53In Appendix A.3, as a robustness check, I look at outcomes (incumbency rates) at the deputy level, conditional on party nomination. Results are consistent in sign and significance with the main findings from the party-district analysis. 54Electoral returns from the 2016 legislative elections and 2015 municipal and regional elections, aggregated to the constituency level, are available at http://www.elections.ma/.

36 account for the high level of dispersion. I generate this measure separately for written and oral queries. For historical comparison and estimation at the party level, I use 2011 electoral returns, including party level vote share and number of seats won.55 Does having higher-performing deputies improve electoral prospects at the party level? I use OLS regression to evaluate the relationship between written questions submitted and electoral outcomes and study whether parties are rewarded for their representatives’ effort in office; results are shown in Table 1.3. For models 1 and 2, the outcome is party level vote share in 2016 as a function of activity by the party’s deputies from a given district, controlling for the number of seats the party occupied (model 1) and its 2011 vote share (model 2). For models 3 and 4, the outcome is the number of votes a party received, controlling for the number of seats or number of votes obtained in 2011. The results are strongly supportive of the idea that voters reward constituency service: the use of written questions by a party’s deputies has a positive and significant association with both vote share and number of votes received in a district. A log unit increase in the number of written questions asked is associated with a 1-2 percentage point increase in the party’s expected vote share – a substantively important difference in a setting in which a vote share in the single digits is usually sufficient to earn a seat. To contextualize this effect, in the 2016 elections, the winner of the last seat in a quarter of districts was determined by less than one percentage point; in 41% of districts, the margin was smaller than two percentage points.56 Interestingly, oral queries do not have the same positive association – more oral queries are associated with lower vote share and fewer votes, though these coefficients estimates are not significant at the conventional α = 0.05 level. At the least, this suggests that voters do not reward parties for engaging in the more public oral query sessions. While somewhat

55Because these results are no longer available on the government elections site, I obtained them from Tafra, an NGO focused on civic information and elections that hosts records from previous election years. Data were downloaded from http://tafra.ma/data2002-2016/ on April 1, 2017. 56Morocco uses multimember districts with between two and six seats. I estimate this margin by comparing the vote share of the strongest loser to the weakest winner in each district.

37 Dependent variable: Party Voteshare 2016 Party Votes 2016 (1) (2) (3) (4) Log Written 0.021∗∗∗ 0.010∗∗∗ 1,243.943∗∗∗ 497.671∗ (0.004) (0.003) (286.331) (199.909) Log Oral −0.015∗∗ −0.007 −517.782 −383.174 (0.005) (0.004) (390.233) (267.501) Number of District Seats 0.139∗∗∗ 10,530.670∗∗∗ (0.019) (1,364.299) 2011 Voteshare 0.934∗∗∗ (0.051) 2011 Votes 1.167∗∗∗ (0.057) Mean Voteshare 0.19 0.19 Mean Votes 12518 12518 Observations 263 263 263 263 R2 0.307 0.631 0.296 0.669 Note: ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001 Table 1.3: Legislative Activity and Electoral Outcomes. Coeﬃcients represent estimates of OLS regression of party-level vote share (models 1 and 2), and party level number of votes (models 3 and 4) on legislative activity by party deputies within a district. The unit of observation is the party-district.

38 counterintuitive, this may result the emphasis in oral query sessions on broader, nationally- focused policy issues and from a voter belief that addressing national considerations takes deputies’ time and attention away from constituency-level issues. Such an explanation is consistent with other ﬁndings that voters who preference constituency service may perceive national and local legislative activities to be substitutes (Wantchekon, 2003; Fujiwara and Wantchekon, 2013; Adida et al., 2016).

1.5.1 Opposition-speciﬁc electoral beneﬁts

In Section 1.3, I argue that both opposition and regime-affiliated parties seek to deliver similar services in order to attract electoral support, but while regime parties can exploit their connections to deliver patronage and satisfy voter demands, opposition actors must work through the institutions in which they are participating. Consistent with this line of reasoning, Section 1.4.4 demonstrated that legislative authorities are used with differing frequency by the two types of politician. In this section, I explore differences in how voters respond to this behavior for royalist and non-royalist parties. To do so, I estimate differential effects for royalist and opposition parties of written question submission on the change in district-level vote share (Table 1.4). For regime-affiliated parties (models 1 and 4), constituency service via written questions was not associated with improved electoral prospects in 2016: the coefficients on activity are small and insignificant for both changes in vote share and number of votes. In contrast, there is a strong, positive association between legislative activity and votes for opposition parties (models 2 and 5), suggesting that voters respond favorably to more institutional effort in providing constituency services for these political actors. A log unit increase in the number of written questions submitted by district representatives is linked to a 1.5 percentage point increase in vote share and more than 1000 additional votes for the relevant party. The results of activity are very different for the two types of parties, as confirmed by the interaction terms in models 3 and 6.

39 Dependent variable: Change in Voteshare 2011-2016 Change in Votes 2011-2016 (1) (2) (3) (4) (5) (6) Log Written −0.005 0.015∗∗∗ 0.015∗∗∗ 17.683 1,123.253∗∗∗ 1,123.253∗∗∗ (0.007) (0.003) (0.003) (471.154) (218.720) (233.308) Royalist 0.067∗∗∗ 4,444.005∗∗∗ (0.019) (1,317.750) Log Writ*Royalist −0.020∗∗ −1,105.570∗ (0.007) (496.362) Constant 0.009 −0.058∗∗∗ −0.058∗∗∗ 2,443.282∗∗ −2,000.723∗ −2,000.723∗ (0.013) (0.014) (0.014) (931.971) (930.626) (992.696) Sample Royalist Opposition All Royalist Opposition All Observations 115 148 263 115 148 263 R2 0.005 0.132 0.073 0.00001 0.153 0.083 Note: ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001

Table 1.4: Legislative Activity and Electoral Outcomes by Party Type. Coefficients represent estimates from OLS regression on the change in party vote share (models 1-3) and the change in party votes (models 4-6) within a district. The unit of observation is the party-district. Models 1 and 4 reflect estimates for royalist parties only, while models 2 and 5 reflect estimates for opposition parties.

The results in Table 1.4 offer clear support for the assertion that opposition parties’ use of their institutional authorities is important in generating electoral support. Voters in Morocco expect their representatives to prioritize constituency service, and written queries offer an important avenue for opposition legislators in particular to ensure bureaucratic attention for their districts. For the opposition, taking advantage of these regularized avenues for service provision translates into increased electoral success. Failing to do so is associated with electoral losses: in districts where their deputies submitted no written questions, opposition parties did worse in 2016 relative to their performance in 2011, as evidenced by the intercept terms in models 2 and 5. But how should we interpret the null relationship for regime-affiliated parties? For royalist parties, there is no link between written queries and electoral gains, which may help explain why, on average, they engage in less of it (Table 1.2). But it is noteworthy that while such parties do not benefit from submitting more written questions, they also are not punished for submitting fewer questions. This is consistent with the supposition that

40 institutionalized constituency service oﬀers a substitute for connection-based patronage. In the preceding section, I posit that citizens evaluate deputy responsiveness in addressing local concerns in choosing how to vote.57 If regime-linked politicians satisfy voter demands through other means, then it may simply be the fact that institutionalized action is a bad proxy for constituency eﬀort for this type of politician. On the other hand, if written queries are the primary way for opposition deputies to hold the regime accountable in their district, then we would expect them to be very closely linked to electoral support. In other words, the broad consensus in the literature that regime-backed candidates win support through non-programmatic, under-the-table distribution of favors remains unchallenged by the results shown here. However, the evidence in Table 1.4 suggests that opposition candidates use by the book, institutionalized action to satisfy the electorate.

1.6 Localized Support for Local Eﬀort

In this section, I examine a possible mechanism underlying the relationship between written queries and vote share. In the preceding sections, I find that written queries are associated with electoral gains, especially for opposition parties. I argue that this is not the result of voters directly observing deputy actions within parliament, but rather that written queries proxy for deputy effort on behalf of their constituency. In other words, the results in the preceding sections do not simply represent a case in which voters reward good, hard-working politicians – but, more specifically, good politicians that expend their effort to provide locally-focused service. I have shown that written queries are primarily used to address local concerns. I argue that voters are not responding to effort in general: if they were, we would also expect to see increased vote share for other forms of parliamentary effort, such as oral queries. Instead, voters are responding to localized effort that benefits their community. If this is the case, we might expect corresponding electoral support to vary across communities based on the level

57I provide further evidence that this is the mechanism underlying the relationship between written queries and vote share in the next section.

41 of attention each receives from a given deputy. Legislative districts in Morocco typically comprise multiple municipalities (communes), spread across a large geographical area, and busy deputies may not divide their efforts evenly across their constituency. I expect that the more attention a community receives from its MPs, the more motivated it will be to support that party in elections. And local references within written queries provide a useful metric of this local attention. To test this, I extract municipal references from written questions to create a metric of subdistrict effort by party deputies. I then use this to construct a variable at the party- municipality level summarizing the number of questions that address a given community. I combine this with electoral data to test whether more questions improve party vote share at the local level. One challenge in evaluating this mechanism is that the Moroccan government does not release subdistrict electoral data, so it is not possible to examine subdistrict variation in support for MPs during a legislative election. However, in September 2015 – midway through the 2011-2016 parliamentary term – Morocco held local elections for which electoral outcomes were published at the municipal level.58 The municipal electoral data allows for subdistrict analysis on party performance. I therefore examine whether local attention by party deputies corresponds to increased electoral support in midterm local elections. This party-level analysis again requires the assumption that voters evaluate and support political parties on the basis of the performance of their legislative deputies. As elsewhere, I conduct OLS regression to examine the relationship between municipality- specific activity and party vote share. The predictive variable of interest is the number of questions raised with references to a given municipality, either by all party deputies or exclusively by those elected to represent the relevant community.59 A municipal-focused question

58Municipal elections in Morocco are held every six years. Smaller municipalities elect municipal council members via a first past the post system, while larger municipalities use PR with party lists. 59Because municipal elections are not direct referenda on legislative performance, I consider two different measures of party effort. The first combines all party activity on behalf of a municipality, while the second restricts attention to activity by party deputies representing that district.

42 that was submitted collectively by more than one deputy is coded as a single reference, and a local reference if at least one of the submitting deputies represented the relevant district. I sum all references and local references to generate the measure of municipal-level effort and use the log transformation of this measure as the main predictor in the specifications that follow.60 Some municipalities share toponyms with higher level administrative units; for example, the municipality of Settat is located in the province of Settat. This generates some ambiguity as to whether a question is in reference to the subdistrict or district entitiy. To ensure that results are not driven by mis-attributed references, I include specifications that subset the data to include only municipalities with unique toponyms for which no ambiguity is possible. In all specifications, I include controls for log population and the number of parties competing. I also include party-district fixed effects to absorb other activity by local deputies; the model comparison is thus across localities within a district. Finally, to address potential concerns regarding reverse causality, such as the possibility that parties might direct the majority of resources and effort to core supporters, I include a control for the party’s share of seats won in the 2009 municipal elections.61 Results are presented in Table 1.5. Models 1 and 2 regress municipal vote share on the log number of unique questions submitted by any party deputies on behalf of a municipality. Models 3 and 4 instead use the log number of unique questions submitted by the municipality’s actual elected representatives.62 Models 2 and 4 subset data to include only municipalities that do not share a name with a province. In all models, the coefficient on municipal references is positive and significant to the 0.001 level: municipalities show increased electoral support for parties that use written

60As before, the transformed variable is equivalent to ln(y + 1) to avoid dropping observations where no questions were submitted. 61Ideally, I would control for each party’s vote share rather than the somewhat noisier measure of seats won. Unfortunately, the Moroccan government did not release vote totals at the municipal level for the previous municipal elections (Pellicer and Wegner, 2013). 62Note that this produces a small (approximately 5%) reduction in sample size, as it requires the omission of districts for which a party did not win seats in 2011.

43 Dependent variable: 2015 Municipal Voteshare (1) (2) (3) (4) Log References (any) 0.013∗∗∗ 0.017∗∗∗ (0.003) (0.004)

Log References (local) 0.024∗∗∗ 0.035∗∗∗ (0.005) (0.007)

Log Pop −0.002 −0.002 −0.010∗ −0.009 (0.003) (0.003) (0.005) (0.005)

No. Parties −0.015∗∗∗ −0.015∗∗∗ −0.016∗∗∗ −0.016∗∗∗ (0.001) (0.001) (0.002) (0.002)

2009 Seatshare 0.279∗∗∗ 0.276∗∗∗ 0.295∗∗∗ 0.289∗∗∗ (0.009) (0.010) (0.014) (0.014)

Party-District FE Yes Yes Yes Yes Unique Toponym Subset No Yes No Yes Observations 8,179 7,795 3,519 3,377 R2 0.384 0.384 0.371 0.370 Adjusted R2 0.331 0.329 0.329 0.326 Note: ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001 Table 1.5: Effect of Local Activity on Municipal Vote share. Coefficients represent estimates from OLS regression on party vote share at the municipal level. Models 2 and 4 are conducted on subsets of the data with unique toponyms. All models include party-district fixed effects.

44 questions to address local issues. Notably, coefficients are largest when data are limited to local deputies: the model predicts that a log unit increase in local questions submitted by a municipality’s deputy corresponds to an additional 2.5-3.5 percentage points in party vote share. This may reflect the greater visibility of such deputies within the district. Taken collectively, the results in Table 1.5 provide evidence that voters are most supportive of parties when deputies’ effort is expended directly on behalf of their locality.

1.7 Regime Strategic Considerations

The preceding empirics provide evidence that, in Morocco, legislative activity related to local concerns and constituency service is linked to voter support in future. Yet it is important to consider deputy activity and voter response in light of the autocrat’s strategic considerations: why does the regime allow opposition parties such a viable path to voter support? If these legislative actions allow the opposition to overcome their structural disadvantages relative to regime-backed parties, we might expect the regime to take steps to block such actions or reduce their eﬃcacy in the eyes of voters. One possible explanation is that even in autocracy, leaders are bound by the institutions they have created. Though constitutions may give the regime theoretically unlimited discretion (such as in declaring a state of emergency that allows suspension of existing laws or elected bodies), such actions are potentially costly if they inspire defection rather than cooperation with the existing political environment or if they increase grievances and thus the risk of uprising. In other words, actions that directly contravene existing institutions are risky and may have lasting implications for economic growth and regime survival. We would thus expect such actions to be deployed infrequently and only in cases where there is a direct threat.63 Alternately, it may be the case that the regime actually beneﬁts from this behavior

63Though there may be larger costs to removing these constitutionally allocated authorities, we might expect that the regime ﬁnds subtler ways to obstruct action by opposition parties. I explore this possibility in Chapter 2 and ﬁnd evidence that regime-appointed technocrat ministers are less responsive to these deputy requests than partisan members of the cabinet.

45 by opposition actors. Other scholars have hypothesized that competitive elections among members of the selectorate allow the regime to farm out responsibility for citizen satisfaction (Blaydes, 2010). The autocrat may prefer to have these activities undertaken by ideologically aligned actors, but he must balance an array of strategic considerations, including those that allow for the inclusion of moderate opposition members (Lust-Okar, 2005). As long as opposition actors are not posing a direct existential threat to the regime, their actions in support of the electorate may produce a net beneﬁt for the state.64 The importance of these actions is especially visible in Morocco’s rural areas, where economic neglect has led to widespread protests in recent years: the list of demands from participants in the ongoing Rif Movement include increased regional investment in health and manufacturing.65 By allowing the opposition to take on responsibility for constituency service in exchange for votes, the regime reduces its own accountability in the event of dissatisfaction, as frustration over government incompetence may be directed at parliament rather than the palace. Additionally, the fact that voters reward extremely localized, municipal-focused action suggests that vote-seeking parties must spend the bulk of their energy and resources on such activies rather than others that would address their partisan agenda. To this end, it seems that the regime has been successful in corralling the opposition into competing on an electoral playing ﬁeld that privileges local responsiveness over national policy concerns. If voters perceive that one action substitutes for the other, they may even punish parties that seem too focused on policy, alleviating a potential source of pressure for the regime.

1.8 Conclusion

This paper seeks to explain legislative behavior and voting patterns in an autocratic context, with an emphasis on the distinction between regime-aﬃliated and opposition parties.

64This threat might stem from policy demands (opposition actors that call for the removal of the head of state) or level of popular support (an opposition party that achieves a large majority of support). In Morocco, the regime beneﬁts from a diverse set of opposition actors that divide the anti-regime vote and struggle to form coalitions. 65“Morocco’s unrest is worsening,” The Economist, July 8, 2017.

46 One underlying assumption, corroborated by a broad body of scholarship as well as by interviews with Moroccan politicians, is that a majority of voters in this context expect their elected representatives to focus on constituency level concerns. While individuals with close ties to the regime might do so through their personal connections, I offer evidence here that opposition politicians use the legitimate authorities of their office to attend to the concerns of their district and pressure other government entities to ensure they are not neglecting constituents. Though the authorities examined here – written and oral queries directed at government ministries – are common to parliamentary systems globally, their relevance for legislative performance in an autocratic setting is neither trivial nor obvious. Yet I find evidence that these institutional features present the opposition with a particularly useful way to provide constituency service and attract electoral support. These results support the idea that institutions regularize certain aspects of political competition and help to level the playing field between regime insiders and outsiders. Because of these institutional opportunities, opposition parties can hope to achieve parity with or even surpass regime-affiliated candidates by maintaining high levels of effort in office and relentlessly pressuring government ministries to follow through on projects relevant to their district. And yet they may do so at the expense of their policy agenda: the results here suggest that voters are less supportive of activities that address national issues. One intriguing implication of this path to victory is that by demonstrating the legitimate powers of their office, opposition actors may succeed in developing institutional trust amongst the electorate. Voters who observe that these outsider parties are using their office in legitimate ways (in contrast to the rent-extracting methods of regime insiders) may place more trust in the legislature to represent their interests. This could in turn have broader ramifications for their ability to build support for political change – and to attract continued support following democratization. If regime-linked parties are dependent on networks of power and less capable of exploiting legitimate authorities, they may be disadvantaged when

47 those personal networks are disrupted.

48 Chapter 2: Horizontal Accountability under Autocracy

2.1 Introduction

Why does the opposition under autocracy allow itself to be coopted, given the minimal returns and ample associated risks? And how does the presence of technocrats – non- partisan political figures accountable only to the regime – affect governance? The literature on autocracy holds that autocrats create institutions in order to facilitate bargaining with the political elite: in exchange for cooperation, they offer rents and minor policy concessions via competitive elected institutions (Gandhi and Przeworski, 2006; Magaloni, 2008). And yet, much of the empirical literature suggests that the actual policy influence of outside participants is very minimal, with autocratic legislatures and ministers from outside the regime dismissed as “rubber stamp” actors (Sweet, 2001; Lust-Okar, 2006; Blaydes, 2010). If this is the case, we must revisit the incentives of opposition actors to consent to cooptation: are there additional benefits to engaging with the regime? In this paper I study the ramifications for horizontal accountability of outside (non- regime) participation in government in an autocratic setting with parliamentary institutions. The government entity of interest is the cabinet, an indirectly elected body: legislative elections determine the weight each party brings to the negotiating table and, in this case, the partisan identity of the head of government, but ministerial appointments are subject to regime interference. As a result, key ministries are held by non-partisan ministers, typically technocrats, selected by and therefore linked to the regime. Scholars have speculated about the importance of this appointment power in autocratic settings (Mauceri, 1997; Schleiter, 2013), but little systematic evidence has been brought to bear on the question of their impact on political interactions. Here, I examine how ministerial appointments affect executive

49 responsiveness to legislative requests. Autocratic constitutions today borrow heavily from democracies; the result is a set of governing institutions that superficially resemble those of their democratic counterparts. And yet there are important differences. In particular, the formal checks and balances that establish horizontal accountability in democratic settings are often absent or weakened under autocracy. As a result, competitively elected bodies are politically sidelined in favor of institutions over which the regime exerts more control – such as the aforementioned ministerial posts. Though there do exist systems of oversight between the elected legislature and executive branch, these lack the formal enforcement mechanisms that gird democratic institutions. I therefore argue that the accountability of the executive branch to the elected legislature is dependent on the incentives of participants, and that ministers will respond to requests only if it is in their own interests. Non-partisan ministers owe their position to the autocrat that appointed them rather than to elected legislative deputies. And the regime may have an incentive to subvert this system of horizontal accountability: by refusing deputy requests, it can diminish legislative effectiveness and potentially reduce support for this elected institution among citizens. On the other hand, by responding to requests, partisan ministers improve deputies’ effectiveness in addressing constituent issues, so I expect them to be more responsive to legislative demands. I test these hypotheses by examining executive response to legislative requests during the 2011-2016 parliamentary term in Morocco, a hybrid regime with a relatively long history of opposition participation in government. To address potential concerns over selection issues in the data generating process, I use both matching and difference-in-difference approaches to identify predictors of ministerial responsiveness. I find substantively significant differences in responsiveness as a function of minister identity. Questions submitted to ministries headed by regime appointees (technocrats from technical sectors or the civil service that lack a partisan affiliation) were 20-30 percentage points less likely to receive a response than comparable

50 requests submitted to partisan-led ministries. Technocrat ministers also responded more slowly than their partisan counterparts. The results here provide systematic evidence that regime control over ministry appointments undermines legislative effectiveness in practice. On the other hand, I do not find evidence that partisan alignment predicts increased responsiveness: partisan ministers do not appear to preference requests made by their co- partisan deputies. Though there are multiple possible explanations for this null finding, it may reflect the fact that this activity takes place in an environment requiring substantial inter-partisan cooperation. In the history of the Moroccan parliament, no party has ever won an outright majority; legislative effectiveness for a given party therefore requires the support of others in office. This fragmentation may create an incentive to respond to requests from deputies of all stripes. These results have a number of implications for our understanding of institutions under autocracy. First, they reveal an important benefit to outside participation in government and a possible explanation for opposition willingness to be coopted by the regime. Though there are other gains to office – including patronage benefits (Truex, 2014; Reuter and Robertson, 2015) and the possibility of negotiating with the regime over even minor policy adjustments – the results here suggest that non-regime participation in government helps to ensure responsiveness from the executive branch and thereby facilitate legislative function. While ceding control to regime appointees allows opposition actors to distance themselves from unpopu- lar policies, it runs the risk of decreasing legislative effectiveness and, relatedly, constituent satisfaction. As the opposition weighs the decision to join in a coalition government, it may take into account the fact that outside (partisan) cabinet control is crucial for ensuring accountability between branches of government. Additionally, these results provide hard evidence of the impact of technocrat appointees in autocratic settings. Other scholars have speculated that ministry control is an important source of autocrat authority in Morocco and elsewhere (Maghraoui, 2018; Mauceri, 1997; Yom, 2009; Lee and Schuler, 2019). Here, I show that technocrats are less likely to cooperate

51 with institutions of horizontal accountability to the legislature; their presence thus serves to undermine instruments of legislative effectiveness on behalf of constituents. Finally, the empirical evidence here demonstrates that superficially democratic institutions function differently under autocracy. This paper represents an effort to extend scholarship on autocratic institutions beyond explaining the existence of such institutions and instead explores the way they operate in practice. The legislative authorities examined here are wholly analogous to those found in other parliamentary settings, but I find that they are dealt with quite differently by the executive in this context. And though the focus in this paper is on executive accountability in a monarchic setting, the results likely hold for other autocratic structures so long as they include the possibility of coalition government with the regime or ruling party. In other words, in scenarios with weak institutions, ceding control to regime appointees can lead to meaningful obstruction to elected deputies’ ability to serve their constituents. The paper proceeds as follows: in the next section, I describe the structure of government in Morocco and institutions of legislative oversight. I then outline my argument regarding horizontal accountability in autocratic institutions and the testable hypotheses that follow. I describe the main empirical strategy and data sources used to test the argument and present results from these analyses. Finally, I discuss the limitations and broader implications of these findings for governance in settings with weak institutions.

2.2 Moroccan Institutionalism

Morocco is a constitutional monarchy often classiﬁed as a hybrid regime due to its use of superﬁcially democratic parliamentary institutions (Diamond, 1999). The king is head of state and possesses the bulk of political power, even after constitutional reforms in the wake of the Arab Spring protests (Maghraoui, 2001). The prime minister is head of government and appointed by the king.1 The executive branch is rounded out by a cabinet of around 30

1In previous years, the king had substantial discretion in selecting a prime minister; as a result of constitutional reform in 2011, he must now be appointed from the party that received the most votes in legislative

52 ministers and delegate-ministers. The significance of ministry control for governance is the subject of some scholarly debate (Darif, 2012). As elsewhere, ministers in Morocco oversee policy implementation on a national level, though the perceived importance varies with the segment of the bureaucracy concerned (for example, the Ministry of the Interior has a more politically salient portfolio than the Ministry of Handicrafts). Together with the prime minister and the king, they form the Council of Ministers – the supercabinet entity that initiates roughly half of all legislation, including the annual bill of finance. Yet the king presides over these policy discussions, meaning it is essentially impossible for opposition cabinet members to submit legislation to which the palace does not consent (Madani, Maghraoui and Zerhouni, 2012).2 The king also has considerable influence over cabinet formation. Following legislative elections, the king appoints a prime minister from the leading party and tasks him with forming a government. But the king is also involved in selection and approval of the roster of cabinet members, a departure from cabinet negotiations in more democratic settings. This constitutionally-enshrined ability to appoint (in consultation with the prime minister) and dismiss cabinet members has long been seen as an important source of authority, leading to frequent unofficial palace interference in negotiations over government formation. And this interference is highly targeted. In particular, the palace has made a point of controlling appointment to the most important ministries, which have generally included the ministries of the interior, foreign affairs, justice, and Islamic affairs (Sater, 2003; Darif, 2012).3 Palace control of the so-called ‘ministries of sovereignty’ is a practice with a decades- long history.4 Per McDonnell and Valbruzzi (2014), a technocrat is a government appointee

elections (though it is not required that he be the leader of that party). 2Per the 2011 Moroccan constitution, the Council of Ministers, led by the king, has direct authority over the Council of Government, a subsidiary body led by the prime minister. 3This list includes other ministries on occasion, such as the Ministry of National Education. 4The concept of ministries of sovereignty is comparable to practices elsewhere throughout the Arab world in which the head of state exerts control over cabinet formation (Herb, 2002). Though the name suggests it is specific to monarchies, we might expect similar dynamics in any setting in which the de facto executive has outsize influence over cabinet appointments. For example, the head of state often has unique control over the defense ministry and typically appoints military technocrats rather than partisan affiliates to head this body.

53 that lacks a current or former party affiliation and who possesses non-partisan expertise relevant to his appointment. I adopt that definition here, and additionally note that such non-partisan actors are, in this context, considered to be regime nominees. All Moroccan governments formed since independence have included at least one palace-selected technocrat minister (Storm, 2007), typically appointees from the civil service, education, or technical sectors that lack partisan linkages.5 But cabinet composition in Morocco has also evolved over time from a posse of technocrats (Willis, 2002b) to incorporate more members of active political parties. The 1998 parliamentary term saw the first gouvernement d’alternance (government of change), in which the leader of a prominent opposition party served as prime minister. Every subsequent cabinet has included representation from the secular or Islamist opposition. How does the presence of both technocrat and partisan ministers affect governance? I explore the ramifications for executive accountability to the elected legislature.

2.2.1 The elected legislature and executive queries

Legislative elections, scheduled to occur every five years, constitute the main arena of national political contestation, and the lower house of parliament is the only state-level body that is directly elected.6 The palace has no official partisan representation, though several parties are seen as having an affiliation with the monarchy due to personal linkages or a connection to the historical elite.7 There is also a large but fragmented opposition, which has expanded over the years to include secular leftist and centrist as well as Islamist parties.8 In recent election cycles, royalist and non-royalist (opposition) parties have achieved

5In addition to meeting technical criteria for technocrats, such ministers are referred to as such elsewhere in scholarship on cabinet appointees (Zerhouni, 2014; Maghraoui, 2018). 6Though the upper house, an indirectly elected body, also has a role to play in the passage of legislation, the focus of this paper is on interactions between the lower house, or house of representatives, and the cabinet. I therefore use the term ‘legislature’ throughout to refer to the lower house. 7One example is the Party of Authenticity and Modernity, or PAM, created in 2008 by a close friend of King Mohammed VI. See York (2019) for more discussion of royalist party labels and the diﬀering ways royalist and opposition oriented parties behave in oﬃce. 8Because this is an autocratic setting, I use ‘opposition’ to indicate political actors from outside the regime that seek policy changes from the status quo advanced by the regime (Lust-Okar, 2005). In the

54 comparable levels of support. In the 2011 elections – the time period under study here – royalist parties won 46% of valid votes. Yet the Moroccan legislature, though a ‘focal point’ for opposition actors, is comparatively restricted in its power, as in other autocracies (Levitsky and Way, 2002; Schedler, 2002). Autocratic legislatures are often described as ‘façade’ democratic institutions or dismissed as rubber stamp bodies (Levitsky and Way, 2002; Blaydes, 2010) due to their lack of authority relative to the executive: their policy decisions can be reversed, and they can often be dissolved at the autocrat’s discretion (Gandhi and Przeworski, 2006; Gandhi, 2008). The constitution grants the legislature a number of responsibilities and authorities, including law-making power, but provides it with few checks on the executive. And in Morocco, both the directly elected legislature and indirectly appointed cabinet are subsidiary to the monarchy (Waltz, 1995). The king has the ability to dissolve the elected legislature before its term is up, though this authority is rarely exercised. Similarly, cabinet members serve at the king’s discretion. This has led scholars to argue that, despite their similarities to democratic structures, these institutions do not constitute real checks on the regime (Sweet, 2001) and that “real political power...lies in the palace” (Haddadi, 2002). Though the legislature’s authorities with respect to the cabinet are also limited (Madani, Maghraoui and Zerhouni, 2012), avenues of executive oversight do exist that create the potential for horizontal accountability. Among other authorities, legislative deputies are constitutionally empowered to query government ministries both orally, in publicly televised sessions, and via written submission.9 Written and oral queries to ministers are common in both democratic and autocratic parliamentary settings10 – these are directly analogous to

Moroccan context, where regime affiliation indicates royalist sentiment, I use the terms ‘regime-affiliated’ and ‘royalist’ interchangeably to indicate regime supporters. Likewise, I use ‘non-royalist’ and ‘opposition’ to describe parties that lack a regime affiliation. Note that in settings with parliamentary institutions, ‘opposition’ sometimes refers to parties outside the governing coalition; to avoid ambiguities in this paper, I use ‘government’ and ‘non-government’ where necessary to indicate whether or not a party is a participant in the governing coalition. 9In addition to these authorities, the constitution grants the legislature the ability to implement a vote of censure and force the resignation of the cabinet; this activity requires a collective effort, however, unlike the queries which can be submitted by individuals. 10Scholars have explored the use and implications of these actions as an example of political behavior

55 authorities granted to MPs in the UK parliament, for example. In the UK and elsewhere, MPs use these queries for information gathering, to raise the proﬁle of issues important to their constituents, and to pressure ministers to address problems (Rogers and Walters, 2013). MPs in Morocco use them in much the same way: in interviews, deputies described queries as a useful means to gather information about issues important to their constituents and an opportunity to extract public promises from ministers.11 Deputies describe them as an important component of their work in oﬃce, especially as a means to provide constituency service,12 and there is evidence that written queries in particular are rewarded electorally (York, 2019). Scholars argue that ministerial queries constitute a meaningful mechanism of horizontal accountability between branches of government in many settings (Martin, 2011; Herron and Boyko, 2015). These queries should represent an opportunity for bureaucratic oversight: deputies use them to guide policy implementation and ensure that their district is not neglected by the administration. But in Morocco, this institution has limitations that are not present in more democratic settings. The ability to query the executive is constitutionally enshrined, and these actions are legally entitled to a ministry response – yet in practice, this is often not the case. Deputies lack an enforcement mechanism when their question is ignored: there is no precedent for censuring a negligent minister.13 Though in interviews, deputies complained that the long wait times and frequent lack of response exceeds what is legally expected, they also accepted them as standard practice.14 And many queries go unanswered; while in the UK, ministers respond to nearly all

and executive oversight in many democratic contexts: Russo and Wiberg (2010) conduct a comparative examination of 17 European parliaments; Akirav (2011) studies the use of questions in Israel; and Salmond (2004) examines the implications of queries for New Zealand opposition parties. Research in autocracies and hybrid regimes is more limited, but includes such cases as Vietnam (Malesky and Schuler, 2010) and Turkey (Hazama, Faruk Gençkaya and Genckaya, 2007; Ciftci and Yildirim, 2019). 11Deputy interview January 30, 2019 12Deputy interviews, January 23, 2018 and February 12, 2018. 13As previously noted, the legislature does have the ability to collectively force the resignation of the cabinet through a successful vote of censure, but this requires a large collective action eﬀort and is an outsize response relative to the action of ignoring a deputy query. 14Deputy interview, January 30, 2018.

56 queries,15 ministers in Morocco left approximately one third of the written queries submitted unanswered through the end of the 2011 parliamentary term.16 Response was generally quite slow: in most cases, deputies waited more than half a year for those that did receive a response.17 This lack of response elicits frustration from party leadership: one political party went to the press to complain about the number of pending queries it had submitted.18 These complaints reveal that though the Moroccan legislature has the same avenues for holding the cabinet accountable for its performance as in democratic parliamentary settings, the eﬀectiveness of these actions are hampered by the lack of formal enforcement mechanism. In the next section, I develop a theoretical explanation for variation in the degree to which ministers respond to deputy requests.

2.3 Executive Oversight with Weak Institutions

The Moroccan institutions described in the preceding sections illustrate a common feature of autocracies today: constitutional rules create an “array of democratic-looking institutions” (Sweet, 2001) that lack the checks and balances that are a crucial aspect of democracy. In- stead, they privilege non-competitive, regime-appointed positions over those more subject to voter choice. Though the Moroccan legislature is the source of substantial political competition – it includes a diverse set of opposition parties that routinely succeed electorally – it is seen as a rubber stamp institution (Waltz, 1995), largely due to its lack of inﬂuence over the other branches of government. I argue that the subordination of competitive bodies means that executive oversight is

15During the 2003-2004 parliamentary session, for example, MPs in the UK House of Commons tabled nearly 50000 written queries. Only 819, or 1.64%, went unanswered. From the “Ministerial Accountability and Parliamentary Questions: Fifth Report of Session 2004-2005” report produced by the Public Adminis- tration Select Committee of the House of Commons, March 17, 2005. 16During the 2011-2016 term, MPs submitted 27196 written queries; responses are recorded for only 18361, or 67.5%. 17Among written queries that received a response, the median time between submission and response was 187 days. 18Reda Zaireg, “Plus de 8.500 questions de parlementaires au gouvernement restés sans réponse,” Huﬃng- ton Post, July 12, 2016 (https://www.huffpostmaghreb.com/2016/07/12/questions-maroc-parlement_ n_10943522.html, accessed May 10, 2018).

57 essentially discretionary, and thus the institutions of horizontal accountability that exist are weaker than they appear on paper. In other words, the legislature in this context has legal authorities of oversight toward executive agencies but lack the additional ‘de facto’ sanctioning power required for polyarchy (O’Donnell, 1998). Though formal channels of influence may exist between elected bodies and appointed institutions, if they lack provisions for enforcement (or if those provisions are subject to other regime-controlled institutions, such as the judiciary), they are only effective if the recipient body chooses to cooperate. What strategic considerations might ministers weigh in determining which queries to prioritize? Queries cost resources to answer, though in general, the costs should be fairly minimal. The majority of queries submitted address district or sub-district issues.19 Questions generally address issues of policy implementation rather than policy and are appropriately targeted at the relevant ministries: for example, the Ministry of Transportation fielded a query concerning “The condition of buses departing from Chichaoua” while the Ministry of Health received a question about “The status of public health in Bouchabel Municipality.”20 And while oral query sessions occasionally include contentious exchanges regarding sensitive issues, these are less common in written queries.21 Answering questions may also have broader implications for government performance at both the deputy and institutional level. Written queries constitute a key form of deputy constituency service, especially for members of opposition parties that lack access to patronage networks (York, 2019). Responding to queries may empower the asking deputy (likely from an opposition party) to do a better job serving his or her constituents; refusing to respond may undermine their effort to perform well. Likewise, a higher rate of response may have the general effect of enhancing the credibility of governing institutions in general: the legislature’s reputation may benefit from its improved performance on constituency issues, and

19Around 70% of queries reference an administrative unit at the legislative district or subdistrict level (including province, prefecture, municipality, city, or village). 20The speciﬁcity of requests to a particular bureaucratic proﬁle may also limit opportunities for forum shopping in targeting queries. 21In the Moroccan context, highly sensitive issues are rarely discussed. For example, protests and demon- strations were referenced in less than 0.1% of written queries submitted, as compared to 1% of oral queries.

58 the ministries themselves may be seen as more credible for behaving as legally mandated. Yet there is substantial variation in responsiveness at the ministry level – the Ministry of Health answered 88% of the queries it received, while the Ministry of Culture responded to a mere 25%. What might determine ministerial incentives and how they evaluate these strategic considerations in determining whether to answer a given query? I predict that this is conditional on the affiliations of ministry leadership and the parties to whom they are accountable. Though in Morocco, all ministers are accountable in some way to the palace (Benchemsi, 2012), the extent to which this is the case varies based on whether they are partisan or technocrat appointees. Technocrat ministers have the backing only of the king: because they have no partisan affiliation, there are no non-regime political actors to whom they answer. Their appointment resulted from regime support and requires it to continue. Were the king to exercise his discretionary ability to dismiss them, it is unlikely that any partisan actors would object. On the other hand, partisan appointees were put forward by their party leadership as a condition of joining the governing coalition. While in office, they remain affiliated with that same party. And that party would likely object to their dismissal; having consented to join the government on the condition of ministry control, they would request a partisan replacement for the demoted minister.22 Other scholars have argued that techocratic governments should differ substantially from partisan governments in behavior due to the lack of interest in furthering party policy objectives (McDonnell and Valbruzzi, 2014). Yet under autocracy, such individuals serve at the regime’s discrtion. Thus not only are they less incentivized to promote party interests, we should expect them to actively promote regime interests within the cabinet.23 The appoint-

22This has, in fact, occurred in recent terms. In October 2017, in response to ongoing protests in northern Morocco, the king dismissed four cabinet ministers. Two were technocrats, while the other two were affiliated with the Party of Progress and Socialism (PPS). The replacement nominees for the two PPS ministers were both from that party. 23Anecdotally, there is ample evidence that this affects their approach to policy in Morocco – for example, the technocrat Minister of National Education sided with the monarchy in opposing the PJD’s efforts to continue arabization of the national education curriculum, much to the chagrin of PM Benkirane. See Omar Brouksy, “Le Maroc enterre trente ans d’arabisation pour retourner au

59 ment of technocrats is common under autocracy, with China and Russia serving as prominent examples of regimes that frequently select bureaucrats without political experience (Cheng and White, 1990; Huskey, 2010). How should we expect the regime to view systems of executive oversight such as written queries? From the regime’s perspective, the strategic costs to responding to questions may exceed any general benefits in government performance. The legislature is the primary source of political contestation, and these requests are an important means by which deputies perform constituency service.24 Refusing to answer requests or delaying their response may therefore diminish legislative effectiveness and potentially reduce support for this elected institution among citizens. This in turn helps the regime neutralize the potential political threat posed by hosting competitive elections. I thus predict that alignment with the regime generates an incentive not to respond. By contrast, partisan ministers are accountable to the other members of their party and especially to party leadership.25 In many cases, partisan ministry nominees are drawn from the pool of party leaders and thus should have a vested interest in promoting their party within the government and to voters. By addressing deputy requests and thereby enabling legislative effectiveness, they strengthen legislative legitimacy – much needed in a context where voters are often disenchanted about the significance of elected institutions and where spoiled ballots are a frequent occurrence (Arab Barometer, 2019; National Democratic Institute, 2011). This creates an incentive for partisan ministers to respond to requests in a timely manner. In addition to this general partisan incentive, we might also anticipate a co-partisan effect: if partisan ministers have an interest in promoting their party over others, français,” Le Monde, February 19, 2016 (https://www.lemonde.fr/afrique/article/2016/02/19/ maroc-le-roi-mohamed-vi-enterre-trente-ans-d-arabisation-pour-retourner-au-francais_ 4868524_3212.html, accessed July 10, 2019). 24Deputy interview, January 30, 2018. 25This assumption draws on party-focused work in other contexts, e.g. Katz (1987). Though the partisan environment in Morocco differs from that of the most developed democracies, many of the criteria developed for party government by Katz apply to the specific case of partisan ministers. In particular, the officials of interest (ministry leaders) are recruited and nominated by party leadership. Further, due to the use of closed list PR in national elections in Morocco, party leaders are also responsible for nominating candidates for office, which means they oversee both elements of the inter-governmental interaction.

60 we might expect them to treat requests from affiliated deputies more favorably. These expectations generate the following hypotheses about responsiveness to deputy queries as a function of minister affiliation, holding other attributes – specifically, the costs of answering an individual query – constant:

Hypothesis 1. Technocrat ministers without a partisan aﬃliation will be less likely to respond, and respond more slowly, to deputy requests than partisan ministers.

Hypothesis 2. Partisan ministers will be more likely to respond, and respond more quickly, to requests submitted by co-partisan deputies.

2.4 Data and Approach

The preceding hypotheses predict differing levels of responsiveness based on minister identity – partisan versus technocrat, and co-partisan versus not – in answering requests submitted by legislative deputies. To test these predictions, I use a dataset of all 27,196 unique written queries submitted by deputies during the 2011-2016 parliamentary session, gathered from the official parliamentary website in March 2017. These data are the result of self-reporting from an autocratic government, as there is no independent tracking of deputy queries. Yet there is reason to expect that this represents a complete record of written queries submitted during the given legislative term. Data match summary statistics reported separately by the government. Additionally, in interviews, deputies described these queries as a documentation of their own work, so they have a strong incentive to ensure accurate record-keeping.26 The government database reports the subject of each query, the asker(s), ministry being queried, date submitted, and date of response, if one was received.27 I supplement this information with separately gathered data on minister affiliation, deputy characteristics, and content analysis of the queries themselves. Ministry affiliations were obtained through media research and verified with the assistance of a local NGO fo-

26Deputy Interview, February 12, 2018. 27Summary statistics at the query level on these and other attributes are reported in Appendix Table B.2

61 cused on citizen-parliamentary relations. Deputy characteristics were obtained from the oﬃcial parliamentary website. Finally, I used dictionary-based text analysis techniques to code content for each query, including geographic and administrative references as well as whether or not a query clearly addressed an issue concerning an individual.

2.4.1 2011 elections and cabinet control

The 2011 elections, held following the Arab Spring protests that rocked the region, resulted in a limited victory for the opposition: the Islamist Party of Justice and Development (PJD) won a plurality of votes and its charismatic leader, Abdelilah Benkirane, was appointed prime minister. Yet the fact that the PJD had won just over a quarter of the total seats in parliament meant that it needed to partner with three other parties in order to acquire the majority necessary to govern.28 The coalition that formed was able to secure control of three of four ministries of sovereignty as well as the important Ministry of Na- tional Education.29 Especially notable was the fact that the list of newly partisan ministries included the Ministry of the Interior – a notoriously sensitive position and one that had previously been held only by non-partisan individuals (Sater, 2016). Though it was led by an appointee from the royalist-leaning Popular Movement party, the fact that it was under partisan rather than technocrat control was unprecedented, and perhaps represented a calculated concession by the palace in light of the unstable political environment in the region, as protests and civil war engulfed other autocratic regimes. But this government was short-lived: in mid-2013, only a year and a half into the governing term, the opposition-oriented Istiqlal party announced it would withdraw from the coalition over tensions between its leadership and that of the PJD (Fakir, 2018). The departure of Istiqlal meant the PJD had to initiate new cabinet negotiations, replacing it with a second royalist-linked party (the National Rally of Independents, or RNI) after months

28The resulting coalition was comprised of the PJD, two secular opposition parties – Istiqlal (PI) and Party of Progress and Socialism (PPS), and the royalist-oriented Popular Movement (MP). 29The exception was the Ministry of Islamic Aﬀairs, which remained under technocrat control.

62 of negotiations. Importantly, in the ensuing reshuffle of cabinet ministers, regime-endorsed technocrats replaced partisans as heads of the ministries of education and the interior – evidence of considerably increased palace influence.30 In the empirical analyses that follow, I examine the implications of differing ministerial affiliations – as well as these significant shifts from partisan to technocrat leadership within important cabinet profiles – for responsiveness to written queries submitted during this term. 31

2.4.2 Independent and dependent variables

I consider two metrics of ministry responsiveness to queries. The ﬁrst is a binary indicator for whether or not a question received a recorded response.32 The second is the time- to-response (TTR) measured as the number of days between the initial question and its response, if one was issued.33 Queries receive an answer only once, so each appears as a single observation in the dataset. Queries are always assigned to the minister to which they were initially posed. But the exodus of Istiqlal from the governing coalition in 2013 and subsequent cabinet reshuﬄe introduces complications with respect to outcome measures. In particular, in cases where the identity of the minister changed hands mid-term, a query that was submitted to one minister and not answered by the changeover may have ended up being addressed – or not – by his

30A full list of ministry positions and partisan or technocrat aﬃliation during the 2011-2016 parliamentary term is in Appendix Table B.1. 31Nearly all non-partisan cabinet appointees from the 2011-2016 term meet the technocrat deﬁnition (I discuss the two exceptions in Appendix B.8). For example, Rachid Belmokhtar, the Minister of National Education from October 10, 2013 through the remainder of the 2011 term, a trained engineer and former university president, has never held a public party membership. He was appointed to the same post on one prior occasion from 1995-1998. Other technocrats appointed during this term include career civil servants, engineers, and historians. 32Note that this characterizes formal responsiveness. It remains possible that ministries in some cases might informally respond to deputies through personal channels. I discuss the likelihood that this occurs and its implications in sections 2.5.2 and 2.6. 33Of the 27,196 queries submitted, exactly one was missing a ministry designation and eight were missing information on the deputies that submitted them. I exclude the former from all subsequent analyses and the latter from all analyses including deputy characteristics. Additionally, 68 queries – 0.3% of the total dataset – had a reply date that appeared to have been entered incorrectly such that it preceded the date of submission; I drop these queries from all analyses using TTR outcomes.

63 successor.34 For questions asked on or after October 10, 2013, when the new cabinet took office, metrics are straightforward: queries are coded as 1 if they have a response recorded before the subsequent parliamentary elections and 0 otherwise. Likewise, TTR measures the lag in days between query submission and recorded response; I treat queries that were never answered as right-censored with a final observation date of October 7, 2016, the date of the 2016 national elections. For queries asked during the first cabinet period, I adopt a cutoff of the date on which the new cabinet officially took office, such that the response variable refers only to whether the original minister answered the given query, and how quickly he did so.35 I therefore treat questions asked before and answered after the changeover as right-censored observations in the subsequent survival analysis. The two hypotheses enumerated in the preceding section require two different independent predictors. To test H1, I use Technocrat, a binary indicator for whether or not the ministry queried was headed by a technocrat.36 For H2, I use Aligned, a binary indicator for whether or not the minister shared a partisan affiliation with the deputy or deputies raising the query.37 The treatment in this case is conceived of as assignment of a query to a ministry with or without the predictor attributes. The control condition for Technocrat is assignment to any partisan minister, while the control condition for Aligned is assignment to a partisan but non-aligned minister.38

34In the new cabinet, eleven ministries and delegate ministries changed hands (either to a different party or to technocrat leadership) and a further five delegate ministries were created and distributed among partisan members of the new coalition. See Appendix Table B.1. 35I adopt this approach in order to ensure as much comparability as possible between ministries that changed hands, and those that did not. As a robustness check for this coding decision, I construct a response variable using the October 7, 2016 date as the cutoff for all queries and re-run the primary specifications on a subset that excludes queries submitted 144 days or fewer before the new cabinet took effect – the median response time among queries that received a response. Results are consistent with the main results. 36Nearly all non-partisan appointees historically and during the 2011 term meet technocrat criteria (Mc- Donnell and Valbruzzi, 2014). There were two non-partisan appointees during the 2011 term that did not; I exclude them from the technocrat analyses and explore the impact of a more general non-partisan treatment in Appendix B.8. 37More than 90% of queries were raised by a single deputy (see Table B.2). In all cases with multiple cosigners onto a single question, the deputies were members of the same party. 38Note that the comparison case for the Technocrat treatment thus includes queries submitted to both aligned and non-aligned partisan ministers.

64 2.4.3 Additional covariates

In addition to the main predictors, I consider an array of other covariates that may impact response patterns, including features of the query itself; the individual(s) that asked it; and the ministry to which it was directed. The timing of a query may impact the ability and incentives for ministers to respond. Overall responsiveness was highest in the first year of the term and declined as the term progressed. Additionally, queries raised near the end of a cabinet period were less likely to be dealt with before the cutoff. I therefore code a factor variable Year identifying the year a query was submitted, as well as a binary variable Late indicating whether or not a query was submitted within 144 days (the median response time among queries that were answered within their assigned cabinet period) of the cutoff for its cabinet period. Finally, I create First Cabinet, a dummy indicator for the cabinet period during which each query was submitted. Additionally, because queries address a range of different issues and some may be easier to respond to than others, I create variables describing the content of a given query. Commune is a binary indicator for whether a query includes a reference to a commune (the smallest administrative division in Morocco). Georeference is an indicator for whether a query references any geographic division in Morocco, including communes as well as provinces, prefectures, and regions. Casework identifies queries that address issues on behalf of an individual, an important component of deputy activity in this context (Benstead, 2008).39 Ministry responsiveness may also vary based on the characteristics of the individual(s) that submitted a given query. Queries that are cosigned by multiple deputies may produce more pressure on ministers to respond, but this may also be an indicator that a question addresses a broader, more complex set of issues that requires more time to answer. Solo thus indicates whether a query was signed by a single deputy. Likewise, senior party members may

39Casework focused queries are coded based on references to an individual constituent issue. The following query is an example of casework: “Request to recover the taxi license in Tangier for Ms. X in Rabat.”

65 have more connections and hold more sway within government ministries. Leader indicates whether a query was signed by at least one committee chair or vice chair, positions typically reserved for party leaders and senior members. I also code Former, an indicator for whether or not at least one of the submitting deputies was a member of the 2007 parliament, and National list, which identifies whether at least one of the submitting deputies was elected via the national list, a set of seats reserved for female and youth deputies that lack direct geographic constituencies.40 Finally, I consider the party of the deputy or deputies involved: different parties have differing policy objectives, and this may influence the focus and type of query they submit. I construct a set of dummy variables (Group) for the eight largest parliamentary groups.41 Additionally, Royalist indicates whether the party of the asker(s) has an affiliation with the monarchy. Ministry-specific attributes are largely captured by the main IVs as well as ministry fixed effects in most specifications. However, I also code New Minister, an indicator for whether or not the ministry leadership changed or was initiated during the cabinet reshuffle for queries submitted during the second cabinet period.

2.4.4 Approach

The data used in this study are observational, which poses potential concerns for inference. In particular, we must consider possible issues of selection on two fronts: assignment of individuals to ministry leadership, and question assignment to ministries. With respect to the former consideration, there is evidence that ministers of diﬀerent types are not randomly assigned to ministries. As discussed in the preceding sections on Moroccan cabinet negotiations, some ministries are seen as being of higher consequence than others (Madani,

40In addition to lacking a direct electoral constituency, national list deputies are generally less experienced politicians serving their ﬁrst term in parliament. 41A total of 18 parties won seats in the 2011 parliament, though only eight won more than four seats (1% of the total house of representatives). For caucusing purposes, these smaller parties formed coalitions with the larger parties or amongst themselves. Here, I include indicators for the parliamentary groups representing the eight parties that won at least 18 seats. The base category is membership in one of these smaller additional parties.

66 Maghraoui and Zerhouni, 2012), and this may be related to the difficulty of answering the typical query submitted to such a ministry. If there is a correlation between complexity of topic and ministry importance, and the regime is more likely to maintain control of important ministries, we may incorrectly attribute the impact of complexity – making it harder to answer the given questions – to technocrat ministers.42 There are additional considerations with respect to the queries themselves and their non- random assignment to ministries headed by different leadership. It is possible that deputies raise different types of questions with different types of minister. Deputies may self-censor if they expect that a query will not be answered, asking only questions that are easy to answer, which would attenuate effects. Alternately, deputies may be more or less confrontational with different types of minister, asking difficult questions of regime-affiliated technocrats and submitting softball questions to ministers from their own party. Because written questions are not broadcast by default to other politicians or the public, they are more suitable for genuine constituency service than as signaling devices of deputies’ commitment to a party platform or stance relative to the regime. This is in contrast to a second legislative authority: the ability to directly question ministers during televised oral query sessions, which has been found in other contexts to involve public criticism of the regime (Malesky and Schuler, 2010).43 With another outlet for direct criticism of the regime, deputies are thus less likely to couch their criticism of regime policy in written queries, and these are instead used to address district-level problems in order to satisfy constituent requests.44 Nevertheless, we cannot dismiss the possibility that attributes that affect the ease and speed with which a question can be answered are correlated with minister characteristics. To address these selection issues, I adopt two approaches: matching on observable characteristics and difference-in-differences estimation using the midterm change in ministry lead-

42I do, however, undertake observational analysis of responsiveness as a function of minister aﬃliation in Appendix B.3. As with the results to follow, the observational data support Hypothesis 1 but run counter to Hypothesis 2. 43See York (2019) for more discussion of the two types of query and the diﬀering ways deputies use them. 44Deputy interviews, February 12 and 13, 2018.

67 ership.45 For the former, I use propensity score matching to pair treated observations with control observations based on shared background covariates; to identify pairs, covariates are weighted according to their predictive capacity for treatment condition (Ho et al., 2007).46 I create matched datasets for each treatment category using nearest neighbor matching on eighteen background covariates.47 The goal of this approach is to generate a counterfactual group for treated observations that is as similar as possible with respect to other observable attributes, such that the only visible difference between queries in one group or another is whether or not they were submitted to a minister with the relevant treatment characteristic. To create this matched dataset, I first regress the treatment variables (Technocrat and Aligned, respectively) on the matching covariates and use the resulting predicted values to generate a propensity score for treatment for each observation.48 I then use 1:1 nearest neighbor matching without replacement to identify matches for each of the treated queries (Ho et al., 2011). The resulting dataset allows estimation of the average treatment effect among the treated. The matched datasets are well-balanced across treatment and control on these observable characteristics, in contrast to the full, original dataset (Table 2.1). In particular, the absolute standard bias is under 0.25 for all covariates (and very close to 0 for most), considered a ‘good match’ (Ho et al., 2007), and the average across the set of matching covariates is 0.005 for techno and 0.04 for aligned data. Throughout, I use ordinary least squares regression in specifications using a binary re-

45I discuss the latter approach in more detail in section 2.5.1. 46As a robustness check, I also construct datasets using exact matching; this approach drops fewer observations overall, but instead sorts observations into subclasses based on existing combinations of covariate values and weights them according to the likelihood of treatment within categories. Results from this analysis are in Appendix B.4 and are consistent with the ﬁndings from nearest neighbor matching. 47The covariates used in nearest neighbor matching are: Year, Late, First Cabinet, Georeference, Casework, Solo, Former, Leader, National List, and Group - an unordered factor variable with nine levels. 48Note that in the Technocrat case, the comparison is between technocrat and partisan ministers; I exclude queries submitted to non-partisan appointees that do not meet the technocrat criteria (I include these ministers in a separate exploration of Non-partisan treatment in Appendix B.8). In the Aligned case, the comparison is between aligned partisan and non-aligned partisan ministers; I exclude queries submitted to technocrat ministers because they are by deﬁnition ineligible for this treatment condition.

68 Technocrat Matched Data Treated Mean Control Mean Diff-in-means Std. Bias Prop. Score 0.29 0.29 0.00 0.00 2012 0.02 0.02 0.00 0.00 2013 0.10 0.10 0.00 0.02 2014 0.53 0.53 -0.00 -0.00 2015 0.24 0.24 -0.00 -0.00 2016 0.11 0.11 -0.00 -0.00 Late 0.07 0.07 -0.00 -0.01 First Cabinet 0.04 0.04 0.00 0.00 Georeference 0.79 0.79 0.00 0.01 Casework 0.04 0.04 0.00 0.01 Solo 0.91 0.91 0.00 0.00 Leader 0.20 0.20 0.00 0.00 Former 0.26 0.26 0.00 0.01 National List 0.12 0.13 -0.01 -0.02 Group: MP 0.03 0.03 -0.00 -0.00 Group: PAM 0.01 0.01 -0.00 -0.00 Group: PI 0.06 0.05 0.00 0.00 Group: PJD 0.43 0.44 -0.00 -0.00 Group: PPS 0.01 0.01 -0.00 -0.00 Group: RNI 0.01 0.00 0.00 0.01 Group: UC 0.00 0.00 0.00 0.00 Group: USFP 0.44 0.44 0.00 0.01 Aligned Matched Data Treated Mean Control Mean Diff-in-means Std. Bias Prop. Score 0.42 0.41 0.01 0.09 2012 0.18 0.22 -0.03 -0.09 2013 0.26 0.28 -0.02 -0.03 2014 0.30 0.28 0.02 0.05 2015 0.16 0.13 0.03 0.07 2016 0.10 0.10 0.00 0.00 Late 0.16 0.17 -0.01 -0.03 First Cabinet 0.36 0.42 -0.06 -0.12 Georeference 0.59 0.56 0.02 0.05 Casework 0.08 0.05 0.03 0.10 Solo 0.84 0.84 -0.00 -0.01 Leader 0.07 0.08 -0.01 -0.02 Former 0.27 0.25 0.02 0.04 National List 0.19 0.19 -0.01 -0.02 Group: MP 0.02 0.02 0.00 0.01 Group: PI 0.04 0.04 -0.00 -0.01 Group: PJD 0.92 0.92 0.00 0.01 Group: PPS 0.01 0.01 -0.00 -0.02 Group: RNI 0.00 0.00 0.00 0.00 Table 2.1: Summary Statistics – Nearest Neighbor Matched Data. Table presents summary statistics across a range of covariates for treatment and control groups generated using 1:1 nearest neighbor matching on treated observations. Top panel describes data generated based on technocrat treatment (n = 11858, with exactly 5929 observations in each category), while bottom panel is generated based on alignment treatment (n = 7224, with exactly 3612 observations in each category). Difference-in-means as well as standard bias estimates are included to demonstrate comparability between categories.

69 sponse predictor.49 In survival analyses using TTR outcomes, I use the semi-parametric Cox proportional hazards model, which requires fewer parametric assumptions than other models (Box-Steffensmeier and Jones, 2004). In all cases, I report clustered standard errors to account for the fact that the treatment variables are assigned by ministry (for technocrat) and a combination of ministry and party (for partisan alignment). My preferred specification includes ministry fixed effects and adjustment for a set of covariates included in matching.50 The ministry fixed effects in particular are intended to address concerns that question complexity and sensitivity may vary at the ministry level. I also report effects from regression on the treatment indicator alone as well as a specification including treatment and ministry fixed effects.

2.5 Results

I first consider effects on the straightforward outcome of whether or not a query was answered before the given cabinet was dissolved. Table 2.2 reports treatment coefficients from OLS regression with and without controls, using nearest-neighbor matched datasets.51 Cluster robust standard errors (HC1) are constructed at the ministry-cabinet level for all regressions on the Technocrat variable and at the group-ministry level for regressions on the Aligned variable. Results support Hypothesis 1 and suggest that technocrat ministers responded to queries at significantly lower rates than partisan ministers: coefficients from both nearest neighbor and exact matching are negative in all specifications and significant at the 0.001 level when fixed effects and controls are added. The difference in responsiveness is substantively large: queries submitted to technocrat ministers were approximately 23 percentage points less likely to receive a response. While about 70% of queries submitted to partisan ministers were

49I prefer OLS to nonlinear regression models due to its better accommodation of fixed effects specifications (see (Lancaster, 2000)). Likewise, Angrist and Pischke (2008) demonstrate that the Conditional Expectation Function from OLS approximates that of logit and probit models across a wide range of specifications. 50The vector of control attributes includes Year, First Cabinet, Late, Georeference, Casework, National List, Leader, and Solo. 51See Table B.4 in the appendix for similar analysis using exact matched datasets.

70 Dependent variable: Received Response (1) (2) (3) (4) (5) (6) Technocrat Minister −0.233 −0.227∗∗∗ −0.229∗∗∗ (0.144) (0.016) (0.038)

Aligned Minister −0.069 −0.016 −0.015 (0.071) (0.026) (0.029)

Mean DV, non-techno: 0.700 0.700 0.700 Mean DV, non-aligned: 0.705 0.705 0.705 Ministry FE: XXXX Controls: XX Matched Sample: Techno Techno Techno Aligned Aligned Aligned Observations 11,858 11,858 11,858 7,224 7,224 7,224 Adjusted R2 0.055 0.233 0.320 0.005 0.130 0.343 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Table 2.2: Responsiveness – Matched Data. Table presents results from OLS estimation of whether a query received a response on treatment indicators in matched datasets, with and without ministry ﬁxed eﬀects and other controls, for both technocrat (models 1-3) and partisan alignment (models 4-6) treatments. Datasets were constructed using nearest neighbor propensity score matching. Robust standard errors are clustered at the ministry- cabinet level for the technocrat treatment, and the group-ministry level for the alignment treatment.

71 eventually answered, technocrats answered under 50% of the queries they received. On the other hand, there is no support for the expectation from Hypothesis 2 that deputies are more likely to receive a response when a query is submitted to a co-partisan minister. Coefficients on alignment are small, negative (suggesting partisan ministers were slightly less likely to answer members of their own party), and statistically insignificant in all specifications. I discuss possible explanations for this unexpected result following the TTR analysis below. I next consider effects of minister characteristics on the time taken to respond to a query. I first examine non-parametric Kaplan-Meier plots of survival as a function of treatment condition, using the nearest-neighbor matched datasets (Figure 2.1).52 Consistent with Hy- pothesis 1, the plots reveal that technocrat ministries respond more slowly to deputy queries: it took partisan-led ministries 118 days to respond to 25% of queries, but technocrats took three times as long at 366 days. Counter to Hypothesis 2, queries were answered more quickly when deputies were not co-partisans, but the difference was not as substantial: 25% of co-partisan queries were answered in the first 120 days, compared to 93 days for queries submitted to unaligned ministers. Results from estimating Cox proportional hazards model on the TTR outcome, shown in Table 2.3, bear out these non-parametric observations. For the technocrat treatment (models 1-3), coefficients are negative and significant in all specifications, indicating that the hazard (likelihood of receiving a response) is reduced for queries submitted to technocrat-led ministries. Coefficients can be interpreted in exponentiated terms: in the model including fixed effects and controls (model 3), the coefficient of -1.35 implies that the probability of response is about 74% lower relative to the baseline hazard function for queries submitted to technocrats. This supports the assertion in Hypothesis 1 that technocrat ministers respond more slowly to deputy requests.

52While I use the semi-parametric Cox proportional hazards model including covariate adjustment to estimate the statistical diﬀerence in survival across treatment conditions, I include the Kaplan-Meier plots in order to provide a descriptive and easily interpretable visualization of response patterns across treatment conditions.

72 a) Technocrat b) Aligned

1.00

0.75

0.50 Proportion answered 0.25

0.00

0 250 500 750 1000 0 250 500 750 1000 Days after submission

Treated No Yes

Figure 2.1: Kaplan-Meier Curves – Matched Data. Figure depicts Kaplan-Meier survival estimates for technocrat treatment (left panel) and partisan alignment treatment (right panel), using nearest neighbor matched datasets constructed for each treatment condition.

On the other hand, the results again fail to support the expectation from Hypothesis 2 that partisan ministers will respond more quickly to requests from co-partisans. Coefficients on the Aligned indicator are negative in all specifications, indicating a lower hazard rate and slower response time, though they are also insignificant and close to 0 when controls and fixed effects are included. What might explain this finding, or lack thereof? In the theory section, I argued that the incentives of ministers to respond to requests are related to the entities to which they themselves are accountable – and posited that partisan ministers should be most accountable to their own party leadership, which should spur them to privilege requests from co-partisan deputies. But it may be the case that inter-partisan cooperation is more important than expected in this setting. No party won an outright majority during the 2011 elections, meaning that legislative activities requiring majority support necessitate coalition-building. This may create an incentive for partisan ministers to treat deputy requests equally, in order

73 Dependent variable: Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat Minister −0.799+ −2.333∗∗∗ −1.351∗∗ (0.410) (0.407) (0.452)

Aligned Minister −0.301 −0.037 −0.021 (0.187) (0.233) (0.108)

Ministry FE: XXXX Controls: XX Matched Sample: Techno Techno Techno Aligned Aligned Aligned Observations 11,839 11,839 11,839 7,211 7,211 7,211 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Table 2.3: Cox Proportional Hazards Model – Matched Data. Table presents results from estimating a Cox proportional hazards model of time-to-response on treatment indicators in matched datasets, with and without ministry ﬁxed eﬀects and other controls, for both technocrat (models 1-3) and partisan alignment (models 4-6) treatments. Models are estimated on datasets constructed using nearest neighbor matching. Robust standard errors are clustered at the ministry-cabinet level for the technocrat treatment, and the group-ministry level for the alignment treatment.

74 to avoid accusations of favoritism that may undermine future deal-making. Alternately, it may simply be the case that the ﬁrst priority of partisan deputies is to enhance legislative performance – and thereby the importance of the legislature in the eyes of citizens – and that the most eﬀective way for them to do this is to answer deputy requests to the extent possible, regardless of their origin.

2.5.1 Diﬀerence-in-diﬀerences

The matching approach implemented in the previous section is useful in that it removes control observations that lack a valid counterfactual among the treated cases, but it cannot account for unobserved confounders. To bolster the findings from the matching analysis, I also exploit the cabinet reorganization that occurred in mid-2013 to conduct a difference- in-differences (DiD) analysis of the effect of having a technocrat minister on responsiveness. The treatment in this case is defined as a switch in ministers from partisan to technocrat.53 In this section, I restrict attention to the technocrat treatment due to the lack of finding with respect to partisan alignment. This approach has the advantage of holding ministry constant and allowing us to differentiate between response rates for questions that would have been assigned to a non-technocrat minister had the cabinet change not taken place. We can thus compare outcomes for queries submitted to the same ministry under differing leadership conditions. Here, I subset the dataset to include only queries submitted to ministries that were partisan during the first cabinet.54 Under the assumption that a given question would always be assigned to a particular ministry (generally plausible, given that query topics constrain

53Because no ministries changed from technocrat to partisan hands in 2013, I am unable to test for effects with the reverse treatment (that is, whether we see an equal and opposite effect in query response for ministries that became partisan). 54This excludes 2290 queries (8% of the total dataset) submitted to the five ministries and delegate- ministries that were held by technocrat or non-partisan ministers throughout the parliamentary term (Ap- pendix Table B.1). Note that no ministries switched from technocrat to partisan in the 2013 cabinet shift, meaning it is not possible to test a difference-in-differences model in which the treatment is a change from technocrat to partisan leadership.

75 them to a particular branch of the bureaucracy),55 I test whether a shift in minister identity corresponds with a change in responsiveness to subsequent queries. In this approach, the questions asked before the change in ministers act as a reference category for the subsequent queries, and help to account for other non-treatment related aspects of ministry responsiveness including any time-based variation (Abadie, 2005). For this analysis, I deﬁne the following variables: post is a dummy indicator for whether or not a query was raised after October 10, 2013 – the date on which the cabinet reshuﬄe took place. Treat is a dummy indicator for whether or not the question was assigned to a ministry that became technocrat in 2013.56 I then estimate the following equation:

Yijt = τ(treatj × postt) + µj + Xi + ijt, (2.1)

where i denotes individual queries, µj denotes ministry fixed effects, and Xi indicates a battery of additional control variables.57 I estimate equation 2.1 with and without ministry fixed effects and additional controls. As before, the dependent variable is an indicator for whether or not a question received a response by the dissolution of the cabinet to which it was submitted. In all specifications, I construct standard errors clustered at the ministry level using bootstrapping (Bertrand, Duflo and Mullainathan, 2004). I also introduce sample restrictions to isolate the treatment and reduce the potential for confounding. In particular, I remove queries submitted to ministries that were led by technocrats throughout the entirety of the 2011-2016 term (i.e. were treated in both the first

55For example, I assume that a concern about a public school (such as ‘The deteriorating conditions of Salman Farsi secondary school in Salé’) would always be addressed to the Ministry of National Education. This would not hold if multiple ministries are able to address a given issue and deputies are able to selectively assign queries. Though it is not possible to conclusively rule out this possibility for all issues, content analysis and submission patterns suggest that deputies target queries to ministries based on topic rather than on partisan affiliation. 56Two ministries – the Ministry of National Education and the Ministry of the Interior – switched from partisan to technocrat control. These two ministries fielded 6843 queries throughout the 2011 term (around 25% of the total dataset). No ministries switched from non-partisan to partisan control in the reshuffle (see Table B.1), making it impossible to estimate the ‘reverse’ treatment. 57The controls used here include year fixed effects, Late, First Cabinet, Georeference, Casework, Solo, Leader, and National List

76 and second periods), such that only ministries that were non-technocrat for the duration of the 2011 cabinet are included. This subset generates a comparison in response rates between ministries that became technocrat and those that remained partisan in both periods.58 The DiD design relies on the assumption of parallel trends (Abadie, 2005), which requires that prior to the shift in cabinet leadership, outcomes among the two groups followed similar trajectories. I test this by plotting monthly averages in response rates among treated and untreated queries leading up to the cabinet reshuffle (Appendix Figure B.1). The plots support the parallel trends assumption: trendlines are very similar among the two groups and, in fact, mostly overlapping leading up to the change in ministers. Given the data generating process involved, I also look at patterns in query submission across treated and untreated ministries, to test for a change in question submission patterns following the appointment of technocrat ministers (Appendix Figure B.2). Though the proportion of queries submitted to treated ministries varies across individual months, the trendline suggests that the relative distribution was quite consistent over time, and there is no evidence that the switch to technocrat ministers produced a short or long-term change in the relative rate of query submissions. Models 1-3 in Table 2.4 present estimates of τ, the average treatment effect among the treated, derived from estimating equation 2.1 on the binary response outcome in specifications with and without controls. Standard errors are clustered at the ministry level. Results are consistent with those using the matched dataset (Table 2.2). In all specifications, queries submitted to ministries that had fallen into technocrat hands were less likely to be answered than they would have had the ministry remained under partisan control; these results are significant to the α = 0.001 level. And the impact was substantively large: coefficient estimates suggest that the probability of response declined by around 30 percentage points.

58As a robustness check to address a possible exclusion restriction violation associated with the fact the the treated category experienced a change in leadership in addition to the change in minister type, I create a further restricted sample that includes only queries submitted to ministries that changed hands in the 2013 reshuﬄe. Results from this subset are shown in Appendix B.6 and bear out those in the main text in both magnitude, sign, and signiﬁcance.

77 Dependent variable: Received Response Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat x Post −0.260∗∗∗ −0.327∗∗∗ −0.330∗∗∗ −1.116∗∗∗ −1.678∗∗∗ −1.671∗∗∗ (0.065) (0.061) (0.048) (0.324) (0.214) (0.237)

Ministry FE: XXXX Controls: XX Model: OLS OLS OLS Cox PH Cox PH Cox PH Observations 24,625 24,625 24,625 24,625 24,625 24,625 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Table 2.4: Difference-in-differences Specification – Technocrat Ministries. Table presents estimated treatment effects from estimation of equation 2.1, with and without ministry fixed effects and additional controls. Models 1-3 represent OLS estimation using a binary indicator for response received; models 4-6 estimate a Cox proportional hazards model with a time-to-response outcome. Robust standard errors are clustered at the ministry level and constructed through bootstrapping.

In models 4-6, I present results from estimating a Cox proportional hazards model. Again, I find evidence that technocrat ministers respond more slowly than their partisan counterparts: as with the estimates from the matched observations, τ is negative and significant in all specifications, demonstrating that the hazard rate is reduced for queries submitted to ministries after they were taken over by technocrats. To help illustrate the substantive impact, I offer a visualization of the survival analysis in Figure 2.2. I plot Kaplan-Meier survival estimates for queries by treatment condition and cabinet period in order to compare the early and late-term survival estimates for ministries that changed from partisan to technocrat and those that remained in partisan hands throughout the 2011 parliamentary term. Responsiveness changed substantially following the change in ministers: during the 2012 cabinet (solid lines), Kaplan-Meier curves are similar for treated and untreated ministries, with treated ministries responding slightly more quickly to most queries. Though responsiveness for both categories declined during the 2013 cabinet (dotted lines), the change was much more dramatic among treated ministries. During the 2012 cabinet, 25% of queries

78 1.00

0.75

0.50 Proportion answered

0.25

0.00

0 250 500 750 1000 Days after submission

Cabinet a) Untreated b) Treated 2012 2013

Figure 2.2: Kaplan-Meier Curves – Difference-in-differences (Technocrat Treatment). Plot shows nonparametric survival estimates for the 2012-2013 and 2013-2016 cabinets within treated (those that switched from partisan to technocrat control) and untreated (those that remained partisan through both cabinets) ministries. submitted to untreated ministries were answered within the first 91 days after submission; this declined to 120 days for queries submitted during the 2013 cabinet. By contrast, 25% of queries submitted to treated ministries during the 2012 cabinet were answered within an impressive 66 days. But in the 2013 cabinet, this declined to 389 days – more than a year following query submission for these newly technocrat ministries.

2.5.2 Heterogeneous eﬀects – Royalist versus nonroyalist

Given the royalist/nonroyalist cleavage that divides Moroccan parties and the expected technocrat-regime linkage, we might wonder whether the regime selectively responds to requests based on the orientation of the deputy submitting them. If the regime seeks to support parties that are more closely aligned with it and discredit members of the opposition in the eyes of voters, we would expect technocrat ministers to favor requests submitted by roy-

79 Dependent variable: Received Response Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat Minister −0.227 −0.219∗ −0.216+ −0.778+ −2.313∗∗∗ −1.288∗∗ (0.160) (0.106) (0.122) (0.436) (0.426) (0.469)

Royalist Deputy −0.008 0.028 0.045 0.281 0.293 0.105 (0.047) (0.049) (0.048) (0.183) (0.207) (0.241)

Technocrat x Royalist −0.125 −0.217 −0.213 −0.564 −0.751 −0.829 (0.188) (0.186) (0.181) (0.610) (0.575) (0.563)

Ministry FE: XXXX Controls: XX Matched Sample: Nearest Nearest Nearest Nearest Nearest Nearest Model: OLS OLS OLS Cox PH Cox PH Cox PH Observations 11,858 11,858 11,858 11,839 11,839 11,839 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Table 2.5: Heterogeneous Effects – Royalist Deputies. Table shows output from OLS and Cox proportional hazard model regression of Technocrat treatment interacted with indicators for whether the submitting deputy was from a royalist-affiliated party, using nearest neighbor matched datasets constructed for Technocrat treatment (see Section 2.4.4). Standard errors are clustered at the ministry level and constructed using bootstrapping. alist deputies. If this is the case, we should find heterogeneous treatment effects in which the technocrat treatment has a more negative effect for queries submitted by deputies from nonroyalist parties. Table 2.5 shows output from interaction models using the binary and TTR outcome variables with the nearest neighbor matched datasets. The results provide no evidence that regime-appointed cabinet members systematically favor formal queries submitted by royalist parties.59 On the contrary, queries submitted by royalist deputies are, on average, somewhat less likely to receive a response from ministers of all types and especially from technocrat ministers. The interaction coefficient is, however, insignificant in all specifications.

59This does not, of course, preclude the possibility that these ministers might favor informal, personalist requests submitted by such deputies – and indeed, other research argues that these personal linkages are important in driving electoral support for these parties (Liddell, 2010; Sater, 2012).

80 In light of these results, it is worth revisiting the role that these queries play for members of different types of party as well as their likely relationship with the regime. Though royalist deputies held 40% of seats in the 2011 parliament, they were behind only 6% of the queries submitted throughout the term.60 The comparative lack of usage suggests that these queries do not constitute a particularly popular activity for royalists, and it may therefore be the case that the regime simply does not see this as an avenue by which to empower friendly political actors – just as a means to disempower the opposition, or even parliament as a whole so that the populace do not see it as a viable alternative to continued regime governance. An alternate interpretation is that this interaction effect – though statistically insignificant – is consistent with the expectation that technocrats and loyalist deputies have personal linkages that help the latter succeed politically (Liddell, 2010). Technocrats may sometimes engage with such deputies through informal means, such as a phone call, rather than an official response. Given this possibility, as a robustness check, I test whether the main results on the Technocrat treatment are consistent to a specification including only non-royalist deputies and find that results are largely unchanged (see Appendix Table B.7).

2.6 Discussion

The findings presented in the preceding sections suggest that, consistent with Hypoth- esis 1, technocrat ministers deal very differently with deputy requests than their partisan counterparts. These effects are substantively large: queries submitted to technocrats were more than 20 percentage points less likely to be answered. We must apply some caution in interpreting results: because both difference-in-differences and matching rely on assumptions about ignorability of assignment, the two approaches may not be able to fully address selection issues related to the queries being submitted. I have sought to mitigate threats to identification by adopting two distinct strategies for estimating the effect of minister characteristics, and it is reassuring that results are consistent across both approaches. Nevertheless,

60At the deputy level, the average royalist submitted an order of magnitude fewer queries than his non- royalist counterpart (York, 2019).

81 it is not possible to entirely rule out the concern that unobserved covariates bias the resulting inference. A second concern stems from the fact that the primary outcome variables of interest are derived from whether and when a query received a response, which overlooks potential nuance in the quality of that response. Responsiveness does not necessarily imply that answers are satisfactory in addressing deputy concerns. And yet, it seems reasonable to expect that a poor response is preferable to none at all, since this at least creates a basis for subsequent follow-up. To this extent, then, the estimates from responsiveness may offer a reasonable proxy measure of deputy ability to hold ministries accountable. As noted above, the outcome measures used are derived from official recorded ministry response. Though this reflects the reality of administrative records in this context, it is possible that this measure fails to capture ‘backdoor’ politics in which ministries reach out personally to deputies without ever submitting a formal response.61 This seems most likely to occur in two cases: first, in the previously discussed case of royalist deputies submitting queries to regime appointees, and second, in the case of deputies submitting to co-partisan ministers. To address the first possibility, I conduct a robustness check of the Technocrat treatment on a subset that excludes royalist queries. The second case supposes that ministers are more likely to have direct connections to MPs from within their party and could theoretically respond informally via party networks. If this is the case, it would downwardly bias estimates of the Aligned treatment (and could perhaps explain the negative, insignificant effect), meaning response rates are artificially low for some partisan ministries. But given the direction of effects found, I would expect this to bias the Technocrat findings toward zero, meaning the results may actually underestimate the size of the effect. In other words, this is unlikely to threaten the principal finding that regime-affiliated ministers are less responsive than partisan ministers. These results have important implications for theories of cooptation in autocratic regimes

61Though I cannot exclude this possibility, it is worth mentioning that deputies did not describe receiving unoﬃcial answers in interviews.

82 and our understanding of autocratic control. In particular, the empirical measures of ministry responsiveness examined here allow us to go beyond speculation about how technocrat appointees might undermine democratic institutions: I find that they functionally impede critical institutional authorities for elected representatives in practice. These results also suggest a rationale for outside participation in government, despite the risks involved. If the regime is allowed to appoint all members of the cabinet, legislative influence over the executive (and, correspondingly, bureaucratic responsiveness to citizen needs) would be substantially reduced. Partisan representation helps to buttress accountability in this setting. Additionally, because opposition parties in particular rely on these actions to provide constituency service (York, 2019), reduced responsiveness disproportionately affects their performance in office. These queries represent a systematic measure of legislative requests and minister cooperation, but they are just one form of horizontal accountability in this setting. These results may proxy for general responsiveness by technocrats toward elected deputies in other areas, including the public oral query sessions and committee demands as well as any informal requests. If regime-affiliated ministers are incentivized to obstruct rather than cooperate with MPs in general, then their pervasiveness may serve as an indicator of the level of accountability of the government as a whole. This further provides an additional incentive for opposition parties to participate in and seek influence over the government formation process: though their ability to generate policy may be limited, their impact on policy implementation helps to promote legislative effectiveness. Future work may pay closer attention to the sectors under technocrat control in an effort to understand how the lack of cooperation with legislative oversight might affect other, downstream outcomes in government performance. Finally, the findings here also help to resolve outstanding questions from York (2019): why does the regime allows the existence of a path to increased popular support for the opposition, and how do deputies benefit from an authority that lacks an enforcement mechanism? The results here suggest that the lack of enforcement mechanism is resolved by having

83 partisans in office. Though one explanation for elected institutions under autocracy holds that the regime benefits by delegating responsibility for citizen satisfaction to other actors through the competitive election process (Blaydes, 2010), the regime also has an incentive to undermine the effectiveness of elected deputies in order to avoid increasing public confidence in a competitive institution or in any one political party. It can do this by ensuring that affiliated, technocrat ministers obstruct deputy action to some degree. As a result, by controlling important ministries, the regime can place a ceiling on deputy effectiveness in performing constituency service. Because voters assess parties on the basis of their constituent service, this may have important electoral implications when key ministries are held by regime agents.

2.7 Conclusion

The findings here offer an empirical examination of executive interaction with a competitively elected legislature under autocracy. That autocrats structure the governing system to limit the amount of authority granted to elected institutions is clear, but little work has examined the functional implications of this aspect of autocracy. Here, I show that this discretionary responsiveness may serve as an important lever of power for the regime in the Moroccan case, allowing it to limit action by deputies on behalf of constituents. Because regime-affiliated ministers are not incentivized to enable effective action by elected deputies, queries submitted to such ministers are less likely to receive a response, and are answered more slowly, than those submitted to their partisan counterparts. These findings help to illuminate cooptation in practice, and suggest that outside participation under autocracy is not mere window dressing: it plays an important role in horizontal accountability given the weak institutions involved. They further underscore the importance of close observation of institutional function and interactions under autocracy for future scholarship.

84 Chapter 3: Distribution under Autocracy

3.1 Introduction

Democratic institutions are an increasingly common feature of autocracies. Globally, a majority of authoritarian states today contain legislatures that are elected through constrained or fully multi-party competition, and, as in democracies, law-making is generally delegated in whole or in part to these bodies. At least on paper, autocratic legislatures often look quite similar to their democratic counterparts. But there are important distinctions that differentiate these bodies from legislatures in democratic settings. In particular, the autocrat is often granted outsize authorities relative to the system of checks and balances employed by democratic polities. In many states, the autocrat has the ability to appoint or disband the cabinet, or dissolve the legislature itself and call for new elections. As a result, though laws are amended and passed within the legislature, the autocrat himself might exert influence over the propositions under discussion through these levers of control over the government. Scholars have devoted considerable attention to the question of why dictators choose to constrain themselves by constructing democratic institutions. A prominent theoretical strand within the literature argues that such institutions help to absorb political pressures by providing allies and moderate opposition actors with an opportunity to compete for rents and policy concessions (Gandhi and Przeworski, 2006; Boix and Svolik, 2013). These institutions apparently serve as the autocrat’s credible commitment not to renege on supporters or cooperative opposition (Magaloni, 2008), and as such, we should take their rules seriously. How do these self-imposed constraints affect distributional politics? Under what conditions are dictators able to exert more control over state allocations, and how do specific insti-

85 tutional concessions (such as those enacted following the Arab Spring protests) alter the budgetary distribution in more or less equitable ways? Studies of distributive politics that reference government type have produced some evidence that regional or ethnic favoritism in distribution is heightened under autocracy (Hodler and Raschky, 2014; Burgess et al., 2015). But in general, such studies adopt a broad view of autocracy, grouping governments by high level indicators and classifications and paying relatively little attention to their governing structure. In other words, while there is suggestive evidence that autocracies may produce more asymmetric patterns of distribution, we have little insight into how this is affected by the institutional features they employ. We might expect, in general, that autocrats do better than their democratic counterparts, but why is this the case? In this paper, I explore the mechanisms through which autocrats are able to institutionally advantage themselves. Much of the theoretical literature on autocratic governance is oriented around the dictator’s construction of a ruling coalition under threat of coup.1 I adopt a different approach here, beginning with a common model used in democratic politics and introducing a set of asymmetries specific to the autocrat. The rationale for this approach is straightforward: rather than creating electoral institutions from scratch, autocrats generally modify and adapt existing democratic institutions to suit their needs. As a result, on paper, many autocratic legislatures bear strong resemblance to their counterparts in democratic polities, though they may seem to function quite differently in practice. The constitutionally specified law- making procedures are often quite similar to procedures used in democratic contexts. There is, of course, substantial variation in how institutions are structured across autocracies – as with democratic states. It would be impossible to capture this broad variation in a single model; I therefore focus on a globally prevalent feature of autocracy and adapt its closest approximation from the formal literature on democratic institutions. I construct a model of legislative bargaining in an autocratic context based on a par-

1See for example Myerson (2008); Svolik (2009).

86 liamentary structure in which the autocrat has particular influence over the formation or dissolution of a governing coalition. This minimalist conception of autocracy is used to illustrate how a common but seemingly unremarkable feature of autocracy might influence distribution in the autocrat’s favor. In this model of authoritarian politics, the autocrat seeks to maximize his share of the budget, and he is granted an undemocratic amount of power to do so. But he is also subject to certain constraints, imposed by the institutional procedures of the legislature, which mimic those in ordinary parliamentary settings. The game is divided into organizational periods, in which players bargain over proposal power, and policy periods, in which members bargain over distributional policy. The autocrat, though a player like others, has unique authorities: he alone has the ability to dissolve an existing coalition and return the game to an organizational period. This reflects a common feature in autocratic states in which the head of state is granted the ability to dismiss cabinet members or dissolve an elected legislative body. In addition, his recogition probability in organizational periods may exceed that of other players, meaning he may have a greater level of influence over government formation. This is akin to a regime with the ability to manipulate electoral contests in its favor. Finally, in a variant of the basic game, his decision to reorganize is linked to his vote on policy proposals. These modifications are intended to capture some of the constitutional privileges and manipulations practiced by many dictators. I find that the ability to reorganize alone creates an incentive for other players to include the autocrat in governing coalitions, since it is only by including him that they can ensure coalitional stability and retain the benefits of rule with certainty. This, in turn, creates a distributional distortion in the autocrat’s favor, as he is assured of access to proposal power. Notably, this outcome does not depend on the autocrat’s ability to manipulate elections, and, in fact, payoffs are equivalent in expectation to a scenario in which the autocrat has complete control over government formation but cannot dissolve the coalition once formed. In the variant with coupled decision-making, other members of the ruling coalition have

87 an incentive to include the autocrat in their policy proposals – again, this incentive arises from their interest in staying in the existing coalition. This results in consistent policy inclusion for the autocrat and, when his recognition power in organizational periods is sufficiently high, an increased expected payoff relative to a game in which he votes separately on policy and coalition dissolution. In both cases, payoffs are asymmetrically distributed relative to a democratic setting. This model borrows from a rich literature on legislative bargaining in democratic settings. Diermeier and Feddersen (1998) use a similar organizing and proposal framework to develop an understanding of the significance of the ruling coalition and institutional rules that affect its formation. In another seminal work, McCarty (2000) examines how executive authority – specifically, the presidential veto – may distort legislative allocation. The model of autocratic policy described here is situated in between these two approaches, borrowing the procedural features from the former but incorporating the asymmetry of the latter. This framework allows for an exploration of the impact of different institutional choices on distributional distortions and a comparison with more democratic institutional arrangements.

3.2 Autocratic Dissolution Power

It is increasingly common for autocratic states to assign the passage of legislation to elected legislatures (Figure 3.1). But relatively little is understood about how the unique features of autocracy – speciﬁcally, a balance of power that privileges the head of state relative to other political ﬁgures – impacts law-making within such bodies. As many scholars have noted, there is substantial diversity in autocratic regimes and the institutions they adopt (Gandhi and Lust-Okar, 2009). Though Figure 3.1 makes it clear that elected legislatures are common in these polities, the exact institutional structure may vary substantially from one autocracy to another. There are, however, some features that generalize to a wide subset of autocracies. Here, I identify a common feature that serves as motivation for the modeling approach used: the authority of the autocrat to dissolve

88 1.00 ● ●●●● ● ●●●●●● ● ●●●●● ●●●● ● ● ● ●● ●● ● ● ●● ●● ●●●●●●● ●● ● ●● ● ●● ● ● ●● ● ●● ● ● ● ● ●● ●●● ● ●● ● ● ● ● ● ●●● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ● ● ● ● ● ● ●●● ●●● 0.75 ● ● ● 1.00 ●

● ● ● ● ● ●● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ● 0.75 ●● ● ● ●●●●● ● 0.50 ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ●● ●● ● 0.50 ●

● 0.25 ●● ● ●●● ●● ●● ●● ●● ● ●● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● 0.25 Proportion of Legislatures that Legislate

0.00 0.00 1900 1925 1950 1975 2000 1900 1925 1950 1975 2000 Year Year

Figure 3.1: Use of elected legislatures within autocratic states, 1900-2017. Left panel shows the proportion of autocratic states by year with elected legislatures. Restricting attention to those states with elected legislatures, right panel shows the proportion for which legislative approval is required by law to pass legislation. Plots are constructed using data from the Varieties of Democracy project, using the “Regimes in the World” measure to identify autocratic states. the legislature, necessitating new elections and the formation of a new government. Other privileges may have additional distributional consequences, but the autocrat’s ability to reshuﬄe the government is one focus of this paper. The prevalence of this feature has increased along with the use of elected institutions in authoritarian contexts (Figure 3.2): since the 1960s, the heads of state of roughly half of all autocracies have possessed the ability to dissolve an elected legislature at their discretion. This is in stark contrast to democratic states, for which this authority has generally been present in less than 10% of regimes. What does this mean in practice? Many aspects of election and government formation under parliamentary autocracy may parallel typical procedures from a parliamentary democracy: legislative elections are held, the head of government is appointed from the winning party and forms a government, and this government proposes laws for MPs to vote on. But whereas in democratic settings, an elected legislature typically serves until its term is complete (or until a coalitional breakdown, such as in the event of a conﬁdence vote precipitated by legislators themselves), in this context, the autocrat himself has a discretionary ability

89 Nonautocratic States Autocratic States

1.00 1.00

0.75 0.75

● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ●● ●●● ● ●● ●● ● ●● ●●● ● ●●●●● ● ●● ● ●● ● ●●● ● ●● 0.50 0.50 ●● ● ● ● ● ●● ● ●●●● ● ● ●● ● ● ●● ● ● ●●●● ● ● ● ●● ●● ●● ● ●● ● ● ● ●●● ●● ●● ●● ●●● ● ● ● ●●●● ● ● ●● ● 0.25 0.25

● ●● ● ● ●●●●●● ●● ●●●●● ●●● ● ● ● ● ●●●●●● ● ●●●●●●●● ●● ●●●●●●●●●● ●● ●●● ●●●●● ● ● ● ●●●●●● ●● ●●● ●●● Proportion Legislature that can dissolve ●●●●●●●●●●●● ●●●●●●● ●● Proportion Legislature that can dissolve 0.00 ●●●●●●●●●●●●●●●●●● 0.00

1900 1925 1950 1975 2000 1900 1925 1950 1975 2000 Year Year

Figure 3.2: Executive powers in nonautocratic versus autocratic states. Figure shows the proportion of states for which the head of state has the power to dissolve an elected legislature at his discretion; left panel includes only non-autocratic states, while right panel includes only autocratic states. Plot is constructed using data from the Varieties of Democracy project, using the “Regimes in the World” measure to identify nonautocratic and autocratic states.

to terminate the elected government.2 I explore how this ability may influence behavior in such a setting, including the coalitions and policies that are proposed. Though some attention has been paid to distributional differences between democracy and autocracy, the evidence is somewhat mixed and does not reference specific institutional features.3 Mulligan, Gil and Sala-i Martin (2004) find little evidence that autocracies are distinct in their total expenditures on public sector activities, including government con- sumption, and education and social spending. On the other hand, several studies have found evidence that autocratic governments facilitate increased favoritism. Do, Nguyen and Tran (2017) argue that in Vietnam, access to positions of power leads to local patronage. Similarly, Hodler and Raschky (2014) find that autocracies are associated with more regional favoritism relative to democracies. Finally, the literature on ethnic favoritism suggests that

2Others have explored the use of the confidence vote or other procedures that generate early elections in parliamentary democracies (Diermeier and Feddersen, 1998; Becher and Christiansen, 2015). A key distinction between the ability of the autocrat to terminate the legislature and these democratic institutions is the assumption that, while a prime minister must risk his own position in a vote of confidence gambit, the new legislative elections do not change the status of the autocrat in the game. 3There is a larger literature on broader, state-level economic outcomes under different regime types – see, for example, Boix (2003); Wright (2008); Haggard and Kaufman (2012).

90 leaders’ ability to preference their coethnics is heightened under autocratic governments (Franck and Rainer, 2012; Burgess et al., 2015). There is thus some support in the literature for the idea that distribution plays out differently in autocratic settings, with evidence in support of heightened favoritism. This may well correspond to greater asymmetry in distribution. However, there is somewhat less clarity on why we should expect this to be the case. In particular, the cited studies do not focus on the specific institutional circumstances or mechanisms that facilitate these patterns: in the cross-national studies, governments are grouped by high-level categorizations such as Polity IV score. Are institutionalized authorities such as legislative dissolution power associated with distributional outcomes under autocracy? Autocratic regimes often lack transparency with respect to policy, and data on distribution in such settings is generally scarce. It is also difficult in many cases to identify the autocrat’s preferred spending policy on a cross-national basis; he may in differing cases favor his region of origin, his coethnics, or his extended family and their interests. To address these challenges, I examine state expenditures in a limited case: that of military regimes, in which the head of state is a military officer. The assumption here is that the military constitutes a preferred group for the head of state in such settings, and higher military expenditures are thus more reflective of his distributive preferences. In Figure 3.3, I plot annual military expenditures relative to total government spending for military regimes in which the head of state has or lacks the discretionary ability to dissolve the legislature. A comparison of the distributions suggests that military spending (relative to total government spending) is about 3 percentage points higher in regimes with dissolution power – a substantial difference and one that suggests that this authority is not a trivial one.4 Though this evidence is descriptive, it is consistent with the idea that institutional choices are not superficial, and that they may have important implications for governance outcomes. 4I further explore this relationship in Appendix C.2 and find that it is statistically significant and robust to the inclusion of a variety of controls.

91 50 ●

●

● ● ●

30 ● ● ● ● ● ● ● 20

10 Military Spending (% of Total Gov. Spending) Gov. Military Spending (% of Total 0 No Dissolution (n = 191) Dissolution (n = 263)

Figure 3.3: Military expenditures relative to total government spending as a function of legislative dissolution power. Data are in country-years and include only regimes in which the head of state is a military oﬃcer. Plot is constructed using data from the Varieties of Democracy project, the World Bank, and the Database of Political Institutions.

3.2.1 Legislative procedures in practice

In the model that follows, I consider a setting in which legislative voting power is assigned to a broad set of actors, but proposal power is limited to a subset of politicians. This approach reflects the influence of the executive over many elected legislatures. Though the legislature’s approval is typically required to pass legislation, it is common in many autocratic regimes for legislation to originate with the executive, including the head of state and cabinet.5 The fact that legislation is passed by, but often not initiated by, elected deputies has led to the characterization of many such bodies as ‘rubber stamp’ institutions (Blaydes, 2010; Malesky and Schuler, 2011). And yet, due to data limitations in these settings, there is relatively little evidence regarding the actual distributional consequences of these institutional features. In this paper, I seek to offer a theoretical argument for how autocratic privileges might impact

5This is also, of course, a feature of democracies with parliamentary institutions, in which both the government and elected MPs have the authority to propose legislation – but most legislation with larger distributional consequences, such as the annual budget, is initiated by the government.

92 policy-making. To illustrate the validity of the modeling framework used here, I consider the example of Morocco, a constitutional monarchy that has had an elected legislature since 1963. Today, Morocco has a competitive political environment with a large number of political parties competing for seats in the legislature and influence over cabinet ministries. Though the constitution grants the king substantial authorities over governmental function, laws are debated on and passed within these legislative institutions, making this a classic example of a hybrid regime. While legislative seats are highly sought after, a majority of bills are proposed by the cabinet: during the 2011-2016 term, 56% of laws passed by the lower house were initiated by the government. This includes some of the most critical laws with distributive consequences, such as the annual finance law.6 And elections are instrumental in determining the composition of the government, despite the monarch’s perceived backroom influence over appointments: the constitution requires that the prime minister be appointed from the party that attains the most votes, and it is he who proposes the remaining members of the cabinet. In recent years, national elections overseen by international observers have been characterized as competitive and largely free of irregularities,7 though this has varied historically. Due to the fragmented party system, no party has ever achieved an outright majority, so government has been formed by coalition. Cabinet ministries are a highly sought after commodity among party leadership due to their influence over proposed policy and implementation. However, also enshrined in the Moroccan constitution is the ability of the king to dissolve one or both chambers of parliament at his discretion, initiating new elections and the formation of a new government. This is an essentially unfettered power: all that is required is that he “consult” with the constitutional court and inform the prime minister.8 A new

6Deputy interview, January 30, 2018. 7See, for example, this 2011 legislative election report by an international NGO National Democratic Institute (2011). 8The only additional restriction is that this power cannot be deployed more than once in a year (i.e. immediately used on the replacement legislature), per Article 98 of the constitution.

93 election is to take place within two months of the dissolution. Historically, this privilege has been employed on only a few occasions: in 1965, shortly after the first parliament was elected, in response to uprisings, and in 1972 and 1983, following similar periods of unrest that led to constitutional revisions. It has been decades since the last such action, and since 1993, parliamentary terms have followed their expected five-year schedule. But lack of use does not equate to lack of importance; the mere existence of this monarchic authority may influence both government formation and the policies that are voted on in the legislature. Figure 3.2 makes it clear that this is a common feature of autocratic constitutions. It is worth noting that, in practice, legislative dismissal manifests in a variety of different ways. In some autocracies, the autocrat exercising this option is seen as normal, while in others, it may lead to greater upheaval, such as wide-scale protests or major revisions to the country’s governing structure. In Peru, in 1992, President Fujimori disbanded an uncooperative legislature and followed that action with a new constitution that expanded his own power. In Kuwait, by contrast, the ruling emir has dissolved the national assembly six times since 2006 without substantial modification to the existing governing structure and generally following the constitutionally prescribed procedures for holding new elections. Scenarios are also distinguished by what follows dismissal: in some cases, dissolved parliaments are quickly replaced with newly elected bodies, while in others, a state of emergency is declared and the autocrat rules by decree for an extended period. The objective in this paper is not to disentangle the intricacies of each autocratic setting, but rather to abstract from any specific context and to consider the potential significance of this feature in general.

3.3 The Model

I consider an n-person legislature (n ≥ 3 and odd), with players indexed by i. One member of the legislature is the autocrat; this player is granted certain authorities as a result of his position which are detailed below. For convenience, the autocrat is always player 1. Let L = {1...n} designate the set of legislators.

94 The game is divided into a total of T periods, where T ≥ 4 and is exogenously determined. In each period, one player is recognized to make a proposal, which is voted on by the remaining players; the nature of the proposal depends on the type of period, either organizational or policy. During organizational periods, nature recognizes one player in the legislature to propose a ruling coalition Q ⊆ L. During policy periods, one member of the ruling coalition is recognized to propose a distributional policy, which similarly passes if a majority of players vote in favor. Policies consist of divisions of a dollar among the group

Pn of legislators, and take the form x¯ = {x1, x2 . . . xn} such that i=1 xi = 1. Recognition in policy periods is random: for a ruling coalition of size q ≤ n, the probability that legislator i is called on to make a proposal is 1/q if he is a member of the ruling coalition and 0 otherwise. Whether coalitional or distributional, proposals pass with a simple majority of support. The autocrat is distinguished from other players by two key features. The ﬁrst concerns the ability of the autocrat to appoint the government: the probability that he is recognized

1 in an organizational period is rˆ ∈ [ n , 1], while the probability of recognition for every other 1−rˆ player is r˙ = n−1 . This is intended to capture the strength of autocratic control over government formation, whether via constitutional privilege or electoral manipulation. In the most competitive case, proposal recognition in the organizational period is fully egalitarian

1 (rˆ =r ˙ = n ). As the autocrat’s ability to manipulate elections increases, so does the likelihood that he determines the makeup of the ruling coalition. The autocrat also has the privilege of reorganization: uniquely among the players, he has the ability to dissolve the existing ruling coalition at the end of a policy period and return the game to the organizational stage. This reﬂects the common authoritarian feature allowing autocrats to dissolve an existing legislature at their discretion. In the basic version of the game, following a policy vote, the autocrat decides separately whether to dissolve the existing ruling coalition. In a second variant, decisions are linked such that a vote by the autocrat in favor of a policy is also in favor of the existing coalition, while a vote

95 T T-1 T-2 ...1

L− L− L− OT OT −1 OT −2 ...O1

L+ L+ L+ A− A−

P T −1 P T −2 ...P 1 A+ A+

Figure 3.4: Game structure. Figure shows progression of a game. Each period is either organizational O or policy P . Transitions are determined by approval (+) or rejection (-) of an existing coalition or coalition proposal and indicated by arrows. Transitions that depend on majority support from the entire legislature (L) are indicated by solid lines, while those that depend only on the autocrat (A) are indicated by dotted lines. against a policy proposal automatically converts the next period to organizational; I refer to this throughout as ‘coupled’ decision-making. Note, however, that the ability to reorganize the ruling coalition is distinct from veto power: in this game, a policy may pass without autocratic support. For notational purposes, periods are enumerated in reverse order such that the ﬁrst period is period T and the ﬁnal period is period 1. The game begins with an organizational period, and proceeds as follows: in an organizational period, if the proposed coalition is accepted, the game progresses to a policy period. If the proposed coalition is rejected, the next period is organizational and a new player is recognized to propose a coalition. This repeats in subsequent periods until a coalition proposal passes or the game ends. Once the game transitions to a policy period, subsequent periods are also policy periods regardless of whether or not the distributional proposal passes, unless the autocrat dissolves the coalition and returns the game to the organizational stage or until the game ends. Figure 3.4 illustrates the structure of the game and the possible transitions between periods. In organizational periods, a majority of the legislature voting on a coalition proposal determines whether the next period is organizational or policy; in policy periods, the autocrat alone decides whether or not to dissolve the existing coalition and thus what type of period follows.

96 1 In organizational periods, all players receive a default payoﬀ of n . In policy periods, if

a proposal passes, each player i receives xi, the amount allocated to them by the proposal. If the policy fails, players receive a default payoff equivalent to the organizational period.9 There are no pre-existing factional preferences within the legislature; policy proposers seek to maximize their own payoff across all periods and will, all else equal, offer concessions to those players whose support is cheapest to purchase. I further assume that players have a strict preference order with respect to the other legislators to avoid indifference; this is randomly assigned in each period. The history of play ht determines the type of period and the set of players eligible for proposal recognition. I characterize subgame perfect equilibria. Equilibria are symmetric with respect to all non-autocrat players. Strategies are defined as follows. For an organizational period t, nature recognizes a player as a function of r. The recognized player chooses a coalition Q ∈ S, where S represents all possible combinations of legislators in L. The coalition strategy for player i in period t is defined as Q : Ht → S. Following the coalition proposal, legislators vote to accept (A) or

t reject (R); their voting strategies ai : H × S → {A, R} map possible coalitions onto this choice. If a majority of players vote to accept the proposal, the coalition is accepted. In policy periods, one player i ∈ Q is recognized to propose a policy x ∈ X, where X represents all possible distributions across the set of legislators. A policy strategy for player

t i in period t is deﬁned as x¯i : H → X. Voting strategies on a policy proposal are deﬁned

t as bi : H × X → {A, R}. If a majority of players vote to accept the proposal, the policy is enacted. In the coupled variant, the autocrat voting to reject the proposals moves the game into an organizational period. In the decoupled game, the autocrat separately chooses whether the next period will be an organizational one d : Ht × X × {A, R}n → {O,P }.

9As in other models of legislative voting, I assume that legislators have a lexicographical preference for voting in support of a policy.

97 3.4 Results

A key objective of this model is to examine the distributional implications of institutional rules under autocracy and the distortions that occur relative to a symmetric (‘democratic’) legislature. Additionally, I seek to identify behavior in organizational periods: particularly the circumstances under which the autocrat will be included in a ruling or policy coalition and the stability of such coalitions over multiple periods. I begin by exploring results in the basic version of the game with decoupled decision-making. In characterizing results here and elsewhere, I denote expected payoﬀs by period, relative

t to the final period 1. Continuation payoffs vp are differentiated based on the player (autocrat vˆ or non-autocrat v˙), period (t ∈ {T...1}), and period type and player status (whether it is an organizational period or, in a policy period, whether the player is in the ruling coalition or not p ∈ {O,C,N}).I also find it convenient in some instances to describe single period

expected payoﬀs up as a function of player status and type of period.

3.4.1 Policy Periods

I first consider strategies in the final period of the game. If the last period is a policy period, players will support any policy that offers them at least the default payoff.10 To attain a majority of votes, then, the proposer must offer at least the default payoff to a

n−1 minimum of 2 other players. The recognized player maximizes his payoff by offering the default payoff to exactly that many other players based on his randomly drawn preference

n+1 ordering, leaving him a surplus of 2n . It is important to note that in the last period of a game, neither the autocrat’s reshuffle privilege nor his recognition probability impact play, as they have no effect in the immediate term (i.e. on the payoff for that period). He therefore has no unique attributes in this period and will be treated symmetrically to other players. As a result, payoffs in this period depend only on period type and whether or not a player

10If the last period is organizational, strategies are irrelevant: as each player receives the default payoﬀ regardless of the coalition proposed or whether or not it passes, and the game subsequently ends. Hence any coalition proposal Q is an equilibrium strategy, and payoﬀs are as described in equation 3.3

98 is a member of the ruling coalition. Given that proposal recognition probability is 1/q, and unrecognized players have a 1/2 probability of being oﬀered their default payoﬀ, it is straightforward to calculate the expected utility for all players given a ruling coalition of q players.

n + q vˆ1 =v ˙ 1 = (3.1) C C 2nq 1 vˆ1 =v ˙ 1 = (3.2) N N 2n 1 vˆ1 =v ˙ 1 = (3.3) O O n

I next consider behavior and payoffs in preceding policy periods, taking into account the possibility of reorganization and players’ status as members of the ruling coalition or not. In policy period 2 and all preceding policy periods, the autocrat can, at his discretion, exercise his ability to reshuffle leadership. The autocrat’s ability to reorganize does not impact the policy proposals themselves. Suppose the recognized player seeks to buy the autocrat’s support through inclusion in a policy offer. Because his decision to reorganize is independent of his vote on a policy proposal, the autocrat can accept the offer but still move the game to an organizational period.

Lemma 1. When the autocrat’s decision to reorganize is independent of and follows his vote on a given policy (decoupled), this decision depends only on his expected payoﬀ in subsequent periods.

There is thus no additional incentive for other players to include the autocrat in a policy proposal. The recognized player’s optimal policy proposal strategy is thus equivalent to that of the ﬁnal period, and symmetric across all players in the ruling coalition, regardless of type. As a result, it is possible to characterize single period payoﬀs up as a function solely of player and period status (p ∈ {O,C,N}). For any organizational period t, the expected

1 payoﬀ for all players is uO = n . In any policy period t, the single period expected payoﬀ

99 is simply a function of whether or not a player is in the ruling coalition and equivalent to the final period payoffs above. Expected single period payoffs for any policy period are thus dependent only on whether a player is part of the ruling coalition:

n + q u = (3.4) C 2nq 1 u = (3.5) N 2n

3.4.2 Coalition Formation

Finally, I consider behavior in organizational periods. What coalition will the recognized player propose, and what will other players accept? It is clear from uC that the expected utility for members of the ruling coalition in any given policy period is decreasing in q, the

n+1 size of the coalition, yet a majority of votes ( 2 ) is required to pass a coalition proposal.

It is also obvious that uC > uN , so all players will support any proposal that includes them in the coalition. The recognized player will therefore include himself in any proposal, and

n−1 otherwise minimize the size of the coalition, offering a position to 2 other players. For ∗ n+1 simplicity in describing equations elsewhere in the results, I use q = 2 to refer to this minimum number of members in the ruling coalition.11 Consider again the autocrat’s role: he cannot block a coalition with majority support from taking effect, but only disband an existing government following a policy period (that is, once the legislature is ‘in session’).12 But other players will take the autocrat’s incentives to reorganize into account when proposing a coalition. Under what circumstances will the autocrat reorganize? From Lemma 1, in the second to last period, this decision is a function only of his expected payoff in the subsequent period. This in turn depends on the type of period and whether or not he is a member of the ruling coalition. Because uC > uO > uN ,

11For an explanation and generalization of this minimum winning coalition, see Appendix Section C.1.1. 12As a result, this authority is relevant only if there are at least two remaining periods: if the subsequent period is the last, he cannot aﬀect expected payoﬀs by reorganizing. The two-period game is thus a special case.

100 he will reorganize if and only if he is outside the ruling coalition.

This pattern holds true for preceding periods as well: for any given period, uC offers the highest possible single period payoff, so an autocrat in the ruling coalition will not reorganize. Even if he were assured of returning to the ruling coalition in a subsequent organizing period, he has no incentive to opt for a one-period deviation to his organizational payoff. Similarly, uN is the lowest possible single period payoff, so an autocrat outside the ruling coalition will always reorganize. Even if he expects to be excluded from the next coalition formed, a one period deviation to the organizational period would be an improvement. In an organizational period, then, the recognized player faces a choice between excluding the autocrat from their proposed coalition and having it fall apart after the next policy period, or including the autocrat and ensuring coalitional stability through the end of the game. Again, because payoffs from repeated policy periods as a member of the ruling coalition exceed even a one period deviation to an organizational period, the recognized player will always prefer a stable coalition and can ensure it by including the autocrat in his proposal. The result is that the autocrat is assured of inclusion in any ruling coalition proposal in any organizational period t > 2. This, along with the results in the preceding section, allows me to characterize expected payoffs for this game:

Proposition 1. Coalitional Stability and Payoﬀs under Decoupled Decision-making. Play- ers will form a stable ruling coalition of size q∗ including the autocrat. The ex ante continuation payoﬀ for players in period T is as follows:

3nT − n + T + 1 vˆT = (3.6) O 2n(n + 1) 2nT + T + 1 v˙ T = (3.7) O 2n(n + 1)

The autocrat’s expected payoﬀ is strictly greater than that of other players, and this diﬀerence is increasing in T .

The autocrat’s unique ability to dissolve an existing ruling coalition thus creates an

101 incentive to include him in the ruling coalition, regardless of who proposes it: his recognition probability rˆ is irrelevant to the T -period expected payoffs. His inclusion is driven by the fact that he profits more from policy periods as a member of the ruling coalition than from organizational periods. However, because this authority is exercised independently from policy votes, it has no impact on policy proposals. The autocrat’s ex ante utility is thus greater than that of other players, but this is simply the result of his certain inclusion in the ruling coalition. Ex interim expected utilities are symmetric across all members of the ruling coalition, including the autocrat. As a benchmark, we might compare the autocrat’s expected payoff to that he and others would receive in a more ‘democratic’ game. In a fully symmetric setting (that is, absent the ability to reorganize or the increased recognition probability), ex ante expected payoffs for all players in a T period game are T/n. Using payoffs characterized in Proposition 1, we can thus calculate the autocrat’s surplus relative to this scenario.

Remark 1. An autocrat with the power to reorganize existing coalitions has a total expected

(n−1)(T −1) surplus of 2n(n+1) relative to a player in an equivalent but fully symmetric game. This surplus is increasing in T and decreasing in n.

As the autocrat is treated symmetrically in policy periods, the surplus in Remark 1 reflects the rent from the other players’ endogenous constraint of including him in coalition proposals. It is notable that the recognition benefit rˆ plays no role in this result: because other players will always include the autocrat in the proposed coalition, he benefits regardless of whether he is the one recognized. As a result, we can see that the result in Proposition 1 is equivalent to a game in which the autocrat does not have the ability to dissolve ruling coalitions, but does have complete discretion over coalition formation (rˆ = 1).13 This suggests

13To see this, consider that in such a game, a coalition once formed is immutable. If the autocrat is always recognized in an initial organizational period, he will include himself in any proposed coalition, along with the minimum number of additional players to ensure the coalition passes. Play in policy periods will be unchanged. As in Proposition 1, then, the coalition will be stable, size q∗, and include the autocrat with probability 1.

102 that these two features – dissolution power and electoral manipulation to a deterministic level – are interchangeable with respect to distributive policy.

3.5 Coupled Decision-making

In this section I consider a variant of the basic game in which the autocrat’s ability to reorganize is linked to his vote on policy proposals. The game described in the preceding sections offers the autocrat only one authority – the power to reorganize – but grants him full autonomy in choosing when to deploy it. Suppose legislative institutions are instead structured such that the decision to reorganize is linked to the autocrat’s vote on a policy proposal, akin to a confidence vote exercised by only one of the players.14 In practice, this might be constitutionally enshrined, or it may manifest in the form of backroom deals or contracting between the autocrat and members of the ruling coalition. How does this modification affect behavior and distribution? This variant, which I refer to as coupled decision-making, changes the stakes of a policy proposal for both the autocrat and other players. Here, the survival of the existing coalition depends on the autocrat’s policy vote. For non-autocrat players, this creates an incentive to include the autocrat in policy proposals: players in the ruling coalition prefer to ensure that the coalition remains stable and they continue to benefit in future periods. What must other players offer to obtain autocratic support for a policy? In the previous variant, the autocrat and all other players would support any policy that offered them at least their continuation payoff of 1/n. But the autocrat must now simultaneously consider the policy offer and the impact of the existing coalition on his future payoffs. In other words, the tradeoff for him is no longer between only the one-shot policy offer and his continuation payoff, but instead between the one-shot policy offer plus expected future payoffs with the existing coalition and his continuation payoff plus expected payoffs from reorganization. If he

14Note, however, that unlike in a conﬁdence vote (see Diermeier and Feddersen (1998)), the failure of the coalition is separate from the success or failure of the policy under consideration – it is linked only to the autocrat’s voting decision. In this game, a policy may still pass without autocratic support.

103 is a member of the ruling coalition and expects to benefit from that status in future periods, he has an additional incentive to support a policy and would accept a correspondingly lower offer. If he is outside the ruling coalition and expects a suboptimal payoff in the subsequent periods, he has an incentive to vote against a policy and requires compensation in the form of a larger offer to vote in favor. To see how this works, consider decision-making in the final periods of play. This change has no effect on the final period of play: as with the basic game, the autocrat’s influence is relevant only when sufficient periods remain for him to exercise his ability to reorganize. In the final period, then, players’ expected utility depends only on period type and membership in the ruling coalition and are equivalent to equations 3.1-3.3. In preceding periods, however, this induces a shift in behavior. Suppose the second-to-last period is a policy period and a non-autocrat is recognized to make a policy proposal. If his vote is pivotal, the autocrat will accept any policy that offers him at least the difference between two continuation payoffs and his expected payoff from a final policy period. If he is a member of the ruling coalition, then all else equal, he prefers not to reorganize,

1 1 since vC > vO for all q < n:

2 n + q 2u − v1 = − O C n 2nq 3 1 = − (3.8) 2n 2q

1 Note that this difference is less than the default payoff of n ; the autocrat therefore will accept a lower offer than other players. If, however, the autocrat is outside the ruling coalition, he

1 3 requires at least 2uO −vN = 2n to support a proposal. This is greater than the default payoﬀ required by other players. Preceding periods follow the same pattern: so long as the autocrat expects a higher payoﬀ from subsequent play given the existing ruling coalition (i.e. when he is a member of that coalition), he has an additional incentive to vote in favor of a policy proposal. If he

104 expects a lower payoff in future periods, he has an incentive to dissolve the coalition unless sufficiently compensated in the present – hence he is more costly than other players. Next, consider the motivation of a non-autocrat player with policy proposal power. She prefers that the game remain in the policy phase for the final period, as her expected payoff exceeds what she would receive from an organization period. To ensure this is the case, she must include the autocrat in her policy offer. If the autocrat is a member of the ruling coalition and q < n, his support is cheaper than that of other players. It therefore maximizes her expected payoff to include the autocrat in the ruling coalition with an offer of (3.8) and

n−3 1 ﬁll out the coalition with 2 other players at cost n each. If, on the other hand, the autocrat 3 is outside the coalition, he requires at least 2n to support a policy. This exceeds the surplus the proposing player would receive from remaining in the ruling coalition during the ﬁnal

1 1 policy period (vC − vO). As a result, the autocrat is too costly to buy off; she will therefore exclude him from her proposal and resign herself to reorganization. This pattern continues in preceding periods: because the autocrat is cheaper when he is in the ruling coalition, players will buy him off with policy offers. If he is outside the ruling coalition, his support is too expensive to procure. This in turn informs behavior in organizational periods t > 2. Only a ruling coalition that includes the autocrat can be maintained across multiple periods. The recognized player thus faces a choice between excluding the autocrat and creating a short-lived coalition or including the autocrat in her proposal and ensuring her coalition lasts. As before, members of the ruling coalition do better when the autocrat is included in the coalition, because it is both stable and cheaper to maintain. The nominated player will therefore include the autocrat in her proposed coalition.15 So again, the autocrat will be included with certainty in any ruling coalition formed before the second-to-last period: if he is recognized, he will include himself, and other players will

15Note that this holds for periods t > 2. In t ≤ 2, there is no special incentive to include the autocrat in a coalition proposal as he will not have an opportunity to disband a coalition in a way that aﬀects payoﬀs before the game ends.

105 do the same. Unlike in the basic game, however, he will also be included in subsequent policy proposals prior to the ﬁnal period. When he is recognized to make a policy oﬀer, he

1 n−1 will follow the same strategy as in the decoupled version, offering n to 2 other players and keeping the surplus for himself. When other players are recognized, they will make him an offer that takes into account how much he prefers to keep the existing coalition in the subsequent period.16 Collectively, the best responses of both autocrat and non-autocrat players allow us to derive ex ante payoffs for the full game.

Proposition 2. Coalitional Stability and Policy Inclusion under Coupled Decision-making. In a game with binding institutions such that the autocrat votes simultaneously on a policy proposal and existing ruling coalition, players will form a stable ruling coalition including the autocrat. The autocrat will further be included with certainty in subsequent policy proposals until period T . The ex ante continuation payoﬀs for players in period 1 are as follows:

n5(6T − 3 +r ˆ) + n4(−8T + 6 − 3ˆr) + n3(−4T − 1 + 3ˆr) + n2(8T − 3 − rˆ)− T 2nT + 1 + 1−n (n5(ˆr − 1) + n4(3ˆr − 4) + n3(3ˆr − 3) + n2(ˆr + 1) − 1) vˆT = n+1 (3.9) 4n3(n − 1)2(n + 1) 4n6T − n5(10T − 3 +r ˆ) + n4(4T − 6 + 3ˆr) + n3(8T + 1 − 3ˆr) + n2(−8T + 3 +r ˆ)+ T 2nT − 1 + 1−n (n5(1 − rˆ) + n4(4 − 3ˆr) + n3(3 − 3ˆr) + n2(−1 − rˆ) + 1) v˙ T = n+1 4n3(n − 1)3(n + 1) (3.10)

Proposition 2 clariﬁes the beneﬁt to the autocrat of enacting strong institutions. His authority over coalition survival again produces an incentive to include him in the government, but the restrictions on that authority generate an incentive for other players to include him in policy proposals as well in order to maintain the existing ruling coalition. The autocrat’s

16The minimum oﬀer in the second-to-last period was derived above. In earlier periods, the autocrat’s minimum policy oﬀer is also a function of his recognition probability in organizational periods. This is because in organizational periods where at least two subsequent periods remain, other players will always nominate the autocrat as part of their ruling coalition in order to ensure coalitional stability. Yet in an organizational period 2, however, there is no incentive to include the autocrat in a coalition that will only last one period. His expectation of being in the ruling coalition in period 1 is thus a function of his recognitional probability r.

106 ex ante payoff is strictly greater than that of other players, and the difference is increasing in rˆ. We can also compare the autocrat’s expected payoff to what he expects during the basic game with discretionary dissolution. In the variant with coupled decision-making, the autocrat receives a smaller cut of policy offers when he is not recognized because he is predisposed to vote to maintain a coalition he is a part of. However, because he is always included in policy proposals, his expected payoff from each policy period exceeds that of the basic game under the conditions described in the following remark.

Remark 2. The autocrat’s expected payoﬀ with coupled decision-making exceeds his payoﬀ under decoupled decision-making if and only if:

T 5 4 3 2 1−n 5 4 3 2 n − 2n T + n (6T − 5) + n (5 − 6T ) + 2nT − 1 + n+1 (n + 4n + 3n − n + 1) rˆ > (3.11) T 2 3 2 1−n 3 2 n n − 3n + 3n − 1 + n+1 (n + 3n + 3n + 1)

Remark 2 establishes the circumstances under which the autocrat benefits from more rigid restrictions on the exercise of his dissolution power (what might be termed a ‘strong’ institution Wright (2008)). When T is odd, the above inequality holds for all r ∈ [0, 1], meaning the autocrat does better with coupled decision-making. Where T is even, this is a binding constraint such that the autocrat does better with coupled decision-making only when rˆ is sufficiently large. In other words, coupled decision-making benefits the autocrat in scenarios where he combines dissolution authority with some ability to manipulate electoral outcomes in his own favor. Finally, we can compare ex interim payoffs for the other players in the game to their expected interim payoffs with decoupled decision-making. Ex interim payoffs for non-autocrat

107 players in and out of the ruling coalition under coupled decision-making are as follows:

1 − nT ! v˙ T −1 =[3n7(T − 1) − n6 13T − 13 +r ˆ + (ˆr − 1) + C n + 1 1 − nT ! 1 − nT ! n5 16T − 16 + 5ˆr + (2 − rˆ) + n4 4T − 5 − 9ˆr − (3ˆr − 5) + n + 1 n + 1 1 − nT ! 1 − nT ! n3 −23T + 24 + 7ˆr + (5ˆr − 7) + n2 17T − 14 − 2ˆr + (2ˆr + 2) + n + 1 n + 1 1 − nT ! 1 − nT n 4T − 1 + − 2 + 2] ∗ [2n3(n − 1)3(n + 1)(n − 2)]−1 (3.12) n + 1 n + 1 n2(T − 1) + n(3 − 2T ) − T v˙ T −1 = (3.13) N 2n(n − 2)(n + 1)

T −1 Remark 3. Non-autocrat members of the ruling coalition have a higher ex interim v˙C under coupled decision-making relative to their counterparts in the basic game or an equivalent, fully symmetric game. Non-autocrats outside the ruling coalition have a lower ex interim

T −1 payoﬀ v˙N under coupled making.

The autocrat’s presence in the ruling coalition and all subsequent policy coalitions reduces other players’ probability of being included in policy coalitions, thereby lowering their expected utility for each policy period. For players outside the ruling coalition, this means their ex interim utility is lower than it would be with decoupled decision-making, as in the basic game. However, the autocrat’s cheaper cost to include in policy coalitions oﬀsets this eﬀect for members of the ruling coalition. The recognized player gets a larger surplus, which overcomes the reduced probability of being included in a policy proposal when others are recognized. As a result, non-autocrat members of the ruling coalition do better under coupled decision-making than in the basic game.

3.6 Conclusion

Institutional autocracy borrows heavily from democratic institutions in establishing a governing structure, yet there are important diﬀerences in the powers allotted to the exec-

108 utive. This paper seeks to explore the ramifications of one common feature of autocratic states with elected legislatures. A brief empirical exploration reveals that legislative dissolution power is associated with distributional distortions: in the limited case of military regimes, this authority is associated with increased military expenditures. With the model presented here, I offer a theoretical justification for this finding, demonstrating that the ability of the head of state to dismiss the legislature and prompt new elections should under most institutional conditions incentivize his inclusion in the governing coalition. As a result, his expected share of distributional policy exceeds that of other players, and of players in a symmetric (‘democratic’) game. This is consistent with conventional wisdom: we expect that autocrats, with their outsize share of political power, will receive corresponding distributional benefits. Yet there is much more to understand about the mechanisms through which they acquire these and which are most strategically appealing. One intriguing implication of the variant with decoupled decision-making is that the autocrat’s expected payoffs with dissolution power are equivalent to a game without dissolution power but with deterministic electoral manipulation. This suggests that the two authorities are interchangeable from a distributional standpoint. But there are reasons to expect that dissolution power is a more efficient, less costly institution in practice: it can be constitutionally enshrined and protected, and need not be employed to be effective. Excessive electoral manipulation may be met with popular or international resistance – and runs counter to explanations for the use of democratic institutions.17 If multiple mechanisms can produce the same distributional distortions in the autocrat’s favor, this grants the autocrat an opportunity to selectively identify the least costly. A second interesting implication of the autocrat’s inclusion in the governing coalition is that the given coalition will be a stable one. This suggests that though the authority is widespread, legislative dissolution may be quite rare in practice. It is only when (if) the

17In particular, overly fraudulent elections may result in popular mobilization against the regime (Van de Walle, 2002; Beissinger, 2007). Even if this does not occur, they may fail to bestow the desired domestic and international legitimacy on the regime (Schedler et al., 2006).

109 autocrat is excluded that he is incentivized to use his power. This is, again, in contrast to electoral manipulation, which must be openly employed at the beginning of each game.18 Finally, the results here demonstrate that autocratic authorities might interact in ways that are not obvious. The variant of the game with coupled decision-making results in a scenario in which the autocrat is not only assured of inclusion in coalitions, but also gets a piece of every policy proposal put forward. And his payoff in this game is related to his recognition probability in organizational periods. In other words, being able to manipulate elections in his favor increases his distributional advantage in such an institutional setting. The results here suggest that there is much to learn from a more formal study of autocratic institutions. This paper covers the implications of one authority for distributional policy and government stability; there are many other institutional features of autocratic regimes that bear examination in future research. These might include the regime’s authority over the judiciary, the frequent policing of political parties based on ideology or affiliation, and the structuring of the legislature to include both directly elected and appointed members. This paper also offers a suggested approach for future research of this kind: given that they are often modified from democratic institutions, autocratic institutions can be studied as such, using adaptations of models from the vast formal literature on democratic institutions to allow for greater comparison.

18Electoral manipulation is not restricted to fraud, but can also draw on the use of electoral rules to advantage the regime’s preferred candidates. Though this may be more institutionalized, it is nevertheless arguably more visible than an authority which is never deployed.

110 Conclusion

Democratic institutions are increasingly prevalent in autocratic polities, and yet much about their function remains poorly understood. Previous scholarship has been primarily focused on high level outcomes such as democratization and political liberalization. Though these are important subjects of study, we can also learn a great deal from studying ‘everyday autocracy’ – in other words, these institutions as they function in equilibrium. To do so requires us to expand our frame of reference to consider not only autocratic breakdown as an outcome, but also the winners and losers of politics on a daily basis. Through my research, I seek to explain the micrologic of how governance works in these settings and the role that institutional rules play in determining this. And, though the focus is on a different set of outcomes (political behavior in office, distribution, and vote choice), I hope that this research can also provide insight regarding the aforementioned transition events by identifying possible sources of popular grievances as well as demonstrating how outsiders can increase their political stake to potentially challenge the regime. Existing scholarly work suggests that democratic institutions benefit the autocrat in terms of survival. But underlying this overarching explanation is the fact that these structures also create a space for (limited) contestation between opposition and regime.19 The papers in this dissertation focus on this dynamic space: I investigate the internal function of institutions in one context to illustrate the ways in which these two actors each try to control and expand their own power. In Chapter 1, I show that the participating opposition uses legislative authorities to provide constituency service. This, in turn, contributes to their popularity with the electorate,

19Not all autocracies have adopted institutions, and some have created institutions purely as façade structures that lack internal contestation. My work is scoped to those regimes (a majority of autocracies) that allow for at least some political contestation in the form of competitive elections.

111 providing an avenue for them to expand their political influence in future cycles. In Chapter 2, I find that regime-controlled ministries are more likely to suppress these same legislative authorities, suggesting that executive appointments are a lever through which the regime can prevent the opposition from growing too powerful. Finally, in Chapter 3, I identify the institutional features that allow the autocrat to share resources and provide normalized access to benefits for other political contestants while remaining assured of his share of the distributional pie. The three chapters together thus illustrate the push and pull between political actors and the mechanisms through which institutional features regulate this political contestation. But many outstanding questions remain for future study. First, these papers are situated primarily in the context of interactions between the regime and the political elite, with some attention to citizen behavior captured by measuring electoral outcomes. But there is much more to understand about this latter group – what they demand from political leaders, what they will tolerate from their representatives and the regime, and how they view institutions in this context. In future work, I will address these questions, starting by testing the argument advanced in Chapter 1 that voters primarily expect constituency service from their representatives. Do voters distinguish between the mechanisms politicians use to provide goods to their district, and do they feel differently about politicians that employ patronage versus institutional authorities as their means of service? I intend to undertake additional data collection at the citizen level (using survey instruments and possibly focus groups) to shed light on the attitudes of ordinary citizens in this setting and their perceptions of governance. Additionally, this dissertation draws motivation from a single context. I have sought to focus on globally common institutional features within this setting and identify potential scope conditions for the arguments put forward where applicable, but there are a number of questions to take on in order to expand the applicability of this broader research agenda. First, there are a range of other institutional features, found within Morocco and other au-

112 tocratic regimes, that bear examining: these include electoral systems (and shifting electoral rules) as well as the policing or abolition of political parties. One objective for future study is to examine political behavior within other regimes; in a first foray, I will take on activity within the Kuwaiti national assembly, using data I have already collected on deputy queries, bill proposals, and roll call voting on legislation. I will generate theoretical expectations for how variation in institutional rules, including changes to the electoral system, alter the behavior of voters and political elites. In a second effort, I hope to take on these same questions cross-nationally, using information on institutional rules across contexts to test how these are associated with outcomes such as partisan participation, voting behavior, and distributional asymmetries. Democratic institutions may contribute to regime longevity in autocracies, and yet they require that the regime relinquish some of its authority to other political contestants. The very existence of this arena of political contestation would seem to pose a potential threat – what happens if the contestation gets out of hand and challengers gain too much political support? Understanding the rules that govern this arena and the mechanisms by which the regime manages – and quells, and even benefits from – opposition participation is the focus of this and future research. Ultimately, I hope to provide insight into the micrologic of this increasingly common regime type and how it is sustained.

113 Bibliography

Abadie, Alberto. 2005. “Semiparametric Diﬀerence-in-diﬀerences Estimators.” The Review of Economic Studies 72(1):1–19.

Adida, Claire, Jessica Gottlieb, Eric Kramon and Gwyneth McClendon. 2016. “When Good News is Bad... and When it Isn’t: Voter Coordination, Preferences, and Electoral Behav- ior.” Unpublished manuscript .

Akirav, Osnat. 2011. “The Use of Parliamentary Questions in the Israeli Parliament, 1992– 96.” Israel Aﬀairs 17(02):259–277.

Angrist, Joshua D and Jörn-Steﬀen Pischke. 2008. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton University Press.

Arab Barometer. 2019. Arab Barometer V: Morocco Country Report. Technical report.

Austen-Smith, David and J Feddersen. 2006. “Deliberation, Preference Uncertainty, and Voting Rules.” American Political Science Review 100(02):209–217.

Becher, Michael and Flemming Juul Christiansen. 2015. “Dissolution Threats and Legislative Bargaining.” American Journal of Political Science 59(3):641–655.

Beissinger, Mark R. 2007. “Structure and Example in Modular Political Phenomena: The Diﬀusion of Bulldozer/rose/orange/tulip Revolutions.” Perspectives on politics 5(2):259– 276.

Benchemsi, Ahmed. 2012. “Morocco: Outfoxing the Opposition.” Journal of Democracy 23(1):57–69.

114 Benstead, Lindsay J. 2008. Does Casework Build Support for a Strong Parliament? Legisla- tive Representation and Public Opinion in Morocco and Algeria PhD thesis University of Michigan.

Bertrand, Marianne, Esther Duflo and Sendhil Mullainathan. 2004. “How Much Should We Trust Differences-in-differences Estimates?” The Quarterly Journal of Economics 119(1):249–275.

Blaydes, Lisa. 2006. “Who Votes in Authoritarian Elections and Why? Determinants of Voter Turnout in Contemporary Egypt.” Political Science 55(4):517–549.

Blaydes, Lisa. 2010. Elections and Distributive Politics in Mubarak’s Egypt. Cambridge University Press.

Boix, Carles. 2003. Democracy and Redistribution. Cambridge University Press.

Boix, Carles and Milan W Svolik. 2013. “The Foundations of Limited Authoritarian Govern- ment: Institutions and Power-sharing in Dictatorships.” The Journal of Politics 75(2):300– 316.

Box-Steﬀensmeier, Janet M and Bradford S Jones. 2004. Event History Modeling: A Guide for Social Scientists. Cambridge University Press.

Brancati, Dawn. 2014. “Democratic Authoritarianism: Origins and Eﬀects.” Annual Review of Political Science 17(1):313–326.

Buehler, Matt. 2018. Why Alliances Fail: Islamist and Leftist Coalitions in North Africa. Syracuse University Press.

Burgess, Robin, Remi Jedwab, Edward Miguel, Ameet Morjaria and Gerard Padró i Miquel. 2015. “The Value of Democracy: Evidence from Road Building in Kenya.” American Economic Review 105(6):1817–51.

115 Cheng, Li and Lynn White. 1990. “Elite Transformation and Modern Change in Mainland China and Taiwan: Empirical Data and the Theory of Technocracy.” The China Quarterly 121:1–35.

Ciftci, Sabri and Tevﬁk Murat Yildirim. 2019. “Hiding Behind the Party Brand or Cur- rying Favor with Constituents: Why Do Representatives Engage in Diﬀerent Types of Constituency-Oriented Behavior?” Party Politics 25(3):369–381.

Clinton, Joshua D. 2017. “Coding the Ideological Direction and Content of Policies.” Annual Review of Political Science 20:433–450.

Darif, Mohammed. 2012. “Morocco: A Reformist Monarchy?” The Journal of the Middle East and Africa 3(1):82–103.

Diamond, Larry. 1999. Developing Democracy: Toward Consolidation. JHU Press.

Diermeier, Daniel and Timothy J Feddersen. 1998. “Cohesion in Legislatures and the Vote of Conﬁdence Procedure.” American Political Science Review 92(3):611–621.

Do, Quoc-Anh, Kieu-Trang Nguyen and Anh N Tran. 2017. “One Mandarin Beneﬁts the Whole Clan: Hometown Favoritism in an Authoritarian Regime.” American Economic Journal: Applied Economics 9(4):1–29.

Domínguez, Jorge I, Ángela Fonseca Galvis and Chiara Superti. 2017. “Authoritarian Regimes and their Permitted Oppositions: Election Day Outcomes in Cuba.” Latin Amer- ican Politics and Society 59(2):27–52.

Fakir, Intissar. 2018. Morocco’s Islamist Party: Redeﬁning Politics under Pressure. Technical report Carnegie Endowment for International Peace.

Franck, Raphael and Ilia Rainer. 2012. “Does the Leader’s Ethnicity Matter? Ethnic Fa- voritism, Education, and Health in Sub-Saharan Africa.” American Political Science Re- view 106(2):294–325.

116 Fujiwara, Thomas and Leonard Wantchekon. 2013. “Can Informed Public Deliberation Overcome Clientelism? Experimental Evidence from Benin.” American Economic Journal: Applied Economics 5(4):241–55.

Gandhi, J and A Przeworski. 2006. “Cooperation, Cooptation, and Rebellion under Dicta- torships.” Economics & Politics 18(1):1–26.

Gandhi, Jennifer. 2008. Political Institutions under Dictatorship. Cambridge University Press.

Gandhi, Jennifer and Ellen Lust-Okar. 2009. “Elections under Authoritarianism.” Annual Review of Political Science 12:403–422.

Geddes, Barbara. 1999. “What Do We Know About Democratization After Twenty Years?” Annual Review of Political Science (1991).

Greene, Kenneth F. 2007. Why Dominant Parties Lose: Mexico’s Democratization in Com- parative Perspective. Cambridge University Press.

Haber, Stephen. 2008. Authoritarian Government. In The Oxford Handbook of Political Economy, ed. Donald A Wittman and Barry R Weingast. Oxford University Press.

Haddadi, Said. 2002. “Two Cheers for Whom? The European Union and Democratization in Morocco.” Democratization 9(1):149–169.

Haggard, Stephan and Robert R. Kaufman. 2012. “Inequality and Regime Change: Demo- cratic Transitions and the Stability of Democratic Rule.” American Political Science Re- view 106(03):495–516.

Hazama, Yasushi, Ömer Faruk Gençkaya and Selma Genckaya. 2007. “Parliamentary Ques- tions in Turkey.” The Journal of Legislative Studies 13(4):539–557.

117 Healy, Andrew and Gabriel S Lenz. 2014. “Substituting the End for the Whole: Why Voters Respond Primarily to the Election-year Economy.” American Journal of Political Science 58(1):31–47.

Herb, Michael. 2002. “Emirs and Parliaments in the Gulf.” Journal of Democracy 13(4):41– 47.

Herron, Erik S and N. Boyko. 2015. “Horizontal accountability during political transition: The use of deputy requests in Ukraine, 2002-2006.” Party Politics 21(1):131–142.

Ho, Daniel E, Kosuke Imai, Gary King and Elizabeth A Stuart. 2007. “Matching as non- parametric preprocessing for reducing model dependence in parametric causal inference.” Political Analysis 15(3):199–236.

Ho, Daniel E, Kosuke Imai, Gary King, Elizabeth A Stuart et al. 2011. “MatchIt: Non- parametric preprocessing for parametric causal inference.” Journal of Statistical Software 42(8):1–28.

Hodler, Roland and Paul A Raschky. 2014. “Regional Favoritism.” The Quarterly Journal of Economics 129(2):995–1033.

Hollyer, James R, B Peter Rosendorﬀ and James Raymond Vreeland. 2011. “Democracy and Transparency.” The Journal of Politics 73(4):1191–1205.

Huber, JD. 1996. “The Vote of Conﬁdence in Parliamentary Democracies.” American Polit- ical Science Review 90(2):269–282.

Huskey, Eugene. 2010. “Elite Recruitment and State-society Relations in Technocratic Authoritarian Regimes: The Russian Case.” Communist and Post-communist Studies 43(4):363–372.

Ives, Anthony R. 2015. “For Testing the Signiﬁcance of Regression Coeﬃcients, Go Ahead and Log-transform Count Data.” Methods in Ecology and Evolution 6(7):828–835.

118 Jensen, Nathan M, Edmund Malesky and Stephen Weymouth. 2014. “Unbundling the relationship between authoritarian legislatures and political risk.” British Journal of Political Science 44(3):655–684.

Katz, Richard S. 1987. “Party government and its alternatives.” Party governments: Euro- pean and American experiences 2.

Kilani, Sa’eda and Basem Sakijha. 2002. “Wasta: The declared secret.” Amman, Jordan: Jordan Press Foundation .

Lancaster, Tony. 2000. “The incidental parameter problem since 1948.” Journal of Econo- metrics 95(2):391–413.

Lee, Don S and Paul Schuler. 2019. “Testing the “China model” of meritocratic promotions: Do democracies reward less competent ministers than autocracies?” Comparative Political Studies .

Levitsky, Steven and Lucan Way. 2002. “The rise of competitive authoritarianism.” Journal of Democracy 13(2):51–65.

Liddell, James. 2010. “Notables, clientelism, and the politics of change in Morocco.” The Journal of North African Studies 15(3):315–331.

Lust, Ellen. 2009. “Competitive Clientelism in the Middle East.” Journal of Democracy 20(3):122–135.

Lust-Okar, E. and A. Jamal. 2002. “Rulers and Rules: Reassessing the Inﬂuence of Regime Type on Electoral Law Formation.” Comparative Political Studies 35(3):337–366.

Lust-Okar, Ellen. 2005. Structuring conﬂict in the Arab world: Incumbents, opponents, and institutions. Cambridge University Press.

Lust-Okar, Ellen. 2006. “Elections under Authoritarianism: Preliminary Lessons from Jor- dan.” Democratization 13(3):456–471.

119 Madani, Mohamed, Driss Maghraoui and Saloua Zerhouni. 2012. The 2011 Moroccan Con- stitution: A Critical Analysis. International IDEA Stockholm.

Magaloni, B. 2006. “Voting for Autocracy: Hegemonic Party Survival and its Demise in Mexico.” 296.

Magaloni, B. 2008. “Credible Power-Sharing and the Longevity of Authoritarian Rule.” Comparative Political Studies 41(4-5):715–741.

Magaloni, Beatriz, Alberto Diaz-Cayeros and Federico Estévez. 2007. Clientelism and Port- folio Diversiﬁcation: a Model of electoral Investment with Applications to Mexico. In Patrons, clients, and policies: Patterns of democratic accountability and political competition, ed. H. Kitschelt and S. Wilkinson. Cambridge University Press pp. 182–205.

Maghraoui, Abdeslam. 2001. “Monarchy and Political Reform in Morocco.” Journal of Democracy 12(1):73–86.

Maghraoui, Driss. 2018. “Working under Constraints: The PJD in the Aftermath of the 2016 Elections.” James A. Baker III Institute for Public Policy of Rice University Issue Brief 5.

Malesky, Edmund and Paul Schuler. 2009. “Paint-by-numbers Democracy: The Stakes, Structure, and Results of the 2007 Vietnamese National Assembly Election.” Journal of Vietnamese Studies 4(1):1–48.

Malesky, Edmund and Paul Schuler. 2010. “Nodding or Needling: Analyzing Dele- gate Responsiveness in an Authoritarian Parliament.” American Political Science Review 104(03):482–502.

Malesky, Edmund and Paul Schuler. 2011. “The Single-party Dictator’s Dilemma: Informa- tion in Elections without Opposition.” Legislative Studies Quarterly 36(4):491–530.

120 Martin, Shane. 2011. “Parliamentary Questions, the Behaviour of Legislators, and the Func- tion of Legislatures: An Introduction.” The Journal of Legislative Studies 17(3):259–270.

Masoud, Tarek. 2014. Counting Islam: Religion, Class, and Elections in Egypt.

Mauceri, Philip. 1997. “Return of the Caudillo: Autocratic Democracy in Peru.” Third World Quarterly 18(5):899–912.

McCarty, Nolan M. 2000. “Presidential Pork: Executive Veto Power and Distributive Poli- tics.” American Political Science Review 94(1):117–129.

McDonnell, Duncan and Marco Valbruzzi. 2014. “Deﬁning and Classifying Technocrat-led and Technocratic Governments.” European Journal of Political Research 53(4):654–671.

Mulligan, Casey B, Ricard Gil and Xavier Sala-i Martin. 2004. “Do Democracies have Diﬀer- ent Public Policies than Nondemocracies?” Journal of Economic Perspectives 18(1):51–74.

Myerson, Roger B. 2008. “The Autocrat’s Credibility Problem and Foundations of the Constitutional State.” American Political Science Review 102(1):125–139.

National Democratic Institute. 2011. Final Report on the Moroccan Legislative Elections, November 25, 2011. Technical report.

Nordhaus, William D. 1975. “The Political Business Cycle.” The Review of Economic Studies 42(2):169–190.

O’Donnell, Guillermo A. 1998. “Horizontal Accountability in New Democracies.” Journal of Democracy 9(3):112–126.

O’Hara, Robert B and D Johan Kotze. 2010. “Do Not Log-transform Count Data.” Methods in Ecology and Evolution 1(2):118–122.

Ordeshook, PC and OV Shvetsova. 1994. “Ethnic Heterogeneity, District Magnitude, and the Number of Parties.” American Journal of Political Science 38(1):100–123.

121 Pellicer, Miquel and Eva Wegner. 2013. “Electoral Rules and Clientelistic Parties: A Re- gression Discontinuity Approach.” Quarterly Journal of Political Science 8(March):1–33.

Pepinsky, Thomas. 2007. “Autocracy, Elections, and Fiscal Policy: Evidence from Malaysia.” Studies in Comparative International Development 42(1-2):136–163.

Pripstein Posusney, M. 2002. “Multi-party Elections in the Arab World: Institutional Engi- neering and Oppositional Strategies.” Studies in Comparative International Development 36(4):34–62.

Raunio, Tapio. 1996. “Parliamentary Questions in the European Parliament: Representation, Information and Control.” The Journal of Legislative Studies 2(4):356–382.

Reuter, Ora John and Graeme B Robertson. 2015. “Legislatures, Cooptation, and Social Protest in Contemporary Authoritarian Regimes.” The Journal of Politics 77(1):235–248.

Rogers, Robert and Rhodri Walters. 2013. How Parliament Works 6th edition. Routledge.

Russo, Federico and Matti Wiberg. 2010. “Parliamentary Questioning in 17 European Par- liaments: Some Steps towards Comparison.” The Journal of Legislative Studies 16(2):215– 232.

Salmond, Rob. 2004. “Grabbing Governments by the Throat: Question Time and Leadership in New Zealand’s Parliamentary Opposition.” Political Science 56(2):75–90.

Sater, James N. 2003. “Morocco after the Parliamentary Elections of 2002.” Mediterranean Politics 8(1):135–142.

Sater, James N. 2009. “Parliamentary Elections and Authoritarian Rule in Morocco.” The Middle East Journal 63(3):381–400.

Sater, James N. 2012. “Reserved Seats, Patriarchy, and Patronage in Morocco.” The impact of gender quotas pp. 72–86.

122 Sater, James N. 2016. Morocco: Challenges to Tradition and Modernity. Routledge.

Schedler, Andreas. 2002. “The Menu of Manipulation.” Journal of Democracy 13(2):36–50.

Schedler, Andreas et al. 2006. Electoral Authoritarianism: The Dynamics of Unfree Compe- tition. Lynne Rienner Boulder, CO.

Schleiter, Petra. 2013. “Democracy, authoritarianism, and ministerial selection in Russia: How presidential preferences shape technocratic cabinets.” Post-Soviet Aﬀairs 29(1):31–55.

Shehata, Samer. 2008. Inside an Egyptian Parliamentary Campaign. In Political Partic- ipation in the Middle East, ed. Ellen Lust-Okar and Saloua Zerhouni. Lynne Rienner Publishers Boulder, Colorado pp. 95–120.

Stokes, Susan C, Thad Dunning and Marcelo Nazareno. 2013. Brokers, Voters, and Clien- telism: The Puzzle of Distributive Politics. Cambridge University Press.

Storm, Lise. 2007. Democratization in Morocco: The Political Elite and Struggles for Power in the Post-independence State. Routledge.

Stuart, Elizabeth A. 2010. “Matching Methods for Causal Inference: A Review and a Look Forward.” Statistical Science: a Review Journal of the Institute of Mathematical Statistics 25(1):1.

Svolik, Milan. 2013. “Incentives, Institutions, and the Challenges to Research on Authori- tarian Politics.” APSA Comparative Democratization Newsletter (June 2013) .

Svolik, Milan W. 2009. “Power Sharing and Leadership Dynamics in Authoritarian Regimes.” American Journal of Political Science 53(2):477–494.

Sweet, Catherine. 2001. “Democratization without Democracy: Political Openings and Clo- sures in Modern Morocco.” Middle East Report (218):22–25.

123 Szmolka, Inmaculada. 2010. “Party System Fragmentation in Morocco.” The Journal of North African Studies 15(1):13–37.

Thomas, Melissa A and Oumar Sissokho. 2005. “Liaison Legislature: the Role of the National Assembly in Senegal.” The Journal of Modern African Studies 43(1):97–117.

Truex, Rory. 2014. “The Returns to Oﬃce in a “Rubber Stamp” Parliament.” American Political Science Review 108(2):235–251.

Van de Walle, Nicolas. 2002. “Elections without Democracy: Africa’s Range of Regimes.” Journal of democracy 13(2):66–80.

Waltz, Susan Eileen. 1995. Human Rights and Reform: Changing the Face of North African Politics. Univ of California Press.

Wantchekon, Leonard. 2003. “Clientelism and Voting Behavior: Evidence from a Field Experiment in Benin.” World Politics 55(3):399–422.

Waterbury, John. 1970. The Commander of the Faithful: The Moroccan Political Elite – a Study in Segmented Politics. Vol. 2 Weidenfeld & Nicolson.

Willis, M. 2002a. “Political Parties in the Maghrib: Ideology and Identiﬁcation. A Suggested Typology.” The Journal of North African Studies 7(3):1–28.

Willis, Michael J. 2002b. “Political Parties in the Maghrib: The Illusion of Signiﬁcance?” The Journal of North African Studies 7(2):1–22.

Wright, Joseph. 2008. “Do Authoritarian Institutions Constrain? How Legislatures Aﬀect Economic Growth and Investment.” American Journal of Political Science 52(2):322–343.

Yom, Sean L. 2009. “Jordan: Ten More Years of Autocracy.” Journal of Democracy 20(4):151–166.

124 York, Erin. 2019. “Votes for Eﬀort: Constituency Service and Opposition Support under Autocracy.” Working Paper .

York, Erin. 2020. “Ministries Matter: Horizontal Accountability under Autocracy.” Working Paper .

Zerhouni, Saloua. 2014. ““Smartness” without Vision: The Moroccan Regime in the face of Acquiescent Elites and Weak Social Mobilization.” Stiftung Wissenschaft und Politik Comments .

125 Appendix A: Appendices to Chapter 1

A.1 Moroccan Political Parties

The Moroccan royal family has in the past encouraged party proliferation as a way of countering consolidation of support by a single opposition party (Lust-Okar and Jamal, 2002; Lust-Okar, 2005). As a result, the Moroccan political arena is highly fragmented with more than 30 registered political parties that typically compete independently in elections.1 The majority of parties are very small and achieve success only in local (municipal) elections. In the legislature elected in 2011, only eight parties won more than a handful of seats; the remaining deputies, representing four diﬀerent political parties, formed a small parliamentary faction (the Alliance of the Center) or participated along with one of the larger parties.2 For the purposes of this study, I restrict discussion to the eight largest parties. The list of parties and notes on their history and ideology are summarised in Table A.1.

1In rare cases, parties join forces on the ballot. For example, three small leftist parties – the Socialist Democratic Vanguard Party, the National Ittihadi Congress, and the United Socialist Party – united to form a joint list (the progressive Federation of the Democratic Left) to contest the 2016 national elections. Collectively, they won two seats in the 2016 parliament. 2For example, the Party of the Environment and Development, a small regional party that won two seats, joined with RNI. Its leadership then joined the PAM for the 2016 elections.

126 Type Party Name Ideology and Notes Royalist Popular Movement (MP) Founded in 1958 as a rural, Parties 2011 seats: 32 royalist party with an Amazigh 2016 seats: 27 (Berber) orientation.

National Rally of Independents (RNI) Formed in 1978 by a coalition of 2011 seats: 52 palace-friendly individual politicians. 2016 seats: 37

Constitutional Union (UC) Founded in 1983, like RNI, by a 2011 seats: 23 group of royalist ﬁgures. 2016 seats: 19

Party of Authenticity and Modernity (PAM) Founded in 2008 by Fouad Ali El 2011 seats: 47 Himma, a close friend of the king. 2016 seats: 102 Due to this connection and lack of clear ideology it is seen as being closely linked to the palace.

Opposition Istiqlal Party (PI) Founded in 1937, a historically Parties 2011 seats: 60 important opposition party with a 2016 seats: 46 center-right ideology. In recent years more closely aﬃliated with the monarchy (Fakir, 2018).

Party of Progress and Socialism (PPS) Founded in 1974 as a leftist group, 2011 seats: 18 the party later distanced itself 2016 seats: 12 from communism.

Socialist Union of Popular Forces (USFP) Formed in 1975 in a split from UNFP, 2011 seats: 39 a former socialist opposition party. 2016 seats: 20 Center-left ideology.

Party of Justice and Development (PJD) Formed as a movement in the 60s 2011 seats: 107 and as a political party prior to the 2016 seats: 125 1997 elections, the party won a plurality of seats in 2011. Islamist conservative party. Table A.1: List of Key Political Parties from 2011 Moroccan Parliament. This table provides notes on the eight largest parties, each of which acquired eighteen or more seats in the 2011 parliament. Parties are grouped by royalist/non-royalist orientation and listed in order of creation.

127 A.2 Robustness Check: Negative Binomial Regression

In this section, I conduct a robustness check of the results in Table 1.2 using a count model. Table A.2 reports output from negative binomial regression of the number of written questions on the royalist indicator as well as a range of controls.3 As in the OLS models with a transformed outcome variable, the coefficient on royalist is negative and significant to the α = 0.001 level in all models; it barely changes with the inclusion of controls for other deputy characteristics. Again, coefficients should be interpreted on a logarithmic scale. Being a royalist is therefore associated with a roughly 2.7 log unit decrease in the number of questions submitted. The coefficient is slightly larger in magnitude than in the OLS model presented in Table 1.2. In other words, royalist deputies are predicted to submit 0.07 times the number of questions submitted by opposition deputies – more than an order of magnitude fewer.

3The negative binomial distribution is preferable to Poisson in this case due to overdispersion; mean exceeds variance in all models.

128 Dependent variable: Written Questions (1) (2) (3) Royalist −2.777∗∗∗ −2.718∗∗∗ −2.703∗∗∗ (0.663) (0.695) (0.817) Governing Coalition 0.639 (0.867) Committee Leader 0.974∗ 1.255∗∗ (0.458) (0.429) National List −0.503∗∗ (0.193) Female −0.442∗ (0.212) Youth −0.843∗∗ (0.294) 2007 Parliament −0.348 −0.301 (0.246) (0.268) Civic Engagement 0.358 0.474 (0.283) (0.245) Block Bootstrapped Yes Yes Yes Observations 413 413 413 Log Likelihood −1,880.555 −1,868.673 −1,858.424 θ 0.428∗∗∗ (0.028) 0.451∗∗∗ (0.030) 0.471∗∗∗ (0.031) Akaike Inf. Crit. 3,765.110 3,749.347 3,732.849 Note: ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001 Table A.2: Predictors of Question Submission in the 2011 Parliament – Negative Binomial Speciﬁcation. Models report coeﬃcients from negative binomial regression of the number of written questions submitted on politician characteristics. The unit of analysis is the deputy. Standard errors are clustered at the party level and calculated by block bootstrap.

129 A.3 Eﬀort and Electoral Outcomes: Deputy Level Analysis

In section 1.5, I examine electoral outcomes as a function of legislative activity at the party-district level to account for Morocco’s use of closed list PR. This approach relies on the assumption that voters reward parties for their deputies’ performance in office. In this section, I test whether the same pattern found at the party level in Table 1.3 exists for individual deputies: in other words, is there a posititve relationship between constituency service and incumbency at the politician level? In Morocco, relatively few incumbents are renominated to party lists in subsequent elections. To deal with the fact that voters cannot reward high-performing politicians who are not renominated by their party, I condition on politician nomination to remove cases where voters were unable to support a high performing deputy.4 This tests the hypothesis that, given the opportunity to reward high-performing candidates with re-election, voters will do so. For this individual level analysis, the predictors of interest are the log transformed number of oral and written queries submitted by a given deputy during the 2011 term. As with other individual-level analysis, I control for other deputy characteristics. Results reported in Table A.3 again support the assertion that voters reward representatives that are more active in engaging in constituency service: deputies who submit more written queries are significantly more likely to be reelected. For individuals nominated as the head of a district list, a log unit increase in the number of written questions submitted corresponds to an eight percentage point increase in their probability of being elected to the 2016 legislature. As with the party-district-level specifications, I find evidence that oral queries have the opposite association: the coefficient for this variable is negative in all specifications, suggesting that voters are less enthusiastic about deputies when they engage in too much nationally- focused activity at the expense of district-level concerns. A log unit increase in oral queries

4This restricts the sample of deputies to include only those most preferred by the party, as well as those from the seven parties for which candidate lists were available.

130 Dependent variable: Re-election |Nominated |Head of List (1) (2) Log Written 0.066∗ 0.082∗∗ (0.026) (0.028) Log Oral −0.086∗∗ −0.085∗ (0.030) (0.035) Royalist 0.110 0.080 (0.094) (0.097) Committee Leader 0.278 0.343 (0.168) (0.191) 2007 Parliament −0.020 −0.032 (0.085) (0.091) National List −0.308∗ −0.247 (0.122) (0.177) Mean Re-election Rate 0.55 0.60 Observations 172 136 R2 0.113 0.111 Note: ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001 Table A.3: Legislative Activity and Electoral Outcomes – Deputy Level Analysis. Coeﬃ- cients represent estimates of OLS regression of legislative activity (oral and written queries submitted) on deputy reelection, subsetting to individuals who were renominated by their party (model 1) or nominated as the head of a district list (model 2). Analysis conducted at the deputy level with controls for politician attributes.

131 is associated with an eight percentage point decrease in reelection probability.

132 Appendix B: Appendices to Chapter 2

B.1 Ministry Aﬃliations during the 2011-2016 Term

In Table B.1, I report the partisan affiliations (or lack thereof) for members of the two cabinets that existed during the 2011-2016 parliamentary term. The first cabinet was formed following the November parliamentary elections and took office as of the beginning of the parliamentary term in January 2012. Following the withdrawal of the Istiqlal party from the governing coalition in mid-2013, a new coalition formed with the National Rally of Independents replacing Istiqlal. The original cabinet was dissolved and the new cabinet took office on October 10, 2013. In the new cabinet, the total number of Delegate-Ministries was increased, so some offices did not exist during the lifetime of the 2012 cabinet. Ministers selected by the regime that lack partisan affiliation are identified as technocrats.

133 Ministry Affiliation 2012-2013 Affiliation 2013-2016 Agriculture Non-partisana Non-partisan Communication Party of Justice and Development PJD (PJD)b Culture Party of Progress and Socialism PPS (PPS)c Economy and Finance Istiqlal (PI)d National Rally of Independents (RNI)e Economy (delegate-ministry) PJD PJD Employment PPS PPS Energy PI PJD Energy: Environment — MP (delegate-ministry) Energy: Water — PPS (delegate-ministry) Foreign Affairs PJD RNI General Secretary Technocrat Technocrat Health PPS PPS Higher Education PJD PJD Housing and Urbanism PPS PPS Industry PJD RNI Interior Popular Movement (MP)f Technocrat Interior (delegate-ministry) Technocrat Technocrat Islamic Affairs Technocrat Technocrat Justice PJD PJD Moroccans Abroad PI RNI National Education PI Technocrat National Education — MP (delegate-ministry) Parliamentary Relations PJD PJD Prime Minister PJD PJD Prime Minister: Civil Service MP MP (delegate-ministry) Prime Minister: Defense Technocrat Technocrat (delegate-ministry) Prime Minister: Public Affairs PJD Non-partisang (delegate-ministry) Tourism MP MP Traditional Craft PI RNI Transportation PJD PJD Women and the Family PJD PJD Youth and Sports MP MP

aThis ministry was headed throughout the 2011 term by Aziz Akhannouch, a businessman and former member of the National Rally of Independents, a royalist party. He left the party to take this appointment and rejoined it in October 2016, after the end of the term and the subsequent national elections. bIslamist-oriented opposition party; won a plurality of votes in the 2011 and 2016 legislative elections. cLeftist opposition party founded in 1974. dHistorical opposition party that today has a center-right ideology. eRoyalist party formed in 1978 by a coalition of palace-friendly individual politicians. fRoyalist party founded in 1958 with a rural, Amazigh (Berber) orientation. gThis ministry was led after 2013 by Mohamed Louafa, a former member of the Istiqlal party and previously Minister of National Education who refused to leave the government with his party in 2013 and subsequently resigned from Istiqlal.

Table B.1: Partisan aﬃliations of cabinet134 ministers during the 2011-2016 term. Full Dataset (n = 27195) Sample A (n = 24625) Sample B (n = 11140) Variable Min Max Mean SD Mean SD Mean SD Aligned 0 1 0.133 0.339 0.147 0.354 0.058 0.233 Former 0 1 0.257 0.437 0.256 0.436 0.304 0.460 Leader 0 1 0.193 0.395 0.195 0.396 0.174 0.379 National List 0 1 0.124 0.329 0.124 0.330 0.126 0.332 Georeference 0 1 0.765 0.424 0.764 0.425 0.765 0.424 Casework 0 1 0.033 0.179 0.035 0.183 0.037 0.189 Commune 0 1 0.225 0.418 0.224 0.417 0.300 0.458 Late 0 1 0.111 0.314 0.112 0.315 0.106 0.308 New Minister 0 1 0.452 0.498 0.488 0.500 1.000 0.000 First Cabinet 0 1 0.224 0.417 0.225 0.418 0.237 0.425 Royalist 0 1 0.063 0.243 0.064 0.244 0.050 0.218 Solo 0 1 0.916 0.278 0.919 0.273 0.913 0.281 Technocrat 0 1 0.218 0.413 0.207 0.405 0.457 0.498 Year 2012 2016 2013.993 1.098 2013.990 1.097 2013.982 1.109

Table B.2: Summary statistics. Table shows summary statistics for the full dataset as well as the two subsets used for DiD analysis.

B.2 Summary Statistics: Full Data and Diﬀerence in Diﬀerences Subsamples

This section presents summary statistics from the full dataset as well as for the subsets used in the difference-in-differences specifications in Tables 2.4 and B.6. Statistics are shown in Table B.2. Sample A drops only queries submitted to ministries that were held by technocrats throughout the entirety of the 2011-2016 term (see Section 2.5.1). Sample B also drops queries submitted to ministries that did not exist in the first term (three delegate-ministry positions) as well as those for which no change in minister occurred (see Section B.6). For example, a query submitted to the General Secretary would be excluded from both subsets, because that ministry was held by a technocrat in both cabinets formed. A query submitted to the Ministry of Communication would be included in Sample A but excluded from Sample B, as it was held by the PJD during both cabinet terms.

135 Dependent variable: Received Response Time to Response (Hazard) (1) (2) (3) (4) Technocrat Minister −0.236∗ −0.241∗ −0.939∗ −1.624∗∗∗ (0.113) (0.117) (0.424) (0.254)

Aligned Minister −0.150+ −0.111∗∗ −0.330 −0.324 (0.078) (0.041) (0.452) (0.229)

Ministry FE: XX Controls: XX Model: OLS OLS Cox PH Cox PH Observations 27,188 27,188 27,120 27,120 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Table B.3: Observational Analysis – OLS regression using the full dataset. I report treatment coeﬃcients on main IVs from speciﬁcations with and without controls. Models 1 and 2 are evaluated on the full dataset; models 3 and 4 include only questions submitted after the formation of the 2013 cabinet. Standard errors are clustered at the ministry and parliamentary group level and constructed using bootstrapping.

B.3 Observational Analyses

In Table B.3, I report output from OLS regression of the binary response variable on treatment indicators, using the full dataset. I estimate models on both IVs with and without controls. Controls include cabinet, ministry, and year ﬁxed eﬀects, as well as indicators for whether a query was submitted at the end of the given cabinet term (Late), whether it included a georeference or evidence of casework, and attributes of the question asker(s) (Leader, National List, and Solo). Cox models drop the 68 observations (0.25% of the total dataset) for which the recorded date of reply preceded the date the query was raised.

136 B.4 Matching: Robustness Using Exact Matching

The literature on matching using observational data has generated a number of techniques for generating a valid control set (Ho et al., 2007). In the main text, I use nearest neighbor propensity score matching to generate a set of control observations that are similar in observable attributes to those that were treated. In this section, I present results from an alternative approach that uses exact matching as a robustness check on my choice of matching technique. Exact matching is more restrictive and often requires that researchers reduce the set of matching covariates lest too many treated observations lack an exact counterpart in the control group (Stuart, 2010). I use the same set of matching covariates as in the propensity score matching approach described in section 2.4.4. In this case, because the matching characteristics used are largely binary indicators, most treated cases have a direct analog in the control category. The matched dataset preserves 99% of treated observations and more than doubles the size of the control group for the technocrat treatment (relative to propensity score matching), and preserves 98% of treated observations and expands the pool of control group observations by more than 50% for the aligned treatment.

137 Dependent variable: Received Response (1) (2) (3) (4) (5) (6) Technocrat Minister −0.235 −0.256∗∗∗ −0.294∗∗∗ (0.143) (0.017) (0.041)

Aligned Minister −0.079 −0.028 −0.026 (0.072) (0.031) (0.032)

Mean DV, non-techno: 0.734 0.734 0.734 Mean DV, non-aligned: 0.717 0.717 0.717 Ministry FE: XXXX Controls: XX Matched Sample: Techno Techno Techno Aligned Aligned Aligned IPW: XXXXXX Observations 23,865 23,865 23,865 9,050 9,050 9,050 Adjusted R2 0.044 0.214 0.301 0.006 0.150 0.360 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Table B.4: Responsiveness – Matched Data. Table presents results from OLS estimation of whether a query received a response on treatment indicators in matched datasets, with and without ministry ﬁxed eﬀects and other controls, for both technocrat (models 1-3) and partisan alignment (models 4-6) treatments. Models are estimated on datasets constructed using exact matching with inverse probability weights. Robust standard errors are clustered at the ministry-cabinet level for the technocrat treatment, and the group-ministry level for the alignment treatment.

138 Dependent variable: Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat Minister −0.787∗ −2.251∗∗∗ −1.456∗∗∗ (0.400) (0.180) (0.221)

Aligned Minister −0.282 −0.164 −0.079 (0.195) (0.250) (0.107)

Ministry FE: XXXX Controls: XX Matched Sample: Techno Techno Techno Aligned Aligned Aligned IPW: XXXXXX Observations 23,806 23,806 23,806 9,032 9,032 9,032 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Table B.5: Cox Proportional Hazards Model – Matched Data. Table presents results from estimating a Cox proportional hazards model of time-to-response on treatment indicators in matched datasets, with and without ministry ﬁxed eﬀects and other controls, for both technocrat (models 1-3) and partisan alignment (models 4-6) treatments. Models are estimated on datasets constructed using exact matching with inverse probability weights. Robust standard errors are clustered at the ministry-cabinet level for the technocrat treatment, and the group-ministry level for the alignment treatment.

139 B.5 Diﬀerence in Diﬀerences: Identifying Assumptions

In this section, I present evidence in support of the parallel trends assumption underlying the DiD analysis in Section 2.5.1. I conduct this analysis on the sample used in the main text, which I refer to as Sample A, as well as on a further restricted subset analyzed as a robustness check, which I refer to as Sample B (see Appendix B.6 for discussion of how this was constructed). Figure B.1 plots average response rates, grouped by the month in which a query was submitted and by whether or not it was submitted to a treated ministry for each of the two samples used in this analysis (see Section 2.5.1 for a description of how sample were constructed). I include averages for both the primary response rate used in analyses (‘response by cabinet enddate’) as well as a measure of whether a response was recorded at any point.1 Trendlines are included for the period before and after the cabinet reshuffle. The plots suggest that treated and untreated queries experienced highly similar response rates in the months leading up to the change in cabinet membership: trendlines are largely overlapping and follow similar trajectories, especially for the response by cabinet enddate outcome. Following the cabinet change (and treatment for queries assigned to the treated ministries), the outcomes diverge more starkly. As a secondary check, I examine trends in query submission before and after the cabinet reshuffle in order to assess whether the data generating process differed following treatment. In other words, this addresses the concern that deputies might behave differently when a ministry is led by a technocrat figurehead. Figure B.2 plots the proportion of queries submitted to treated ministries in a given month for Samples A and B, both before and after the assignment of technocrat ministers to treated ministries. The data suggest that submission rates to treated and untreated ministries did not systematically change before and after treatment: the trendline is essentially flat throughout the parliamentary term. This indicates that submission rates to the treated ministries relative to untreated ministries 1For queries submitted after October 10, 2013 - the date of the cabinet reshuffle - these outcomes are identical.

140 Figure B.1: Technocrat DiD – Parallel Trends. Figure shows response rates for queries submitted within a given month to treated and untreated ministries, before and after the cabinet reshuﬄe of 2013, indicated by points as well as a loess trendline. Vertical line indicates the date on which the new cabinet took oﬃce (when treatment occurred). Top and bottom panels depict trend data for Samples A and B, respectively.

141 Figure B.2: Technocrat DiD – Query Submission by Treatment Status. Figure shows the proportion of queries submitted to treated ministries, grouped by month, for Samples A (top panel) and B (bottom panel). Vertical line indicates the date on which the new cabinet took oﬃce (when treatment occurred). were fairly constant across time.

142 B.6 Diﬀerence-in-Diﬀerences: Sample Robustness

I discuss the difference-in-differences approach in Section 2.5.1, including sample restrictions to reduce confounding. In this section, I address a possible exclusion restriction violation associated with the fact the the treated category (ministries that went from partisan to technocrat control in 2013) experienced a change in leadership in addition to the change in minister type. To do this, I create a further restricted sample (referred to as Sample B in Figures B.1 and B.2) that includes only queries submitted to ministries that changed hands in the 2013 reshuffle. This excludes ministries that remained under the control of the same party throughout the 2011 term. This subset also excludes ministry posts that were created in the 2013 reshuffle, as they have no prior point of comparison.2 Results from estimating equation 2.1 on this sample are shown in Table B.6, with models 1-3 representing OLS estimation on whether or not a query received a response, and models 4-6 showing Cox proportional hazards estimation on time-to-response. The estimates on Treatment x Post bear out the results in the main text (Table 2.4). Coefficients are similar in size and significance, with queries submitted to treated ministries around 30 percentage points less likely to receive a response.

2The newly created ministry posts were all delegate ministries (subsidiary posts to the primary minister in a given area), and included: two delegates to the Ministry of Energy, Mining, Water, and the Environment; one delegate to the Ministry of Equipment, Transport, and Logistics; one delegate to the Ministry of National Education and Vocational Training; and one delegate to the Ministry of Industry, Trade, Investment and Digital Economy. See Appendix Table B.1.

143 Dependent variable: Received Response Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat x Post −0.242∗∗∗ −0.296∗∗∗ −0.301∗∗∗ −1.071∗∗ −1.595∗∗∗ −1.591∗∗∗ (0.069) (0.069) (0.073) (0.340) (0.262) (0.306)

Ministry FE: XXXX Controls: XX Model: OLS OLS OLS Cox PH Cox PH Cox PH Observations 11,140 11,140 11,140 11,140 11,140 11,140 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Table B.6: Difference in differences specification – technocrat ministries, restricted sample. Table presents estimated treatment effects from estimation of equation 2.1, with and without ministry fixed effects and additional controls. Models 1-3 represent OLS estimation using a binary indicator for response received; models 4-6 estimate a Cox proportional hazards model with a time-to-response outcome. Robust standard errors are clustered at the ministry level and constructed through bootstrapping.

144 B.7 Robustness: Technocrats and Non-Royalist Deputies

In this section I report results from estimating effects of the Technocrat treatment with a dataset that includes only queries submitted by non-royalist deputies. The basis for the subset is the nearest neighbor matched dataset described in section 2.4.4. Coefficients are similar in size and significance to those reported in Tables 2.2 and 2.3, with queries submitted to treated ministries around 20 percentage points less likely to receive a response.

Dependent variable: Received Response Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat Minister −0.227 −0.226∗∗∗ −0.212∗∗∗ −0.782+ −2.362∗∗∗ −1.189∗∗ (0.152) (0.009) (0.038) (0.425) (0.409) (0.461)

Ministry FE: XXXX Controls: XX Matched Sample: Nearest Nearest Nearest Nearest Nearest Nearest Model: OLS OLS OLS Cox PH Cox PH Cox PH Observations 11,339 11,339 11,339 11,322 11,322 11,322 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Table B.7: Non-Royalist Only – Matched Data. Table presents results from the Technocrat treatment for queries submitted by non-royalist deputies, with and without ministry ﬁxed eﬀects and other controls. Models 1-3 represent OLS estimation using a binary indicator for response received; models 4-6 estimate a Cox proportional hazards model with a time- to-response outcome. Datasets were constructed using nearest neighbor propensity score matching. Robust standard errors are clustered at the ministry level.

145 B.8 Non-Partisan Treatment

Historically, the majority of regime-selected ministry appointees in Morocco have been ‘apolitical technocrats’ (Willis, 2002b) with backgrounds in the civil service, engineering, or higher education and without substantive partisan linkages. This remains true for the two governing cabinets formed during the 2011 parliamentary term: six out of the eight ministries under non-partisan control for some or all of the term were led by technocrats (representing 5929 queries, or 22% of the dataset). There were two exceptions (representing 1734 queries, or 6% of the dataset). First, the Ministry of Agriculture was led throughout the 2011 term by Aziz Akhannouch, a prominent businessman who had previously headed the National Rally of Independents (RNI), a royalist-leaning party. His party was not a part of the ruling coalition formed by Abdelillah Benkirane following the 2011 elections, and Akhannouch subsequently resigned from the RNI in order to take up a post as the (non-partisan) head of this ministry. He rejoined his party as its head in October 2016, following the next legislative elections. Though Akhannouch therefore did not have an active partisan affiliation and was not participating in the cabinet as an official party representative, he had strong linkages to a royalist party both before and after his tenure. The second exception was Mohamed Louafa, a member of the non-royalist Istiqlal Party (PI) who headed the Ministry of National Education as a partisan minister during the first cabinet. When PI announced they would leave the existing governing coalition in mid 2013, Louafa refused to resign his post along with the other party representatives and as a result was expelled from the party. He did not maintain his position at the Ministry of National Education in the government that formed in late 2013, but was instead appointed as the Delegate Minister for Public Affairs. At that point, like Akhannouch, he lacked a current party affiliation and is therefore considered a non-partisan cabinet member. The main text explores the impact of technocrat appointees, using a strict set of criteria

146 Minister Category Response Rate Non-partisan Non-technocrat 0.63 Non-partisan Technocrat 0.46 Partisan (Opposition) 0.68 Partisan (Royalist) 0.67

Table B.8: Average Response Rates by Minister Type. regarding individual ministers’ background and historical affiliations. This designation is consistent with historical patterns of regime appointees in Morocco. Local notables and big men have an opportunity to compete for power via partisan electoral competition (much as Blaydes (2010) characterizes electoral competition in Egypt), acceding either to legislative seats or to ministry appointments as part of a governing coalition. The regime selections for the ministries of sovereignty, by contrast, are typically career civil servants or academics and lack partisan affiliation. The theoretical distinction between such appointees and the typical technocrat appointee described in the main text may be relatively minor, yet some ambiguity remains regarding the partisan linkages of these two ministers. Though they were clearly non-partisan appointees to the cabinet, their historical (and, in the case of Akhannouch, subsequent) linkages to party structures suggests that, at the least, they may have divided loyalties relative to deputy demands. If these two appointees feel pressure from both the regime and their former party leadership and the incentives of those two actors are at odds, we may not see the same attitude toward queries as with purely technocrat or partisan appointees. It is for this reason that I do not include them in the Technocrat treatment condition. From a descriptive standpoint, queries submitted to technocrat ministers were the least likely to receive a response (see Table B.8). Royalist and non-royalist partisan ministers responded to around 68% of queries, the two non-partisan, non-technocrat ministers responded to 63% of queries, and technocrats answered 46%. Do the predictions from Hypothesis 1 hold for a broader class of minister – all those who were appointed by the regime rather than via a partisan coalition? I define a new treatment

147 Non-Partisan Matched Data Treated Mean Control Mean Diﬀ-in-means Std. Bias Prop. Score 0.33 0.33 0.00 0.00 2012 0.04 0.04 -0.00 -0.00 2013 0.11 0.12 -0.01 -0.03 2014 0.50 0.50 0.00 0.01 2015 0.24 0.23 0.01 0.02 2016 0.11 0.11 -0.00 -0.01 Late 0.07 0.08 -0.00 -0.01 First Cabinet 0.07 0.07 0.00 0.00 Georeference 0.80 0.79 0.01 0.02 Casework 0.04 0.03 0.00 0.02 Solo 0.90 0.90 -0.00 -0.01 Leader 0.20 0.20 -0.00 -0.00 Former 0.27 0.26 0.02 0.04 National List 0.12 0.13 -0.00 -0.01 Group: MP 0.03 0.03 -0.00 -0.01 Group: PAM 0.01 0.01 -0.00 -0.01 Group: PI 0.06 0.06 0.00 0.00 Group: PJD 0.44 0.45 -0.00 -0.00 Group: PPS 0.01 0.01 0.00 0.00 Group: RNI 0.01 0.01 0.00 0.00 Group: UC 0.00 0.00 0.00 0.00 Group: USFP 0.42 0.42 0.00 0.00

Table B.9: Summary Statistics – Nearest Neighbor Matched Data – Non-Partisan Treatment. Table presents summary statistics across a range of covariates for treatment and control groups generated using 1:1 nearest neighbor matching on treated observations. Data are generated based on non-partisan treatment (n = 15326, with exactly 7663 observations in each category). Diﬀerence-in-means as well as standard bias estimates are included to demonstrate comparability between categories. condition, Non-partisan, coded 1 for all queries submitted to a minister without a current partisan aﬃliation and 0 for all other ministers. I then follow the same procedures outlined in section 2.4.4 to generate a matched dataset using this more general treatment condition. The comparison cases are again drawn from the pool of queries submitted to partisan ministers. Including the non-partisan, non-technocrat ministries increases the number of treated queries to 7663. As in the main text, the propensity score matching generates a dataset that is extremely comparable across treatment and control conditions, with an average absolute standard bias of 0.01 (Table B.9).

148 Dependent variable: Received Response Time to Response (Hazard) (1) (2) (3) (4) (5) (6) Technocrat Minister −0.188 −0.224∗∗∗ −0.166∗∗ −0.628+ −2.190∗∗∗ −1.093∗∗∗ (0.121) (0.035) (0.064) (0.355) (0.180) (0.288)

Ministry FE: XXXX Controls: XX Matched Sample: Nearest Nearest Nearest Nearest Nearest Nearest Model: OLS OLS OLS Cox PH Cox PH Cox PH Observations 15,326 15,326 15,326 15,303 15,303 15,303 Note: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Table B.10: Non-Partisan – Matched Data. Table presents results from the Non-partisan treatment, with and without ministry ﬁxed eﬀects and other controls. Models 1-3 represent OLS estimation using a binary indicator for response received; models 4-6 estimate a Cox proportional hazards model with a time-to-response outcome. Datasets were constructed using nearest neighbor propensity score matching. Robust standard errors are clustered at the ministry level.

I report regression results from estimation of the responsiveness outcomes on the Non- Partisan treatment in Table B.10. Results are similar to those from the Technocrat treatment – estimated effects are negative in all specification and similar in levels of significance, though coefficients are somewhat smaller in magnitude. In other words, the predictions from Hy- pothesis 1 hold for a more generalized concept of non-partisan appointees.

149 Appendix C: Appendices to Chapter 3

C.1 Proofs

C.1.1 Minimum Winning Coalition

The payoﬀ for a member of the ruling coalition is a function of their recognition probability and the surplus received when they are recognized to make a proposal, as well as their probability of being included in a proposal η in the event they are not recognized:

n + 1 v1 = 1/q(1 − − 1 c ) + (1 − 1/q)(η ∗ c ) RC 2 ¬1 i i

It is clear that this utility is decreasing in q: the payoﬀ from being recognized is strictly greater than the payoﬀ when not recognized. So the ruling coalition proposed will be no

∗ n+1 larger than necessary to secure passage. Deﬁne q ≡ 2 as this minimum winning coalition. This result generalizes to all variants of the game – expected payoﬀs for members of the ruling coalition are always of the form

1/q(¯xi) + (1 − 1/q)(¯x¬i)

where x¯i > x¯¬i because it contains the surplus for the recognized player. Expected payoﬀs for members of the ruling coalition are therefore declining in q and a minimum ruling coalition of q∗ thus represents a best strategy in organizational periods.

150 C.1.2 Proposition 1

Final period payoﬀs are diﬀerentiated based on the type of period and whether or not a player is in the ruling coalition. From equations 3.3, 3.1, and 3.2 in the main text, it is clear that for q < n,

1 1 1 vC > vO > vN (C.1)

From this, it is clear that if the preceding period is organizational, the recognized player

1 will include himself in any coalition proposed. Additionally, because vC is decreasing in q, he will seek to minimize the size of his coalition while ensuring it will pass. As a result, he will

n−1 propose a coalition including himself and 2 other players. Because only one period remains, the autocrat’s ability to reorganize does not impact play, and he is treated symmetrically by other players. The probability of being included in a given coalition for any player is ri + 1/2(1 − ri). As a result, expected payoffs in this scenario are simply the organizational period payoff plus the expected payoff from a final policy period for each player type:

1 rˆ + 1! rˆ + 1! vˆ2 = + v1 + 1 − v1 O n 2 C 2 N nrˆ + 4n + 3 = (C.2) 2n2 + 2n 1 n − rˆ ! n − rˆ ! v˙ 2 = + v1 + 1 − v1 O n 2(n − 1) C 2(n − 1) N nrˆ − 4n2 + 3 = (C.3) 2n − 2n3

If the second-to-last period is a policy period, the player recognized to propose a policy

1 n−1 will offer the minimum required to ensure passage: n to 2 other players, and keep the surplus. Payoffs depend on recognition probability (1/q for members of the ruling coalition, and 0 for all others). As noted in Section 3.4.1, the one-shot payoff for all policy periods is

151 thus equivalent to final period payoffs. From C.1, we can see that the autocrat will reorganize if and only if he is outside the ruling coalition, as he seeks to maximize his final period payoff. Payoffs for each player are thus a function of a) whether or not they are in the ruling coalition and b) whether or not the autocrat is in the ruling coalition. When the autocrat is in the ruling coalition, he will not reorganize and payoffs are as follows:

3n + 1 vˆ2 =v ˙ 2 = + v1 C C∧ 2n2 + 2n C 3n + 1 = (C.4) n2 + n 1 v˙ 2 = + v1 N∧ 2n N 1 = (C.5) n

When the autocrat is outside the ruling coalition and reorganizes, payoﬀs are instead:

3n + 1 v˙ 2 = + v1 C· 2n2 + 2n O 5n + 3 = (C.6) 2n2 + 2n 1 vˆ2 =v ˙ 2 = + v1 N N· 2n O 3 = (C.7) 2n

This result makes it possible to solve for behavior in a preceding organizational period.

2 2 n−1 Note that vˆC > vˆN . Hence, if the autocrat is recognized, he will include himself and 2 2 2 other players in his proposed coalition. Additionally, v˙C∧ > v˙C·. Non-autocrat players thus also prefer to include the autocrat in any proposed coalition. If the autocrat is certain to be in the ruling coalition, the probability that remaining players are included falls to 1/2, given q∗.

152 Expected payoﬀs are thus as follows:

1 vˆ3 = +v ˆ2 O n C 4n + 2 = (C.8) n2 + n 1 1 1 v˙ 3 = + v˙ 2 + v˙ 2 O n 2 C∧ 2 N∧ 3n + 2 = (C.9) n2 + n

This behavior extends to a game with more than three periods. Following from Lemma 1, a player’s single period payoff depends only on whether or not they are a member of the ruling coalition, and these payoffs are equivalent to uC and uN , respectively. Because uC represents the maximum possible single period payoff, the autocrat will reorganize if and only if he is outside the ruling coalition. Likewise, we can see that other members of the ruling coalition benefit from stability. For non-autocrat players, a reorganization leads to a one period deviation to an organizational payoff and possible exclusion from the subsequent coalition formed. Regardless of which player is recognized in an organizational period t > 2, she will therefore include the autocrat in her coalition proposal. Note also that the ex interim payoff for a member of the ruling coalition is decreasing in q:

(t − 1)(n + q) vT −1 = C∧ 2nq ∂ t − 1 vT −1 = − ∂q C∧ 2q2

We therefore expect the recognized player to propose a coalition of size q∗ including the autocrat.

153 For a T period game, the expected payoﬀs are thus:

1 3nT − n + T + 1 vˆT = + (T − 1)ˆv1 = (C.10) O n C 2n2 + 2n 1 1 1 2nT + T + 1 v˙ T = + (T − 1)( v˙ 1 + v˙ 1 ) = (C.11) O n 2 C 2 N 2n2 + 2n

C.1.3 Remark 1

In a fully symmetric T period game, all players ex ante continuation value would simply

T be n . The autocrat’s additional beneﬁt over other players is thus:

3nT − n + T + 1 T (n − 1)(T − 1) − = (C.12) 2n2 + 2n n 2n(n + 1)

This surplus is increasing in T :

∂ (n − 1)(T − 1)! n − 1 = ∂T 2n(n + 1) 2n(n + 1)

This is positive under existing parameter conditions. The surplus is decreasing in n:

∂ (n − 1)(T − 1)! (n2 − 2n − 1)(T − 1) = − ∂n 2n(n + 1) 2n2(n + 1)2

This is negative under existing parameter conditions.

C.1.4 Proposition 2

I ﬁrst solve for the autocrat’s expected single period payoﬀ as a member of the ruling coalition in a period t < T . Assume that in any organizational period t > 2, any recognized player will include the autocrat in his coalition proposal of size q∗. Assume also that in policy periods t > 3, other players will always include the autocrat in proposals in order to preserve their spot in the

154 ruling coalition. If this is the case, then in a policy period t > 3, the autocrat will support any proposal that offers him at least the difference between two continuation payoffs and

t∗ 2 t−1 his single period expected payoff from the following period: xˆC = n − uˆC . A recognized player always seeks to maximize their own payoff, so they will offer the autocrat his minimum acceptable policy. The autocrat’s payoff in any given policy period is then a combination of the payoff if he is recognized plus his expected payoff when another player is recognized:

1 n − 1 1 2 uˆt = (1 − ) + (1 − )( − uˆt−1) (C.13) C q∗ 2n q∗ n C

Policy outcomes in the last three periods of the game can be solved individually. The ﬁnal period of this game t = 1 is as in other variants: the autocrat’s authority is irrelevant for both policy periods and organizational periods. In the second-to-last period, an autocrat

n+q∗ 3 1 in the ruling coalition must be offered at least 2/n − 2nq∗ = 2n − 2q∗ . In policy period t = 3, the autocrat faces the possibility that he would be excluded from the next ruling coalition to form1 and must therefore take into account his expected payoff for the last two periods. His probability of being included in a coalition proposal in t = 2 is thus his recognition probability plus the likelihood another player includes him in their proposal: rˆ∗ =r ˆ + 1/2(1 − rˆ). An autocrat in the ruling coalition in t = 3 must be offered at least:

2 n2rˆ + nrˆ + n + 5 +r ˆ∗(v1 ) + (1 − rˆ∗)(v1 ) − vˆ2 = n C N C 2n(n + 1)2

We can use this information to solve equation C.13 for any period t ≥ 3 and characterize the autocrat’s payoﬀ:

1If the second-to-last period t = 2 is organizational, there is no special incentive to include the autocrat in a ruling coalition proposal (with only one remaining period, he cannot dissolve a successful coalition before the game ends).

155 t t t t t t 3 3 1−n 3 1−n 2 2 1−n 2 1−n 1−n 1−n 3n − n rˆ n+1 + n n+1 − 7n − n rˆ n+1 + 2n n+1 + 5n − 2n n+1 + n+1 − 1 uˆt |t ≥ 3 = C 2(n − 1)2n2 (C.14)

We can see that this payoﬀ is greater than the default payoﬀ for all periods:

t (n3 − 3n2 + 3n − 1) + 1−n (n3 − n3rˆ + 2n2 − n2rˆ − 2n + 1) uˆt − 1/n = n+1 C 2(n − 1)2n2

1−n The ﬁrst polynomial is always positive under existing parameter conditions. | n+1 | < 1, so the magnitude of the second polynomial is decreasing in t. At both t = 3 and t =

t 4,uˆC − 1/n > 0, and this will hold for higher t:

(n − 1)3n (n2rˆ + nrˆ + n + 5) uˆ3 − 1/n = C (n + 1)3 (n − 1)3n (n3(2 − rˆ) + 5n2 + n(ˆr + 2) − 7) uˆ4 − 1/n = C (n + 1)4

We can then characterize the autocrat’s ex ante expected utility for a T period game as the sum of his continuation payoﬀ in an initial organizing period and his expected payoﬀ for all subsequent policy periods:

T −1 T 1 X t vˆ = + uˆC n 1 h = n5(6T − 3 +r ˆ) + n4(−8T + 6 − 3ˆr) + n3(−4T − 1 + 3ˆr) + n2(8T − 3 − rˆ) − 2nT + 1+ # 1 − nT n5(ˆr − 1) + n4(3ˆr − 4) + n3(3ˆr − 3) + n2(ˆr + 1) − 1 ∗ [4n3(n − 1)2(n + 1)]−1 n + 1

156 Ex ante utilities are symmetric across the other players:

T − vˆT v˙ T = n − 1 h = 4n6T − n5(10T − 3 +r ˆ) + n4(4T − 6 + 3ˆr) + n3(8T + 1 − 3ˆr) + n2(−8T + 3 +r ˆ) + 2nT − 1+ # 1 − nT h i−1 n5(1 − rˆ) + n4(4 − 3ˆr) + n3(3 − 3ˆr) + n2(−1 − rˆ) + 1 ∗ 4n3(n − 1)3(n + 1) n + 1 (C.15)

Next, we consider the expected interim payoﬀs for other players under this scenario. Their continuation payoﬀs remain the same, but the assumption that the autocrat is included in every policy proposal means that their own probability of inclusion is slightly reduced. For a player outside the ruling coalition with no hope of recognition, this probability is now

q∗−2 n−2 . It is straightforward to calculate the interim continuation payoﬀ for a player outside the ruling coalition over policy periods t ∈ {1...T − 1}:

−n2 + 2n + 1 ! 1 v˙ T −1 = (T − 2) + N −2n3 + 2n2 + 4n 2n n2(T − 1) + n(3 − 2T ) − T = (C.16) 2n(n − 2)(n + 1)

The autocrat’s interim continuation payoﬀ across policy periods was derived above. The non-autocrat members of the ruling coalition split the remainder. The interim continuation

157 payoﬀ for these players is as follows:

1 v˙ T −1 = (T − 1 − vˆT −1 − (n − q∗)v ˙ T −1) C q∗ − 1 C N 1 − nT ! =[3n7(T − 1) − n6 13T − 13 +r ˆ + (ˆr − 1) + n + 1 1 − nT ! 1 − nT ! n5 16T − 16 + 5ˆr + (2 − rˆ) + n4 4T − 5 − 9ˆr − (3ˆr − 5) + n + 1 n + 1 1 − nT ! 1 − nT ! n3 −23T + 24 + 7ˆr + (5ˆr − 7) + n2 17T − 14 − 2ˆr + (2ˆr + 2) + n + 1 n + 1 1 − nT ! 1 − nT n 4T − 1 + − 2 + 2] ∗ [2n3(n − 1)3(n + 1)(n − 2)]−1 (C.17) n + 1 n + 1

It is a dominant strategy for a player to include the autocrat in a policy proposal for periods t > 1, conditional on him being a member of the ruling coalition. To see this, note ﬁrst that the cost for other members of the ruling coalition to include the autocrat in a policy

t∗ proposal is strictly less than the cost to include other players: his required payoﬀ x1 |x¯¬1 is less than that of other players in policy periods t > 1. The autocrat’s minimum payoﬀs for periods t > 1 are as follows:

n2 + 2n + 5 x2∗|x¯ = 1 ¬1 2n(n + 1)2 n2rˆ + nrˆ + n + 5 x3∗|x¯ = 1 ¬1 2n(n + 1)2 2 xt∗|x¯ , t > 3 = − uˆt−1 1 ¬1 n C

2∗ 3∗ 1 Both x1 and x1 are obviously less than the default payoﬀ n within the existing parameter t 1 t∗ 1 conditions. For periods t < T , uˆC > n as shown above, so x1 |x¯¬1 < n . The cost diﬀerential creates an incentive to buy the autocrat into any policy proposal. This is in addition to the fact that players’ expected utility from remaining in the ruling

t coalition in future periods vC exceeds their expected utility from a scenario in which the

158 t coalition is dissolved and they may be excluded from future coalitions vO. To see this, note that the non-autocrat ex interim payoﬀs are strictly greater for members of the ruling coalition than for those outside it. A one period deviation from including the autocrat in policy proposals would put the recognized player at risk of ending up outside the next ruling coalition to form. The below inequality holds under existing parameter constraints.

t t t t−1 t−1 v˙C − v˙O =v ˙C − [1/n + 1/2v ˙C + 1/2v ˙N ] < 0

As a result, if the autocrat is a member of the ruling coalition, any recognized player will include him in his or her policy proposal with certainty in any period before the last one. This ensures that the existing ruling coalition will be a stable one. The same logic explains why a player recognized in an organizational period will include the autocrat in her coalition proposal: their payoﬀ as a member of a stable ruling coalition is strictly greater than their payoﬀ as a member of a one-period ruling coalition without the autocrat followed by a subsequent reorganization and possible exclusion.

C.1.5 Remark 2

From proposition 1, the autocrat’s ex ante continuation payoﬀ under the basic game (decoupled decision-making) is:

3nT − n + T + 1 vˆT |d = C 2n(n + 1)

From proposition 2, the autocrat’s ex ante continuation payoﬀ under coupled decision-

159 making is:

" 1 − nT ! 1 − nT ! vˆT |c = n5 6T − 3 +r ˆ + (ˆr − 1) + n4 −8T + 6 − 3ˆr + (3ˆr − 4) + n + 1 n + 1 1 − nT ! 1 − nT ! n3 −4T − 1 + 3ˆr + (3ˆr − 3) + n2 8T − 3 − rˆ + (ˆr + 1) − n + 1 n + 1 1 − nT # 2nT + 1 − ∗ [4n3(n − 1)2(n + 1)]−1 n + 1

The diﬀerence in expected payoﬀ between the two variants is thus:

" 1 − nT ! 1 − nT ! vˆT |d − vˆT |c = −n5(ˆr − 1) + 1 + n4 (4 − 3ˆr) + 3ˆr − 2T + C C n + 1 n + 1 1 − nT ! 1 − nT ! n3 (3 − 3ˆr) − 3ˆr + 6T − 5 − n2 (1 +r ˆ) − rˆ + 6T − 5 + n + 1 n + 1 # 1 − nT h i−1 + 2nT − 1 ∗ 4n3(n − 1)2(n + 1) (C.18) n + 1

This diﬀerence is decreasing in rˆ. The following is negative for all n ≥ 3 and T ≥ 4:

! ! ! ! ∂(ˆvT |d − vˆT |c) 1 − nT 1 − nT 1 − nT 1 − nT C C = −n2 n3 + 1 + 3n2 − 1 + + 3n + 1 − 1 ∂rˆ n + 1 n + 1 n + 1 n + 1

T T And vˆC |d − vˆC |c = 0 where

T 5 4 3 2 1−n 5 4 3 2 n − 2n T + n (6T − 5) + n (5 − 6T ) + 2nT − 1 + n+1 (n + 4n + 3n − n + 1) rˆ > T 2 3 2 1−n 3 2 n n − 3n + 3n − 1 + n+1 (n + 3n + 3n + 1) (C.19)

Where T is odd, this cutoﬀ always falls outside (below) the existing range of rˆ and thus

T T vˆC |d < vˆC |c. Where T is even, it is a binding condition such that for suﬃciently low rˆ,

T T vˆC |d > vˆC |c.

160 C.1.6 Remark 3

From the proof of proposition 2, we know the ex interim utility for non-autocrat players are as follows:

1 − nT ! v˙ T −1|c =[3n7(T − 1) − n6 13T − 13 +r ˆ + (ˆr − 1) + C n + 1 1 − nT ! 1 − nT ! n5 16T − 16 + 5ˆr + (2 − rˆ) + n4 4T − 5 − 9ˆr − (3ˆr − 5) + n + 1 n + 1 1 − nT ! 1 − nT ! n3 −23T + 24 + 7ˆr + (5ˆr − 7) + n2 17T − 14 − 2ˆr + (2ˆr + 2) + n + 1 n + 1 1 − nT ! 1 − nT n 4T − 1 + − 2 + 2] ∗ [2n3(n − 1)3(n + 1)(n − 2)]−1 n + 1 n + 1 (C.20) n2(T − 1) + n(3 − 2T ) − T v˙ T −1|c = (C.21) N 2n(n − 2)(n + 1)

Recall from Section 3.4.1 that with decoupled decision-making, the expected continuation payoﬀ for any member of a stable ruling coalition is simply the expected single period utility times the number of periods remaining. Under decoupled decision-making, the ex interim payoﬀs are thus the same for autocrat and non-autocrat players and are as follows:

(T − 1)(3n + 1) v˙ T −1|d = (T − 1)u = (C.22) C C 2n(n + 1) T − 1 v˙ T −1|d = (T − 1)u = (C.23) N N 2n

For members of the ruling coalition, the diﬀerence in expected payoﬀs between the two

161 games is as follows:

! ! 1 − nT 1 − nT v˙ T −1|d − v˙ T −1|c =[n6 (ˆr − 1) +r ˆ − T + 1 + n5 (ˆr − 2) − 5ˆr − 3T + 3 + N N n + 1 n + 1 ! ! 1 − nT 1 − nT n4 (5 − 3ˆr) + 9ˆr − 16T + 17 + n3 (7 − 5ˆr) − 7ˆr + 22T − 23 + n + 1 n + 1 ! ! 1 − nT 1 − nT n2 (−2 − 2ˆr) + 2ˆr − 15T + 12 + n − + 4T + 1 + n + 1 n + 1 ! 1 − nT 2 − 1 ] ∗ [2(n − 2)(n − 1)3n3(n + 1)]−1 (C.24) n + 1

This is negative where

" ! ! ! 1 − nT 1 − nT 1 − nT rˆ < n6 T − 1 + + 2n5 −3T + 3 + + n4 16T − 17 − 5 + n + 1 n + 1 n + 1 ! ! ! 1 − nT 1 − nT 1 − nT n3 −22T + 23 − 7 + n2 2 + 15T − 12 + n − 4T − 1 − n + 1 n + 1 n + 1 !# " ! ! 1 − nT 1 − nT 1 − nT 1 − nT 2 − 1 ∗ n6 + n6 + n5 − 5 − 3n4 − 3 − n + 1 n + 1 n + 1 n + 1 ! !#−1 1 − nT 1 − nT n3 5 + 7 − 2n2 − 1 (C.25) n + 1 n + 1

1 This condition falls outside the existing parameter constraint n ≤ rˆ ≤ 1, so the inequality holds for all possible rˆ. For players outside the ruling coalition, the diﬀerence in expected payoﬀs is:

(n − 1)(T − 2) v˙ T −1|d − v˙ T −1|c = (C.26) N N 2n(n − 2)(n + 1)

This is always positive under existing parameter constraints, so the expected interim utility is higher under decoupled decision-making.

162 C.2 Empirical Speciﬁcations

In this section I elaborate on the suggestive empirics shown in Figure 3.3 and conduct additional statistical tests of the relationship between legislative dissolution power and military spending in military regimes. I use cross-national panel data assembled from the Varieties of Democracy (V-Dem) project, the Database of Political Institutions, and the World Bank. The outcome measure of interest is annual military expenditures relative to total government spending.2 This should closely proxy the metric of interest from the preceding model. As a secondary measure, I use annual military expenditures as a proportion of national GDP; this indicator is available across a wider set of country-years. The two measures are closely correlated, and I use both in the speciﬁcations that follow. The independent variable of interest is whether or not the head of state has the discretionary ability to dissolve the legislature, which I obtain from V-Dem data. This authority is present in 63% of country-years for which the head of state is a military oﬃcer (see Table C.1).

Statistic N Mean St. Dev. Min Pctl(25) Pctl(75) Max Year 1,290 1,992.585 11.877 1975 1982 2002 2017 Log_GDP 986 22.860 1.732 18.429 21.429 24.153 28.351 Dissolution_Power 1,290 0.634 0.482 0 0 1 1 Mil_Gov 454 11.048 7.258 0.390 6.037 14.175 47.755 Mil_GDP 1,034 3.044 2.666 0.00003 1.426 3.819 28.948

Table C.1: Data on Dissolution Power and Military Expenditures. Table reports summary statistics for a selection of covariates for the empirical analysis reported in Table C.2.

I report results from OLS regression of military expenditures on dissolution power in speciﬁcations with and without controls in Table C.2. Standard errors are clustered at the country level for all speciﬁcations. The results suggest that, as expected, military expenditures are higher in settings where

2This measure is tracked on an annual basis by the World Bank for countries for which total government spending is available.

163 Dependent variable: Military/Government Military/GDP (1) (2) (3) (4) Dissolution Power 3.136∗ 3.274∗∗ 0.942∗∗ 0.811∗∗ (1.744) (1.582) (0.467) (0.342)

Controls: XX Observations 454 448 1,034 986 R2 0.046 0.116 0.029 0.096 Adjusted R2 0.043 0.093 0.028 0.085 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Table C.2: Military expenditures as a function of head of state’s power of dissolution. Table reports coefficients from OLS regression with and without controls for log GDP, head of state’s path to office, and decade fixed effects. Standard errors are clustered at the country level. the military-linked head of state has dissolution power over the legislatures. This is associated with about a 3 percentage point increase in spending as a percentage of total budget and a 0.8 percentage point increase in spending as a percentage of national GDP.

164