SPEAKING OF HOME: LOCAL REPRESENTATION AND REELECTION IN AFRICA By Nikolaos Frantzeskakis A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Political Science—Doctor of Philosophy 2023 ABSTRACT This dissertation examines how parliamentary representation impacts the reelection prospects of MPs in Africa, and how the levels of local inter-party competition affect MPs’ focus on representa- tion. These are crucial questions in African states, where trust in parliaments is very low, and MP reelection rates are often about 30%. These conditions endanger the prospects of democratic con- solidation and reduce the legislative body’s ability to perform its constitutional functions, including lawmaking, checking the executive, and preventing democratic backsliding. Conventional wisdom highlights clientelism and constituency service as the primary linkages between voters and politicians in the region. This state of the literature denotes the primacy of local over national politics in regard to vote choice. I concur with this notion, but my central claim is that the provision of local representation is also key in understanding voting behavior in the region. I argue that parliamentary representation is an important and overlooked determinant of future support at the polls in Africa. In addition, I pose that the decision to prioritize representation depends on whether MPs rely on the party label for reelection. To explore these claims, I developed a unique dataset of parliamentary speeches made in Ghana, Botswana, Malawi, Zambia, Zimbabwe, and Uganda. Statistical analysis of almost 960,000 speeches by 3,800 MPs offers sound support for both claims. In addition, using original exper- imental data from Malawi, I demonstrate that the provision of local representation in parliament affects vote choice. Finally, I provide evidence that citizens are aware of their MP’s level of effort in representing the needs of the constituents in parliament. Copyright by NIKOLAOS FRANTZESKAKIS 2023 To my wife, Marianna iv ACKNOWLEDGEMENTS Looking back at the time I was beginning my adventure in graduate school, it was impossible to fathom how many people would contribute to any success that I had during this time. To begin with, it would not have been possible to complete this project without the steadfast and enthusiastic support of my dissertation committee. I can only hope that I will be able to pay this forward. I am especially grateful to Michael Wahman, who co-chaired my dissertation committee at Michigan State and also chaired my Master’s committee at the University of Missouri. It is hard to overstate his influence on my development as a scholar. Michael devoted more time than any advisee could have hoped for to mentor me, to help me develop my research, and, overall, to provide support in any way, shape, or form. He encouraged me to ask big questions and supported me in pursuing an ambitious and challenging project. I am extremely grateful to Michael, and I feel lucky to have been his student. I also owe much to the mentorship of Jeffrey Conroy-Krutz, who co-chaired my dissertation committee alongside Michael. Throughout my time at MSU, Jeff provided a stabilizing presence for which I am thankful. He always encouraged me to focus on the finer details of my arguments and exposed their blind spots. He also put a lot of effort into helping me develop a cleaner and more concise writing style. My research is much improved for his advice. I would also not have been able to produce this work without Shahryar Minhas and Eric Chang. To Shahryar, I owe most of my methodological training. He always challenged me to do methods the right way as opposed to the easy or even the conventional way. His advice was instrumental in the development of the dataset and of the research design aspects of the project. Eric helped me develop a greater sense of perspective in the way I think about complex theoretical concepts, about abstracting them and identifying their constituent parts. His influence is evident in the development of the theory presented in this project. Beyond my committee, I want to thank Sarah Reckhow, the Director of Graduate Studies, and Sarah Krause, our Graduate Program’s Assistant. They have been going above and beyond to help us navigate administrative hurdles related to issues like coursework, travel, research funding, v fieldwork, and even the pandemic as efficiently as possible. My time at MSU would have been much more stressful without their efforts. In addition, this research benefited from generous financial support from Michigan State University. I am extremely thankful to the Malawians who gave their time to be interviewed in relation to this project. Their experiences and insight were crucial in developing the ideas presented in this project. In addition, the sincere interest that they displayed in my research boosted my confidence in the project and helped me finish it. I am indebted to Laston Segula, the Assistant Editor of the Hansard at the Parliament of Malawi, for his assistance with issues related to data collection during my stay in Lilongwe. I am also grateful to Felix Chauluka. While I was in Malawi, he provided me with impeccable research assistance. In addition, he was a gracious host and companion. He made the trip that much more rewarding. I also thank my US-based research assistants without whom it would have been impossible to collect and code the data on parliamentary speeches used in this dissertation: Layla Brooks, Cooper Burton, Isaac Cinzori, Anthony Luongo, Benedict Nsenga- Rugendo, and Javohn Dyer-Smith. I am also thankful to Merete Bech Seeberg and to George Ofosu who gracefully agreed to include my conjoint experiment in the survey that they fielded in Malawi over the Spring of 2022. This project’s research design is much stronger because of it. In writing the dissertation, I received helpful feedback at various venues from several professors, colleagues, and friends. I am thankful for their input to Kelechi Amakoh, Ana Bracic, Vladimir Chlouba, Justine Davis, Erica Frantz, Andrew Halterman, Kate Knapp, Carolyn Logan, Scott de Marchi, Yonattan Morse, Michael Olson, Anne Pitcher, Victor Rateng, Ani Sarkissian, Tosin Salau, Alejandra López Villegas, and Benjamin Yoel. I am also grateful to Laron Williams who, during my time at the University of Missouri, completely changed the way I perceived quantitative methods and helped me set out on a path I am still walking today. Special thanks to my strong support team within academia: my companion through graduate school, Ha-Eun Choi, my accountability partner, Shayla Olson, my mentor, Brandon Beomseob Park, my close friend and collaborator, Yuko Sato, and the ever-curious about all things political, Max Welch. I am grateful to my family and friends whose support and encouragement helped me get into vi graduate school and then kept me going. I thank my father, Manolis, and my mother, Ioanna, who always fully supported me in pursuing my dreams no matter how far-fetched they seemed to be and were next to me every step of the way. They bred in me the confidence to make big swings and the will to take responsibility for the outcome. For this, I am forever grateful. I thank my brothers, Lefteris, and Argiris, for always being able to lighten the mood. Our group chat has been a source of joy since I left Greece. Also, I want to thank my circle of friends, both in Greece and in the US, for always being present and supportive, and for making these years in graduate school enjoyable and worthwhile. Finally, I am grateful to my wife, Marianna, to whom I dedicate this dissertation. Marianna joined me on this adventure from the very beginning. She has been my best friend and my closest confidante, always there to support me when things looked bad, and ready to celebrate every little success. My time in graduate school would have been much more lonely and joyless if not for her presence. vii TABLE OF CONTENTS LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii CHAPTER 1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Scope Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.2 Research Design Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.3 Overview of the Manuscript . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 CHAPTER 2 THE DYNAMICS OF LOCAL REPRESENTATION PROVISION . . . . 12 2.1 Conceptualizing Representation . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.2 Voter Demand for Local Representation in Parliament . . . . . . . . . . . . . . 16 2.3 Parliamentary Speech-Making as a Form of Representation . . . . . . . . . . . 22 2.4 Choosing Between Principals . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 CHAPTER 3 CASE SELECTION AND COUNTRY PROFILES . . . . . . . . . . . . 32 3.1 Case Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.2 Country Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 CHAPTER 4 LOCAL REPRESENTATION AND REELECTION . . . . . . . . . . . 46 4.1 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 4.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 CHAPTER 5 DEMAND FOR LOCAL REPRESENTATION . . . . . . . . . . . . . . 69 5.1 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 5.3 Do Voters Know if their MP Provides Local Representation? . . . . . . . . . . 81 5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 CHAPTER 6 SERVING AT WHOSE PLEASURE? PARTY DOMINANCE AND LOCAL REPRESENTATION IN AFRICAN LEGISLATURES . . . . . 86 6.1 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 6.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 CHAPTER 7 CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 7.1 Summary of Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 7.2 Academic and Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . 103 7.3 Questions for Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . 104 7.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 viii APPENDIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 ix LIST OF TABLES Table 3.1: Aggregate Information for countries in sample . . . . . . . . . . . . . . . . . 33 Table 4.1: Dataset summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Table 4.2: Descriptive Statistics for Main Variables . . . . . . . . . . . . . . . . . . . . 56 Table 4.3: Descriptive Statistics for Control Variables . . . . . . . . . . . . . . . . . . . 58 Table 4.4: Constituency Representation and Reelection . . . . . . . . . . . . . . . . . . 60 Table 4.5: Predicted Probability of Reelection . . . . . . . . . . . . . . . . . . . . . . . 63 Table 4.6: Constituency Representation and Reelection with Country Dropout . . . . . . 65 Table 5.1: List of Surveyed Constituencies . . . . . . . . . . . . . . . . . . . . . . . . . 71 Table 5.2: Attributes and Levels for Conjoint Survey Experiment . . . . . . . . . . . . . 73 Table 5.3: Example of Candidate Bios . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Table 5.4: Voting Task Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Table 5.5: Descriptive Statistics for Variables Used to Test Voter Knowledge . . . . . . . 83 Table 5.6: Effect of Parliamentary Representation on Respondents’ Estimations . . . . . 84 Table 6.1: Descriptive Statistics for all Variables . . . . . . . . . . . . . . . . . . . . . . 91 Table 6.2: The Effect of Party Dominance on Localist Speeches . . . . . . . . . . . . . . 94 Table 6.3: Predicted Probability of Localist Speeches . . . . . . . . . . . . . . . . . . . 97 Table 6.4: Vote Share and Constituency Representation with Country Dropout . . . . . . 98 Table A.1: Comparative Agendas Project, Major Topics . . . . . . . . . . . . . . . . . . 125 Table A.2: Breakdown of Training Set . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Table A.3: BERT 10-fold Cross-Validation scores, Policy Topic . . . . . . . . . . . . . . 126 Table A.4: BERT 10-fold Cross Validation scores, Localism . . . . . . . . . . . . . . . 126 Table A.5: Balance Statistics for Experiment Sample, Respondent Characteristics . . . . 128 Table A.6: Balance Statistics for Experiment Sample, Conjoint Attributes . . . . . . . . 129 Table A.7: AMCEs for Personal Vote Choice . . . . . . . . . . . . . . . . . . . . . . . 130 x Table A.8: AMCEs for Constituency Choice . . . . . . . . . . . . . . . . . . . . . . . . 131 xi LIST OF FIGURES Figure 1.1: Illustration of the Argument and Research Design . . . . . . . . . . . . . . . 9 Figure 2.1: An MP’s most important role by country . . . . . . . . . . . . . . . . . . . 20 Figure 4.1: Localist speeches and reelection . . . . . . . . . . . . . . . . . . . . . . . . 59 Figure 4.2: Representation focus and Reelection probability . . . . . . . . . . . . . . . . 62 Figure 4.3: Representation focus and Reelection probability with Country Dropout . . . 66 Figure 5.1: Map of Sampled Constituencies . . . . . . . . . . . . . . . . . . . . . . . . 71 Figure 5.2: Determinants of Voting for the Incumbent: Respondent’s Vote Choice . . . . 79 Figure 5.3: Determinants of Voting for the Incumbent: Respondent’s Estimation for the Constituency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 Figure 6.1: Localist speeches and reelection . . . . . . . . . . . . . . . . . . . . . . . . 93 Figure 6.2: Party Dominance and Representation focus . . . . . . . . . . . . . . . . . . 96 Figure 6.3: Party Dominance and Representation Focus with Country Dropout . . . . . . 99 Figure A.1: Diagnostic Plots for Ch. 3 Main Model (Model 1) . . . . . . . . . . . . . . 127 Figure A.2: Diagnostic Plots for Ch. 6 Main Model (Model 1) . . . . . . . . . . . . . . 132 xii CHAPTER 1 INTRODUCTION Over the past three decades, the vast majority of African countries institutionalized multi-party elections. Over time, elections have become the dominant mode of competition, and, ultimately, of ascension to power. The international donor community has supported extensively this introduction of democratic institutions (Norris, 2014) and the scholarly community has studied these develop- ments at length (e.g., Bleck and Walle, 2019; Bratton and Van de Walle, 1997; Lindberg, 2006). As major political players have tried to tilt the field in their favor (e.g., Bogaards and Elischer, 2016; Levitsky and Way, 2010), an extensive literature has developed on topics related to electoral manipulation and malpractice (e.g., Asunka et al., 2019; Norris, 2014; Schedler, 2002; van Ham and Lindberg, 2015). Overall, a vast amount of attention has been devoted to the institution of elections and to the processes related to it. However, in the African context, much less attention has been devoted to the very reason that elections are instituted: to provide a mechanism for citizen representation. It is important to ensure that the institutional pathways of representatives to power function as intended, i.e. that elections are free and fair. Nonetheless, it is also critical to study the linkages between representatives and represented. Even if elections function institutionally as intended, it remains to be seen whether those voted into office strive to provide representation, and also, whether representation is rewarded by the voters. Existing evidence indicates that African voters are not satisfied with their Members of Parliament (MPs). To begin with, surveys of Africans demonstrate that they are unhappy with their parliaments and they trust them less than other institutions. Looking at the most recent Afrobarometer survey, we see that more than 53% of respondents trust the executive and the courts somewhat or a lot but only 44% of respondents trust the parliament.1 Given the low levels of trust in the institution, it is not surprising that the levels of turnover among MPs are extremely high. Compared to reelection 1Looking at the 7th round of the Afrobarometer, 35% of respondents trust the president a lot and 18% somewhat. Similarly, 28% trust the courts a lot, and 25% somewhat. However, only 23% trust the parliament a lot, and 21% somewhat. 1 rates of about 70% in OECD countries, MP reelections in Africa are typically between 30-50% (Warren, 2019). While turnover via elections is generally regarded positively as a sign of healthy democratic competition, at extreme levels it may in fact facilitate democratic backsliding. When parliamentary turnover is exceedingly high, institutional memory is eroded at the end of each parliamentary term. Newcomers in parliament adjust to their new roles over a period of time. The more the balance between veteran and newcomer MPs tilts towards the latter group, the longer we may expect this adjustment period to last. This makes it difficult for parliaments to become strong institutions capable of checking the executive and preventing democratic backsliding (Barkan, 2008; Mattes and Bratton, 2007; Svolik, 2013). This is especially the case since presidents in these same countries get reelected at overwhelming rates. For example, between 1990 and 2015, 87% of presidents who ran were reelected (Bleck and Walle, 2019). This percentage is far higher than the 55% world average for presidential elections (Arriola, 2013). The combination of an entrenched executive and an inexperienced parliament makes power grabs from the former easier to enact (Barkan, 2009; Levitsky and Ziblatt, 2018). In addition, such processes have the potential to set in motion a vicious cycle. In contexts with powerful executives and weak democratic development, parliaments are less productive when it comes to their legislative function (Frantzeskakis and Seeberg, 2022). Therefore, it is important to further study the linkages between voters and MPs and the conditions under which they can be strengthened. When theorizing about the linkages between politicians and voters, existing literature on par- liaments in Africa has focused primarily on resource transfers. The linkages between MPs and their constituents are understood to be mainly clientelistic at the informal level (Kramon, 2016, 2018; Wantchekon, 2003) and constituency service-based at the formal level (e.g., Lindberg, 2010; Ofosu, 2019). As a result, the electoral fortunes of MPs are thought to be tied to the degree they use resources to cultivate a clientelistic base in their constituency (e.g., Arriola, 2009; Wantchekon, 2003; Warren, 2019). The effect of representation is considered to be minimal, with voters re- warding the provision of visible and tangible goods (Harding and Stasavage, 2014; Keefer, 2007). 2 However, given their access to state funds and the perks of their office, incumbent MPs should be better able to provide clientelistic benefits and constituency services than most challengers. Instead, this dissertation approaches the aforementioned outcomes as the manifestation of a crisis in representation, and, more specifically, as a deficit in local representation. The central thesis is that Africans care about representation in parliament. I start from the premise that, in the African context, due to the comparatively low state penetration in the periphery, voters prioritize local political cleavages over national ones. This position is compatible with existing research on political behavior in Africa. Many studies emphasize the importance of clientelism (e.g., Kramon, 2016, 2018; Wantchekon, 2003) and constituency service (e.g., Lindberg, 2010; Ofosu, 2019) for voting behavior. The processes related to these activities take place largely at the local level. Similarly, an extensive literature has focused on the importance of ethnicity for political outcomes (e.g., D. L. Horowitz, 2008). When we consider that ethnic groups tend to be clustered in space, it becomes plausible that many of the “ethnic" priorities can also be understood as “localist" priorities (Boone et al., 2022; Kim, 2017). The literature’s focus on these concepts denotes the importance of the local element for political outcomes in the region, although this unifying link is rarely discussed in an explicit way. Building on this idea of the salience of local politics, and since connectivity with the center is precarious, I argue that voters care deeply about their MP’s ability – and willingness – to communicate to the center the problems that they face locally. The argument proposed in this dissertation departs from the dominant explanations in the literature, which focus on clientelism and constituency service, and focuses instead on the value of local representation. Representation is a foundation of modern liberal democracy (Dahl and Shapiro, 2015) and may be defined as that “the power of the people should be parted with, and given over, for a limited period, to the deputy chosen by the people, and that he should perform that part in the government which, but for this transfer, would have been performed by the people themselves" (Pitkin, 1967: 150). In other words, representatives should take notice of the concerns of the represented, take them to the government, and help address them (e.g., Barkan and Mattes, 2014; Lindberg, 2010). In line with 3 the above, in this manuscript, representation is conceptualized as realized in action. Contrary to static conceptualizations, in order to have representation, the representative needs to purposefully act. Specifically, representation is achieved when representatives make representative claims in the name of the represented (Saward, 2014). The process of representation includes two stages. The first is getting informed regarding the needs and wishes of the represented. In the African context, this translates to holding meetings with constituents during which the MP can hear their concerns. Arranging such meetings is considered a sign of commitment and interest from the MP. On the other hand, not arranging them is viewed very negatively by the voters, who experience this as neglect (Bratton et al., 2005; Nijzink et al., 2006). The second stage is communicating these concerns to the central government. While other options are available, an MP’s most visible platform is the floor of parliament. MPs that want to highlight issues of concern to their constituents can introduce them during parliamentary debates in the form of statements, ministerial questions, or even as contributions to bills. When utilizing this forum, MPs signal to their constituents that they are actively representing their concerns and advocating for solutions (Laver, 2021). In addition, they flag issues of concern for ministers and parliamentary committees, which increases the likelihood that they can be addressed through state-initiated action (Proksch and Slapin, 2012). Existing literature on African politics expects that voters view parliamentary speeches as cheap talk (e.g., Keefer, 2007), while their linkage with their MPs relies on clientelism and constituency service (Lindberg, 2010). However, I pose that, especially in high-need contexts, it is exceedingly difficult for MPs to have sufficient resources to “buy" or “develop" their way to reelection. As a case on point, Lindberg (2010) discusses how the demands of local party activists in Ghana grew exponentially over time. While during the earlier elections of 1992 and 1996, most requests were for used bikes or bike parts, by the 2008 elections, the requests were for motorbikes or even cars (Lindberg, 2010). This rapid inflation renders the clientelistic model unsustainable. A reliance on constituency development can similarly lead to discontent for certain parts of the population if said development projects do not address their own needs. On the other hand, providing local 4 representation is not subject to these limitations. Instead, as I discuss at length in Chapter 2, representation is more abstract and it opens up the potential for services, material benefits, or a more favorable status quo down the line. In order to have such a representation linkage, it is important that information on what happens in parliament can reach the voters. In recent years, this kind of information dissemination has been possible via a number of channels. In several countries, parliamentary sessions are broadcast on the radio or TV and streamed on the parliament’s web page, or even on social media platforms like Facebook. The existence of such broadcasts creates several avenues for the information to disseminate. Local “information brokers", like local news outlets and chiefs can keep tabs on what happens in the capital especially if they expect their MP to speak on issues important to the home constituency. Another channel through which such information reaches the voters is WhatsApp. Group chats devoted to discussing local politics appear to be especially popular among younger voters. There is also anecdotal evidence that this information loop closes with citizens giving feedback to their MPs. In Ghana, for example, MPs report that their constituents call them to give their opinion on what they hear them say on the floor (Lindberg, 2010). Such incidents indicate that some segments of the public monitor the behavior of their representatives. Furthermore, in countries like Kenya and Malawi, the broadcasts have led to the censure of senior politicians when their speeches lead to public outcry. Therefore, I argue that MPs who prioritize constituency representation will benefit at the polls and they will be more likely to be reelected. However, if providing local representation is an effective electoral strategy for MPs, why is it not more prevalent? I argue that the viability of this approach is affected by the level of intra- and inter-party competition at the local level, and from the level of party control. While parties were not discussed in earlier parts of the theory they are a crucial part of the political system. MPs usually compete in elections as candidates for a specific party, which means that, for an MP, the party leadership is an additional principal. Voters and party elites can have different priorities and expectations from MPs, which makes keeping a balance between the two potentially difficult. While I expect that voters will always want their MP to provide local representation, party elites 5 will have different priorities based on how safe they consider the party’s hold to be on a given constituency. I argue that whether MPs prioritize providing local representation depends on the level of party dominance in the constituency. In several African countries, local-level competition is very low (Wahman and Boone, 2018). In many constituencies, a specific party can expect to win the presidential race comfortably regardless of the name of the legislative candidate on the ticket (Demarest, 2021). In such constituencies, MPs will be keen on retaining the party nomination. Therefore, in order to maintain the electoral boost that comes with representing the locally dominant party, MPs elected in such constituencies will be more likely to seek the approval of the party leadership than that of the voters. Under such circumstances, representation may be forsaken for the sake of towing the party line and scoring points with the leadership. On the other hand, when the levels of local competition are high, or when the constituency is expected to vote for a specific MP regardless of their party allegiance, the MP should be more likely to prioritize the wishes of the voters. Under such circumstances, constituency representation is more likely to be a priority. Taken together, these arguments provide a comprehensive account of the dynamics of local representation provision in Africa. This study makes several contributions to the literature. Most importantly, it provides evidence for the existence of an accountability mechanism between politicians and voters that goes beyond the clientelistic linkages that have been the focus of existing literature. Also, it undermines earlier accounts that parliamentary politics have a very limited impact on electoral outcomes including voting behavior. In addition, this dissertation adds to our understanding of democratic consolidation in the continent. It suggests that low levels of subnational competition, which empower central party elites, have often hindered individual MPs from fostering strong links with constituents. This in turn increases the likelihood of MP turnover and erodes the ability of parliament to conduct executive oversight. Beyond the realms of political representation and accountability, the study contributes a unique cross-national dataset that can be leveraged to explore a host of additional research questions. 6 1.1 Scope Conditions This manuscript seeks to explain constituency representation in parliaments in Africa and what it means for electoral support. The manuscript’s empirical focus is on Anglophone Africa since the re-introduction of multi-party elections in the 1990s, but there are no structural reasons why the theory cannot be applied to other countries in the global south with similar institutional arrangements and center-periphery dynamics. The theory is rooted in the premise that voters care about their concerns being heard. There is evidence that providing local representation is considered important in similar contexts, although relevant contributions to the literature stop short of evaluating the effect of local representation provision on reelection (e.g., Dockendorff, 2020; Motolinia, 2021). Furthermore, the focus is on multi-party regimes, but the dynamics discussed here should in large part also play out in legislatures within a single-party regime. As long as MPs are directly elected, the significance of local representation might be even more pronounced in a context devoid of party competition. When there are no national-level alternatives to consider, voters are more likely to concentrate on issues affecting their locality (e.g., Malesky and Schuler, 2010). 1.2 Research Design Overview The manuscript contributes a variety of novel datasets that are being used to test the theoretical claims presented. Given that the theory proposed in this manuscript takes place on multiple levels and includes several moving parts, a diverse set of data and methodological approaches is necessary to test it. I utilize a variety of observational and experimental data, as well as qualitative data from elite interviews. Quantitative tests are conducted using an original cross-national dataset on parliamentary speaking in six countries, as well as original experimental data on vote choice from Malawi. In addition, the development of the theory is informed by open-ended elite interviews. These qualitative data were collected during a fieldwork trip in Malawi in Summer 2022. In this section, I outline how these datasets are used to test the various components of the argument presented in the manuscript in order to improve both internal and external validity and to investigate causal processes. 7 The theory’s main implications are manifested at the MP level. The individual decisions of a constituency’s voters are also aggregated into a collective decision at that level. In addition, that is the level at which the tension between the influence of the individual MP vs their party is substantiated. For theory-building purposes, and to better understand these relationships, I conducted in-depth, open-ended, elite interviews in Malawi. In total, I interviewed sixteen MPs, two members of local NGOs, and one academic based in Malawi. The interviews lasted between thirty five and ninety minutes, with most of them being about one hour long. Throughout Chapter 2, I refer to these interviews to provide illustrative examples of the concepts and the relationships discussed. The focus of the theoretical argument is at the within-country level, but the findings are intended to be generalizable and to apply across countries. To test the theoretical implications in a way that satisfies the above criteria, I developed an original time-series, cross-sectional, MP-level, quantitative dataset on parliamentary speaking in six countries. This dataset covers Botswana (2014-2019), Ghana (2005-2021), Malawi (1999-2019), Uganda (2011-2021), Zambia (2001-2021), and Zimbabwe (2008-2018). In total, this amounts to about 960,000 speeches made by 3,800 MPs. The speeches have been coded based on a number of indicators discussed in detail in later sections. The bulk of the data was coded with the use of a neural network, trained on data that were hand-coded by a group of research assistants. This dataset represents the first attempt to classify parliamentary speech to this level of disaggregation and at such a scale. The empirical analyses conducted at this level generate external validity and attest to the generalizability of the theory. In addition, I conducted a conjoint experiment in order to directly evaluate the importance of local representation in parliament for vote choice at the individual-level. This experiment facilitates testing the causal processes that lead to the aggregate-level implications that are being explored with the use of the cross-national dataset described in the previous paragraph. The experiment was conducted in Malawi. It took place across twelve constituencies and included 2,200 respondents. This test lends credibility to the causal mechanism developed as part of the manuscript’s theory. 8 Finally, a critical assumption made in the theory is that information regarding an MP’s behavior in parliament disseminates to the individual voter. For parliamentary behavior to affect vote choice, it is critical to establish the degree to which voters can accurately estimate whether their elected representative provides constituency representation while in parliament. To test that such an information link exists, I combine data gathered as part of the surveys conducted for the experiment with data on the parliamentary behavior of the MPs for the corresponding constituency. This allows me to test the ability of voters to estimate the levels of local representation provided by their sitting MP at the time that the survey was fielded. A schematic illustration of the arguments made in this manuscript and of the data used to test them are provided in Figure 1.1. In this illustration, the nodes represent the main parts of the argument. The arrows indicate how these are connected. The labels on the arrows specify the data used to establish empirical support for the existence of a relationship between the variables connected by the specific arrow. Figure 1.1: Illustration of the Argument and Research Design p a rli a m e n t a r y s p e e c h e s y+ voters know if surve s it is provided he speec Local Reelection Representation conjo int exp + . voters care parliamentary about it speeches Party Dominance 9 1.3 Overview of the Manuscript Chapter 2 develops the manuscript’s theoretical argument and the research design employed to test it. I begin by explaining why representation, and more specifically local representation, is seen as vital in the African context. I then describe how taking the floor of parliament satisfies citizen demands for local representation. Following that, I develop the main argument, that providing local representation in parliament reinforces an MP’s chances of getting reelected. Afterward, I theorize why local representation is not more prevalent despite the electoral boost that it provides. At this part of the story, I explain how local party dominance affects the dynamic discussed up to this point and how it limits the ability of MPs to provide local representation. Chapter 3 discusses case selection and provides background information on the countries included in the sample. The chapter begins with a discussion of the main criteria that were considered for inclusion in the cross-national dataset. Next, the case selection for the conjoint experiment is motivated. Following that, I present a profile for each country included in the sample in order to provide some background information relevant to the theoretical story argued in this manuscript, and to showcase the diversity of the sample. Chapters 4 through 6 contain tests of the empirical implications suggested by the theory. To com- prehensively test the stated hypotheses, I conduct a series of analyses evaluating both constituency (MP) level and individual (citizen) level outcomes. The analyses include original quantitative data on parliamentary speeches across six different countries for a total of eighty years, and original experimental data from Malawi. The observational data are used to test high-level relationships between local representation in parliament and reelection, and party dominance and local repre- sentation. The experimental data are used to test the relationship between local representation and individual-level vote choice. Finally, a combination of individual-level parliamentary data is used to test citizen knowledge regarding the levels of local representation in parliament. In accordance with the proposed theory, the analysis shows that MPs who focus more on representing their constituents in parliament are significantly more likely to be reelected. In addition, the levels of party dominance in the constituency significantly affect how much the MP 10 will focus on constituency representation. In constituencies where the party is dominant, the MP makes fewer speeches in parliament representing the constituency. On the other hand, when either competition is high or the MP rather than the party is a mainstay of the local political scene, the focus on representation is significantly higher. Furthermore, the experimental data show that citizens are more likely to vote for incumbents who focus on representing the constituency in parliament and that respondents are aware of their MP’s level of focus on constituency representation. Chapter 7 presents the conclusions of this study. First, it summarizes the findings and their implications for the study of Africa specifically as well as for wider scholarship in political science. Then it suggests fruitful areas for further study. 11 CHAPTER 2 THE DYNAMICS OF LOCAL REPRESENTATION PROVISION This dissertation focuses on constituency representation in parliaments in Africa and what it means for electoral support. In several African states, given the low levels of state penetration, voters prioritize local over national cleavages even if the electoral competition at the national level is high. In a sense, this is not lost on the existing literature which, in order to answer questions regarding MP reelection, focuses overwhelmingly on clientelism, and on constituency service – both of which are locality-oriented activities. I argue that an overlooked determinant of vote choice in this context is local representation. While clientelism and constituency service are individual or club goods, representation is a public good. The provision of the public good of representation opens up the possibility that a desired outcome manifests through action at the state level. This outcome may take the form of material goods, legislation, etc. Overall, I argue that voters will favor MPs that are active in representing their constituencies in parliament. However, if providing local representation is a beneficial electoral strategy, we should expect that more MPs would take advantage of it and get reelected. Therefore, why do the MPs not prioritize it more? I argue that a critical factor to consider is support for the party vis-à-vis support for the MP within the constituency. In many constituencies, a locally strong party enjoys overwhelming support from the voters. I argue that MPs elected in such constituencies may be more likely to seek the approval of the party leadership than that of the voters. Under such circumstances, representation maybe takes a back seat to towing the party line in order to score points with the leadership. On the other hand, when the levels of the local competition are high, or when the constituency is expected to vote for a specific MP regardless of their party allegiance, cultivating ties with the voters is critical. Under such circumstances, local representation will be a higher priority. In this chapter, I develop a theory on how, and under what conditions, the provision of local representation can turn the electoral tide in favor of sitting MPs. The chapter starts with a discussion on the conceptualization of representation which is built on recent theoretical discourse 12 on the topic. After developing the conceptual foundation, it discusses the existence of demand for local representation among voters. Following that, I develop my theory regarding the dynamics of the provision of representation. First, I introduce the argument regarding the value of parliamentary speech-making as a form of representation. Then, I analyze the dynamics between parties and MPs, and how they affect the MPs’ focus on local representation. 2.1 Conceptualizing Representation “Representation" is often conflated with the concepts of “accountability" and “responsiveness," which are also central in our understanding of democracy. However, while inextricably linked, these concepts are substantially different. Representation refers to acting in the interests of the represented. Accountability and responsiveness are crucial steps in achieving representation. Responsiveness relates to whether the elected officials change their agenda based on signals they receive from the voters. On the other hand, accountability is related to the sanctioning ability of voters to remove the elected officials who fail to be responsive to their signals (Przeworski et al., 1999). In this manuscript, representation is conceptualized as realized in action. In this sense, represen- tation is achieved when representatives make representative claims in the name of the represented. This is a critical departure from more static understandings of representation and it is key in un- derstanding the importance of parliamentary speeches as an act of representation. In the following paragraphs, I briefly summarize the theoretical debate that led to my preferred conceptualization of representation and I explain why it is superior to alternative conceptualizations. In her seminal work, Hanna Pitkin (1967) discusses four different understandings of represen- tation: formalistic, descriptive, symbolic, and substantive. Formalistic representation focuses on the bestowment of the authority to represent. Representation is achieved simply because a repre- sentative is selected through some form of institutional means. Descriptive representation denotes the similarities between represented and representative. Representation is achieved when the repre- sented and the representative share some key descriptive characteristics (i.e., race, gender, religion, etc.). Symbolic representation resides in how the represented views the representative. This aspect 13 of representation focuses on how the presence of a representative –usually by an underrepresented group– is viewed by others and how it affects them. Finally, substantive representation focuses on how the representative acts in order to address the needs and demands of the represented. While the first three types are static, the latter focuses on ‘acting for representation,’ and is therefore superior (Pitkin, 1967). Traditionally, debates on substantive representation revolved around representatives as delegates vs representatives as trustees. Delegates would receive a clear mandate from the voters that stipulated how they should do politics, virtually diminishing the independence of the representative. On the other hand, trustees would be granted broad discretion to act in the interests of the voters (e.g., Burke, 1774; Pitkin, 1967; Rakove, 2010; Rehfeld, 2005). The distinction can be traced back to Magna Carta, at a time when representatives were only given authority to assent to the king’s request for money, and it has slowly developed as representatives have received increasingly more power over time (Rehfeld, 2005). In modern contexts, representatives tend to alternate between the two roles (Mansbridge, 2003; Pitkin, 1967). Building off of Pitkin’s work, recent scholarship has led to a constructivist turn in our conceptual understanding of representation, i.e. to the notion that representation exists by virtue of being performed. In this school of thought, in order to have representation, it is necessary to have “representative claims" (Saward, 2010, 38). Representatives become so when they actively make the claim that they speak for others. As a result, representation includes performance in a literal sense (Saward, 2014). While primarily normative, this debate has critical implications for the empirical study of representation. The constructivist turn is primarily set up in a back-and-forth debate by Mansbridge and Rehfeld (Mansbridge, 2003, 2011; Rehfeld, 2009, 2011). Mansbridge proposes that we move on from the delegate-trustee dipole. Instead, she introduces four new ideal types of representation: promissory, anticipatory, gyroscopic, and surrogate. Promissory representation focuses on fulfilling promises made on the campaign trail. Anticipatory aims at predicting how voters would like the representative to react to a changing post-election world. A gyroscopic representative is committed to a set of 14 goals/policies, self-reliant in judgment, and relatively unresponsive to electoral sanctions – or the expectation of sanctions. Finally, surrogate representation refers to the representation of the whole as opposed to one’s own constituency (Mansbridge, 2003). Mansbridge’s approach puts the relationship between the constituency and the representative at center stage. The relationship itself is characterized by the representative’s actions (Mansbridge, 2009, 2011). Rehfeld pushes against Mansbridge’s set of ideal types as over-complicated, and instead high- lights the hidden dimensions that are collapsed within the trustee-delegate bipole: “(1) should a representative-decisionmaker pursue the good of the whole or a part?; (2) should he or she rely on his or her own judgment, or that of a third party?; and (3) should he or she be more responsive or less responsive to sanctions for his or her work?"(Rehfeld, 2009, p. 221). This approach offers a better understanding of the individual characteristics of a representative (Rehfeld, 2009). In addition, this individual representative-based approach helps apply concepts of representation beyond the realm of electoral regimes (Rehfeld, 2011). Finally, Saward (2014) adds an extra wrinkle to the debate by expanding on how representatives “position" themselves in their roles. A representative might at different times position themselves as an anticipatory or a gyroscopic representative. However, Saward goes one step further by arguing that representatives can also position themselves in roles usually thought of as static. For example, in different circumstances, a representative might position themselves as a descriptive representative, a moral representative, a champion for a specific cause, etc. The mode in which these changes in positioning can be made is through speech and overt action from the representative (Saward, 2014). In this manuscript, I conform to the view of representation as realized in action. Empirical research has shown that when failing to act, representatives inevitably provide poor service to the people they represent. Currently, most regimes have a formally elected legislature. However, many autocratic regimes use the legislature as well as other democratic institutions as facades. In such cases, the existence of formalistic representation may bring very limited benefits to the represented (Gandhi et al., 2020; Williamson and Magaloni, 2020). Furthermore, in several cases, policies 15 aiming to increase the descriptive representation of certain groups have limited impact on policy outcomes that would be beneficial for the group (e.g., Bird, 2014; Clayton, 2021; Clayton et al., 2014). While being in the chamber – as formalistic and descriptive forms of representation would entail – is a prerequisite, very limited benefits can come from presence devoid of action. 2.2 Voter Demand for Local Representation in Parliament As discussed in the previous section, representation is a multifaceted concept. Formally, an MP’s role entails that they are present when the chamber is in session and that they participate in lawmaking, executive oversight, and constituency representation. Lawmaking and executive oversight are tasks with a broader impact on the entire country. On the other hand, constituency representation denotes a much more localist predisposition. As a result, national policy and constituency work can be seen as two isolated spheres, or archetypes, of activity (e.g., Fenno, 2003; Norton, 1993). As elected representatives, different MPs may follow different paths in order to fulfill their roles. In principle, as well as in practice, MPs are given broad discretion to act between elections. They may divide their energies along the different functions of their office, or they may focus on some of them and neglect others. One factor to consider is the institutional design of electoral institutions, which is widely accepted as a primary determinant of whether MPs will prioritize local representation. In countries with proportional systems, voters often primarily vote for the party, not the MP. The goal is for the national legislature to be representative of the composition of the nation (Lijphart, 2004, 2012; Powell, 2000). On the other hand, in countries with majoritarian systems, like those on which I focus in this project, there is a bigger emphasis on the accountability of the representative specifically to their constituency (Barkan and Ng’ethe, 1998; D. L. Horowitz, 1991, 2003). In these systems, usually, a single MP is elected in each constituency (SMD), and their mandate is much more “localist." The MP is expected to put the needs of the constituency first (Fenno, 1977; Przeworski et al., 1999), and they are encouraged to prioritize local representation (Carey, 2008; Crisp et al., 2004; Morgenstern, 2003). In SMD systems, the prioritization of the constituency comes as no surprise, but it is critical 16 to differentiate between constituency representation and constituency service. In both cases, MPs focus primarily on their constituencies, but representing the constituency does not mean bringing home the pork. Constituency representation is a collective good, provided when the MP brings attention to the problems and challenges that their constituency faces. On the other hand, constituency service manifests in the form of the provision of either club or private goods (Lindberg, 2010). The provision of the public good of representation opens up the possibility of material goods being provided. For example, if an MP regularly draws attention to their constituency’s pressing needs for improved roads, it is more likely for relevant provisions to be included when an infrastructure bill starts being discussed in parliament. Nonetheless, the public good of representation is distinct from and more encompassing than the specific material good –which may itself be a public good, as in the case of a toll-free central road– that it led to. In other words, providing representation is a constant and more open-ended process whose essence is found in giving voice to the will of the represented. While it may lead to the successful provision of material goods which is related to constituency service, it is not tied to it. In studies of the global north, we observe a shift in focus. While older studies used to focus on constituency service (e.g., Cain et al., 1987; Heitshusen et al., 2005; Lancaster, 1986; Lancaster and Patterson, 1990), a more recent wave of studies focuses instead on representation as manifested through formal channels like parliament (Geese and Martínez-Cantó, 2022; Poyet and Raunio, 2021; Schürmann, 2023; Schürmann and Stier, 2023; Zittel et al., 2019). Some argue that as societies develop, most interests should cut across electoral districts (Thomassen, 1994). Through this process of nationalization, national cleavages are expected to take priority over local cleavages (Lipset and Rokkan, 1967), which might be expected to decrease demand for constituency representation (Däubler, 2020). Instead, we see that MPs often look at the needs and the interests of their constituency in order to decide what kind of national policies to pursue (Bertelli and Dolan, 2009; Däubler, 2020). An MP may focus on issues that are currently salient in the constituency or issues they have reason to believe are important given the characteristics of the constituency (Papp, 2021; Poyet, 2021). Achieving policy objectives through initiating and passing legislation 17 would be ideal. However, the immediate goal is often to simply create attention around the issue of interest, even if this does not lead to tangible results in the short term (Bowler, 2010; Vliegenthart et al., 2011). Similarly, parliamentary questions can be used to bring constituency issues to the attention of the chamber (Saalfeld, 2012; Soroka et al., 2009). Overall, there is ample evidence of the importance of constituency representation in the parliament in Western contexts. However, when studying the global south, the focus remains on clientelism and constituency service rather than constituency representation (e.g., Baskin and Mezey, 2014; Keefer and Khemani, 2009; Ofosu, 2019; Wessel Tromborg and Schwindt-Bayer, 2019), with few exceptions (e.g., Motolinia, 2021). As a result, in the African context, much consideration has been devoted to constituency service. Several studies have focused on the proliferating Constituency Development Funds (CDF), which are resources granted to the MP to use for the developmental needs of their constituency (Barkan and Mattes, 2014; Ochieng’ Opalo, 2022). These funds, however, are not enough to cover the substantial needs that are present across an entire constituency. During open- ended interviews that I conducted in Malawi during May and June 2022, some MPs referred to the CDFs as something of a trap. They are sufficient to fund a single larger project or a few smaller ones, but deciding to pursue specific projects is almost guaranteed to alienate parts of the electorate that will not benefit from them. One MP in particular pointed out that it was necessary to build several new bridges throughout their constituency. However, the CDF funds were barely enough for one bridge. No matter where he decided to build the bridge, a large part of the constituency would be upset that it does not address their needs. Pleasing the voters becomes even less tenable when it comes to individual voter demands. Several of the MPs discussed at length the importance of establishing a mindset and remaining consistent. A number of MPs brought up the issue of coffins. Voters would frequently ask the sitting MP to provide coffins for the recently deceased. As the value of coffins is considerable, MPs were often unable to provide them, which was detrimental to their relationship with the requesting party. To avoid this, many of my interviewees said that they made it clear at the beginning of their campaign that they could not afford to provide coffins. In a world of limited resources, setting 18 clear boundaries is key to managing voters’ expectations. These observations point to the reason that clientelistic and constituency service-based strategies ultimately fail to impress the voters. In high-need contexts, it is virtually impossible for an MP to muster a sufficient amount of resources to address the needs and demands of their constituents. On the other hand, the aspect of constituency representation has been neglected in the region. This approach does not seem to stem from citizens’ lack of interest on local representation or from an overwhelming preference for constituency service. On the contrary, there are indications that voters care about the performance of politicians (e.g., Bhandari et al., 2021; Humphreys and Weinstein, 2010). In addition, the limited number of studies that have looked into citizen preferences in the region report that demands representation are quite substantial (e.g., Barkan and Mattes, 2014; Cho, 2012; Mattes and Mozaffar, 2016). Relatedly, a closed-answer targeted question was included in the fourth round of the Afrobarometer, which asked the respondents what they thought was the most important responsibility of an MP. In most countries, the majority of the respondents chose “listen to constituents and represent their needs." The second most frequent response was “delivering jobs and development to your constituency." In Figure 2.1, I present a density plot of these two responses by country. Indeed, in the majority of countries, most respondents indicate a preference for constituency representation over constituency service.1 Furthermore, I argue that constituency representation is even more critical in the context of Africa due to the comparative weakness of the state. Colonial administrative units were not designed with strong central power, and in most cases, parallel sub-national political institutions were preserved. Despite an attempt by post-colonial rulers to centralize power, African central states remain limited in their capacity and territorial reach (e.g., Boone, 2003; Cappelen and Sorens, 2018; Herbst, 2000). Under conditions of low state capacity, formal channels connecting local communities with the center are either limited or nonexistent (Alesina et al., 2011; Bates, 2008; Herbst, 2000). However, under SMD systems, the MP is both formally charged to care 1Interestingly, most countries in which citizens prefer constituency service over-representation, like Benin, South Africa, Mozambique, and Namibia, have proportional electoral systems. 19 Figure 2.1: An MP’s most important role by country 0.6 Percentage of Responses 0.4 0.2 0.0 a a a a Ke ny aw i na ia al i de th r ia ia e o n a ni nd bi wa Se a ha ig M Ve o be am qu Fa ni Af za ga Za n ne ga Mal er r so ib am s Be r ic Ta U m ts Zi m ba l G N e Le Li N bi in a th n Bo M ad ag bw e ap oz rk So as ca C Bu u r M Representation Service Note: The data are from Afrobarometer, Round 4, Q52. for the constituency and has direct access to the center. In this context, citizens expect their representative to provide a link between their localities and the center (Thomas and Sissokho, 2005; Barkan 1979) and value MPs who can utilize this link to make their voices heard on the national stage (Lindberg, 2010). This dynamic is reinforced by the relative scarcity of transportation infrastructure and nationwide mass communication, which leads to comparatively low integration of the periphery with the center, politics being nationalized to a lesser degree, and to local cleavages being prioritized over national ones. Thus, electoral campaigning remains much more local than in Western countries, and local issues and concerns can be of pivotal importance for success at the polls (Carrier and Kochore, 2014; J. Horowitz and Long, 2016). The process of local representation is generally understood to include two stages. The first is getting informed regarding the needs and wishes of the people being represented. In the African context, MPs are expected to hold meetings with their constituencies, during which they will get to know their concerns. Scheduling such time is considered of major importance, with voters often 20 complaining about the lack of opportunity to communicate with their representative (Nijzink et al., 2006). Constituents are less likely to continue supporting an elected representative who grows distant, or an MP whose stated goals seem to change once they travel to the capital. Voters want their representative to understand their problems and empathize with them, not to live in another world in the capital. Bratton et al. (2005) note the existence of a sizable gap in representation. They evoke several quotes from focus groups in Zambia regarding MPs, including that “they don’t attend to pressing national problems but only to issues concerning themselves," and “most MPs don’t go there (to the National Assembly) for charity reasons but to help themselves to a good standard of living" (Bratton et al. 2005: 242). In the same vein, in several countries, the evaluations of MPs correlate largely to whether they “speak about [the constituents’] needs", and “listen to the people" (Bratton et al., 2005). Overall, there are clear indications that Africans demand local representation from their MPs. Furthermore, technological advances help more citizens communicate with their MPs, even if the MPs do not travel to the constituency that frequently (Grossman et al., 2014), and to be better informed about politics (Conroy-Krutz, 2018b). One might argue that, given the diversity of African states, the kinds of problems different communities would bring to the MPs would be too diverse and idiosyncratic for them to coherently address in parliament. This would especially be the case if we ascribe to the narrative that ethnicity is the main driver of vote choice (e.g., D. L. Horowitz, 2008). If voters base their vote choice primarily on ethnic identity, their problems and concerns might otherwise not align. However, there is evidence to suggest that local concerns can be quite homogeneous. Citizens have shown that when deciding how to vote, they take into consideration the local ethnic geography. Even in contests where ethnicity is a salient cleavage, they may vote for a party associated with an opposing group if they expect it to win (Ichino and Nathan, 2013). This willingness to cross the ethnic line may point to the existence of a stronger driver of vote choice. Considering that ethnic groups tend to be clustered in space, it indicates that localized politics might in fact be the critical cleavage. Existing party strategies lend credence to this proposition. In order to increase their support, parties address appeals to individuals that belong to the same socioeconomic cleavage as their core 21 ethnic group of supporters. This strategy becomes viable as both within-country inequality and engagement in specific economic sectors tend to be clustered in space (Boone et al., 2022; Kim, 2017). Overall, these dynamics can lead to constituencies behaving in a homogeneous way. Under such circumstances, it is possible to nurture the shared norms and networks necessary to instill a preference for public goods, like representation (Habyarimana et al., 2007). The second stage of conducting representation is communicating these concerns to the central government. There are various avenues through which MPs may act on such concerns behind the scenes. If the issue is procedural, the MP may get in touch with the officials responsible for addressing it. If no such structure is in place, the MP may pursue a private meeting with the responsible minister, or bring the issue to the attention of the relevant parliamentary committee. However, an MP’s grandest platform is the floor of parliament. MPs who want to highlight issues of concern to their constituents can introduce them during parliamentary debates in the form of statements, ministerial questions, or even contributions to bills. During the debates, MPs have considerable discretion over the content of their speeches. This discretion allows MPs to bring up issues that do not pertain to their party’s agenda, or might even go against it (e.g., Herzog and Benoit, 2015; Schwarz et al., 2017). This is especially the case in SMD systems (B. J. Slapin and Proksch, 2021). When utilizing this forum, MPs signal to their constituents that they are actively representing their concerns and advocating for solutions (Laver, 2021). In the following section, I expand on this second stage and develop my theory on how constituency representation in parliament affects the electoral fortunes of MPs. 2.3 Parliamentary Speech-Making as a Form of Representation As discussed in previous sections, making representative claims is a key part of performing representation (Saward, 2010, 2014). In their seminal work, Proksch and Slapin (2012) argue that MPs tend to take the floor to explain and justify their position to their party and the voters. In this line of argument, legislators directly address the voters, whom they expect will be informed of the contents of their speech (Proksch and Slapin, 2012). This argument finds mixed support. While it is validated by several studies in various countries and institutional setups (e.g., Bulut 22 and İlter, 2020; Fenno, 1977; Fujimura, 2016; Marcinkiewicz and Stegmaier, 2019; Strøm, 1997), other studies dispute the effectiveness of participating in parliamentary activities as a strategy for reelection (e.g., Fiorina, 1981; Sulkin et al., 2015). The consensus in the literature on African parliaments sides with the latter group of scholars. The dominant expectation is that voters regard parliamentary speech-making as “cheap talk,” as they are not able to observe whether it leads to specific outcomes – as opposed to the provision of pork (e.g., Keefer, 2007). The linkage between MPs and voters tends to be understood as clientelistic at the informal level, and constituency-service based at the formal level (Lindberg, 2010). This is based on the understanding that African parliaments are weak institutions compared to an executive who has considerable influence on their composition (e.g., Barkan, 2008; Prempeh, 2008; Thomas and Sissokho, 2005). As a result, according to the literature, the electoral fortunes of MPs are tied to the wishes of the executive (e.g., Mozaffar and Scarritt, 2005; Opalo, 2017; Warren, 2019) and to the degree they manage to cultivate a clientelistic base in their constituency (e.g., Arriola, 2009; Wantchekon, 2003; Warren, 2019). Parliamentary activities overall are not considered significant when it comes to popular support. Contrary to much of the cited literature, I argue that parliamentary representation and respon- siveness are important and overlooked determinants of future support at the polls in Africa. As discussed in the previous section, providing constituency representation is distinct from providing constituency service, even in cases where both aim at the material development of the constituency. Representation is more abstract and has the potential to lead to services, material benefits, or a more favorable status quo down the line. For this potential to be realized, it is important for MPs to speak on the floor of parliament. As Proksch and Slapin (2012) propose, by taking the floor and speaking about local issues, MPs show to their constituents that they take the voters’ concerns seriously and that they are keen on performing representation. In addition, it draws the attention of the chamber to specific issues to be addressed. This allows the opportunity for other MPs to engage on this topic, especially if similar issues constitute problems in their own constituencies. This process increases the likelihood that a solution is implemented at the state level. 23 In several countries, parliamentary sessions are broadcast on the radio or TV and streamed on the parliament’s web page, or even on social media platforms like Facebook. In a period that media freedom is perceived as of decreasing importance in the continent (Conroy-Krutz, 2020) broadcasts of the parliamentary debates have in various occasions led to censure of senior politicians when parliamentary debates lead to public outcry. Very soon after the live coverage of the debates started in Kenya, in 2008, Finance Minister Amos Kimunya was forced to resign due to public pressure after corruption allegations were levied against him during the debate (Wanjicu, 2008). Similarly, in Malawi, in February 2022, Hon. Ulemu Chilapondwa received severe criticism for saying in a speech that cooking oil prices matter to the elite and not to people in the villages. He was forced to issue an apology after video and audio clips of the speech went viral on social media (Mzungu, 2022). Furthermore, attempts to limit the broadcasts meet resistance. In Tanzania, the executive’s decision to stop the coverage of the parliamentary sessions in 2016 was met with fierce reaction from the opposition parties, while previous attempts to stop coverage had been reversed after similar reaction by journalist groups (Mutiga, 2016). The existence of such services does not guarantee that a significant part of the population will actively monitor the behavior of their MP, but it creates avenues for the information to disseminate (Laver, 2021). For example, it makes it easier for local news outlets to monitor MP behavior and inform constituents of their activity. Local chiefs are also anticipated to pay closer attention to the MP’s behavior in parliament and to pass on their observations to the people in their jurisdiction. This is especially the case when such local “information brokers" become aware that the MP speaks in parliament about local issues and concerns and, thus, realizes the mandate to represent their home constituency. There is also anecdotal evidence that this information loop closes with citizens giving feedback to their MPs. In Ghana, for example, MPs report that their constituents call them to give their opinion on what they hear them say on the floor (Lindberg, 2010). Such incidents indicate that some segments of the public monitor the behavior of their representatives. As a result, at least in some contexts, MPs expect that their constituents will be informed of their performance in parliament, and they will hold them accountable for it. To provide an illustrative 24 example, on February 2nd, 2017, Hon. Ras Mubarak and Hon. Frank Annoh-Dompreh made suc- cessive statements in the national parliament of Ghana highlighting the need to enhance the amount and quality of information that the citizens receive regarding the MPs’ work in parliament. The statements came in the aftermath of the publication of MP performance scorecards by ODEKRO, a news outfit focusing on disseminating information about the parliament of Ghana. The speakers noted that similar initiatives had led to sitting MPs being voted out of office if voters considered their performance inadequate. The speeches seemed to strike a chord in the chamber, as no fewer than eleven MPs from both parties stood to associate themselves with the statements and to offer their views on the matter. Given the insistence for participation in the debate, the conduct of all other business was deferred several times. In the end, no other item of business was discussed before the Speaker of the House adjourned the meeting late in the evening. In several of the interviews I conducted, Malawian MPs shared stories corroborating the anec- dotes from Ghana. Out of the sixteen MPs that I interviewed, all but one expressed certainty that at least some of their constituents were aware of whether they spoke in parliament. When asked to explain their response, many MPs provided examples of constituents calling or texting them even within the same day to thank them or to complain about the contents of their speech. Moreover, many MPs explained that before the beginning of the parliamentary session, they would arrange meetings with chiefs, local notables, and voters in their constituency. During those meetings, they would specifically inquire about local issues that those present would like them to bring up in parliament. Parties that made such requests were especially likely to monitor existing information channels to see that their MP spoke about their particular issue. According to the interviews, most citizens learned about parliamentary discussions from two sources. The first is local radio stations, which would cover both national and local politics. Such stations discuss the contributions of the local MP, and they translate the main points into the local language. The second is through WhatsApp groups. Reportedly, most areas have WhatsApp groups about local politics that are especially popular among the younger segments of the population. Within such groups, details about speeches made by the local MP travel fast. Some of the MPs I interviewed said that they 25 were members of such groups themselves, and they made sure to highlight their contributions in the chamber. These anecdotes indicate the existence of a representation-centered linkage between MPs and voters in Africa that up to this point has been completely overlooked by the literature, but is evident in research based on surveys of African voters. Overall, I argue that African voters care about constituency representation in parliament, that they are aware of the level of effort their MP exhibits in representing local issues in parliament, and that this affects the MPs’ chances of reelection. Formally, I derive the following testable hypotheses: Hypothesis 1: MPs who focus on constituency representation in parliament are more likely to be reelected. 2.4 Choosing Between Principals If constituency representation is an effective booster to the likelihood of reelection, it remains unclear why not all MPs emphasize it. Up to this point, I have considered MPs in a state of competition, where they need to independently vie for support. However, MPs usually participate in elections as candidates for a specific party. The existence of parties complicates the relationship between MPs and voters, as the party leadership is an additional principal to which MPs are held accountable (Carey, 2007). Keeping a balance between the two can be a tricky proposition. The eventual winner depends on the results of the general elections, which are decided by the voters. Nonetheless, turnover frequently takes place during the candidate nomination process. While formally these selection procedures may be open, the party leadership tends to have a strong influence on who stands for the party in the general elections (Warren, 2019). The way that MPs address this double principal dilemma will affect whether they prioritize constituency representation. I argue that the degree of control the party has on the constituency affects the opportunity costs of engaging in constituency representation. Taking the floor to discuss constituency concerns is expected to have a positive effect on the MPs standing with their constituents. However, it might also harm their standing with the party leadership. First, time on the floor is a limited commodity. 26 Assuming equal access, when more MPs raise to take the floor, there is less time on average for each individual MP, including members of the party leadership. This is an important consideration even if in many contexts party leaders are institutionally allowed additional time on the floor. Second, it is not always clear how a speech might be perceived by the party leadership. Speeches on local issues are often given in the form of statements or questions to the minister, which are disconnected from general debates in which party discipline can be expected. Regardless, an MP’s willingness to take the floor and conduct local representation may be adversely affected by the way the anticipate the party leadership will perceive their contribution. Poyet and Raunio (2021) make a similar argument in the context of Finland, where they pose that increased inter-party competition will lead to increased speech-making and constituency- focused politics, while inter-party competition will lead to decreased speech-making (Poyet and Raunio, 2021). On the contrary, Dockendorff (2020) shows that Chilean MPs from unsafe seats are less likely to take the floor (Dockendorff, 2020). These accounts call attention to the issue of stronghold constituencies. As electoral contests have become significantly more competitive over time, representing one of the major parties should improve a candidate’s viability. However, a nationally competitive race is often made up of multiple local contests that are dominated by one locally strong party. Indeed, in many African countries with competitive national elections, we see local voting patterns that are highly regionalized (Ichino and Nathan, 2013; Wahman, 2017; Wahman and Boone, 2018). As a result, many constituencies form electoral strongholds for a specific party. From a candidate’s point of view, being nominated as the locally dominant party’s candidate is a huge boon, as it almost guarantees victory at the polls (Demarest, 2021). This is evident from the number of suitors who are keen on receiving the nomination. An increasing number of major parties in the region have started opening up their candidate selection processes by organizing party primaries. The candidates who compete in the contests fielded in stronghold constituencies are more numerous, and they spend considerably more during their campaigns (Wahman and Seeberg, 2022). Research on Africa as well as on other regions indicates that under such circumstances, MPs 27 make every effort to remain candidates for their party (Young, 2014), support the party line (Carey, 1998; Siavelis and Morgenstern, 2008), and try to avoid being perceived as “mavericks" (Searing, 1994). In turn, the party leadership is likely to reward such behavior (Proksch and Slapin, 2012; Yildirim et al., 2019) and punish deviations from it (Opalo, 2019). As a result, MPs who want to continue their career in politics come to view principal senior party elites as the “key" rather than their constituents (Demarest, 2021). Therefore, since time on the floor is a valuable and limited commodity (Proksch and Slapin, 2012; Searing, 1994), MPs should prioritize being on the good side of the party leadership over providing local representation, and they should be more likely to use their time on the floor to support party policy, or even forego speaking altogether to make it easier for senior leadership to have more time on the floor. From the party’s point of view, the calculus is geared towards leveraging its strongholds to maximize the party’s power at the national level. To remain viable, a party’s number one goal is to conserve and strengthen its control inside its own strongholds (Wahman and Goldring, 2020). This means fielding locally popular candidates, but not so popular that they may pose a threat to the established party hierarchy. MPs who start to amass power and influence may appear threatening to the party elites and, as a result, they are likely to be replaced so that they cannot challenge the party leadership from within its own stronghold. In addition, many of the party-safe seats are also used strategically by the party leadership to keep allies in the fold. To maximize the number of people who have access to state power over time, party elites may cycle out candidates (Warren, 2019). If ministers are required to be MPs, this strategy is also used by the executive to make sure that candidates for ministerial office get elected (Bratton et al., 2005; Demarest, 2021; Warren, 2022). Party primaries are also used strategically to achieve these goals. The higher echelons of party leadership yield some of their decision-making power for two possible reasons. The first is that they come under pressure by party members. As the hopeful nominees are willing to spend substantial resources to secure the support of the party grassroots and, therefore, the party nomination, local party members stand to gain. Thus, parties are forced to allow for primaries especially in valuable stronghold constituencies that attract a lot of hopeful nominees who are willing to spend a lot during 28 the nomination race (Ichino and Nathan, 2012). In these cases, parties employ elite primaries, in which the selectorate is much diminished, with only local party activists having a vote. Therefore, employing clientelism in the primaries is often considered much more beneficial than during the general elections (Acheampong, 2020). A second possibility is that the party leadership seeks to extend its viability by allowing a degree of within-party democracy. By organizing more open primaries it is possible for new blood to enter the party and replace incumbent candidates who have lost the favor of the party’s local base, and overall manage within-party conflict and help avoid defections (Warren, 2022). In such cases, the grassroots are more likely to mobilize, and that increases the viability of minority candidates who would not stand a chance of being selected for nomination otherwise (Ichino and Nathan, 2022). However, even in cases where primaries are more open, party elites still often assert great control through extra-institutional means. This has been shown in several countries including Botswana (Warren, 2022), Ghana (Acheampong, 2020), Kenya (Wanyama and Elklit, 2018), Malawi (Seeberg et al., 2018), Tanzania (Sulley, 2022), and Uganda (Kjær and Katusiimeh, 2021). Overall, such practices, including the maneuvering of party elites during primaries, should further decrease the likelihood of MPs in party strongholds engaging in constituency representation. Up to this point, my argument is very similar to the one proposed by Poyet and Raunio (2021), as the expectation would be to have more local representation in competitive constituencies compared to stronghold constituencies. However, compared to parties in Finland, parties in Africa tend to be quite weak, and the party system can be highly volatile. Looking at V-Dem’s party institutionalization index (v2xps_party) which takes values from 0 to 1, in 2014 Finland had a score of 0.939. In the same year, Ghana, whose parties are some of the most institutionalized in the continent, had a score of 0.668. On the other hand, Zambia, whose parties are considered poorly institutionalized, had a score of 0.366. As a result, it is important to take into consideration how the comparatively low levels of party institutionalization affect relationships discussed up to this point. Being nominated to represent a well-institutionalized party confers sizable competitive advan- 29 tages due to the party brand. On the other hand, in areas where their support is not as strong, parties vie to recruit candidates that have a strong local base of support. For example, in the 2015 elections in Tanzania, support for CHADEMA drastically increased in constituencies traditionally dominated by the governing Chama Cha Mapinduzi (CCM) following the defection of influential candidates (Sulley, 2022). Similarly, in a recent study, Arriola et al. find that in Zambia candidates able to fund their own campaigns and raise local support are much more likely to both be recruited by parties and to defect (Arriola et al., 2021). In such cases, at the local level, the candidate confers electoral viability to the party. However, it is not necessary for only one of these dynamics to be in play for a given party. On the contrary, in many African states parties are relatively young and do not have a strong presence across the entire country, which leads to unequal levels of competitiveness and institutionalization (Wahman and Boone, 2018). As a result, in some areas, the party may legitimize the candidate in the eyes of the voters while in others the candidate may legitimize the party. On this basis, I differentiate between party strongholds, in which the party nominee is guaranteed to win, and MP strongholds, controlled by baron-like MPs. In MP strongholds, the same MP wins consistently and is able to personalize power. This can be done by staying with the same party and claiming credit for positive developments. However, it can also be done by persuading the voters that they choose to switch parties in order to better represent constituent interests. Therefore, this latter type of constituency is highly likely to vote for their sitting MP, who is well-loved locally, regardless of party affiliation. These dynamics were evident in several of the interviews I conducted in Malawi, where the results of the primaries are known to change based on the party leadership’s preferences (Seeberg et al., 2018). When discussing their participation in the chamber, several MPs running in party strongholds were quick to point out that they avoided bringing up topics that might be seen as going against the party line regardless of national or local focus. A lack of information regarding the views of the party leadership on a specific topic also had a similar influence. MPs stated that sometimes it was preferable not to participate in the debate if they were not sure about the party 30 leadership’s position. In addition, MPs who represented competitive constituencies indicated that their colleagues elected in party heartlands were less likely to participate in the debates overall. When discussing crossing the floor, a practice that is fairly common in Malawi, many MPs said that given the party’s influence, it would be political suicide. Others, however, said that due to the strength of their bond with the voters, they could potentially run for a different party and win. Thus, I introduce the concept of party dominance. This concept captures the extent to which a party is able to carry a constituency regardless of the name on the ticket. Party dominance is high when support for the party in the constituency is strong even if support for the party’s candidate for the legislature is weak. On the other hand, party dominance in a given constituency is low when support for the candidate is stronger than support for the party they represent. Thus, this concept accommodates all the possible balance-of-power scenarios between parties and MPs. Formally, I propose the following hypothesis: Hypothesis 2: In constituencies where party dominance is higher, MPs will be less likely to engage in constituency representation. 31 CHAPTER 3 CASE SELECTION AND COUNTRY PROFILES In this chapter, I explain the considerations based on which cases were selected for the observational as well as the experimental parts of this study. Following that, I provide some background information regarding the recent political history and the institutional status quo for each country included in the sample. 3.1 Case Selection There are several points to consider when selecting cases for the MP-level analysis. One major goal is to determine the degree to which the findings are generalizable to a broader population of cases and robust to alternative explanations. To accomplish this, I selected countries that are diverse across a number of dimensions including the level of democracy, the level of party system institutionalization, the prevalence of ethnic politics, and candidate selection procedures. This allows me to test the theory across a diverse range of modes of competition. Aggregate levels of the level of democracy, the level of party system institutionalization, and candidate selection procedures for each country are provided in Table 3.1. On the other hand, these countries are comparable across several other dimensions. They are Anglophone countries, and they have presidential systems,1 MPs are elected in first-past-the-post single-member districts. The shared colonial legacy and key formal political institutions offer a consistent baseline for the entire sample. The existence of this baseline facilitates a test of the theory under a strict set of assumptions regarding the institutional setup. The final criterion regards data availability. Since collecting data on parliamentary speaking is essential, it is necessary to take into consideration the availability of Hansards, or parliamentary transcripts, which need to be available in digital form and accessible. Furthermore, they need to be available for an extended period of time and without significant gaps in the period covered. The steps necessary to compile this dataset included the collection of the Hansards, the collection of 1Botswana is an exception here, as it has a parliamentary system. However, the president is very powerful even if not directly elected. 32 data on the MPs (e.g., gender, party, district, etc.), and the coding of the speeches. Table 3.1: Aggregate Information for countries in sample Quality of Party System Candidate Data Country Elections* Institutionalization* Selection* Availability Ghana High High Local 2005-2020 Botswana High High Local 2015-2021 Malawi Medium Low Local 1999-2019 Zambia Medium Low Central 2001-2021 Zimbabwe Low Medium Regional 2009-2021 Uganda Low Medium Local 1996-2021 * The information for these variables comes from V-dem. While the case selection for the MP-level analysis aimed at including a large range of electoral environments, for the voter-level analysis Malawi was selected as the most typical case in the sample (Gerring, 2016). When looking at the measures of central tendency for the main variables of interest, such as rates of reelection and local representation provision, the results for Malawi are the closest match to those of the entire sample.2 Furthermore, Malawi is fairly typical when it comes to important contextual variables like the levels of democracy, candidate selection systems, and party system institutionalization. In the rest of the section, I provide some background information regarding the recent political history and the institutional status quo in each of the countries in the sample. 3.2 Country Profiles 3.2.1 Ghana Ghana is often seen as the poster child of African democracy. The country has held multiparty elections since 1992. Jerry Rawlings, a military officer who had been in power for most of the 1979- 1992 period, became the first president of the 4th Republic of Ghana. Starting from the inaugural 1992 elections and up to the present, all electoral contests have been fought almost exclusively 2Looking at reelection rates, for the entire sample the mean is 0.32 with a standard deviation of 0.47. For Malawi, the mean is 0.28 with a standard deviation of 0.44. When it comes to making localist speeches, the sample is 0.16 speeches per day with a standard deviation of 0.24. For Malawi, the mean is 0.19 speeches per day with a standard deviation of 0.29. These variables are discussed to a great extent in Chapter 4. 33 between the National Democratic Congress (NDC), which was created by Jerry Rawlings ahead of the 1992 elections, and the New Patriotic Party (NPP), which was established at the same time. Between them, these two parties have controlled all of the presidencies and 95% of the parliamentary seats. The NDC emerged victorious in the 1992 and 1996 elections when Jerry Rawlings was the presidential candidate. After his retirement, NPP won the 2000 and 2004 elections. Following that, NDC controlled the presidency after the 2008 and 2012 elections. Finally, NPP won the 2016 and 2020 elections. Ghana’s two main political parties are relatively distinct in terms of ideology. While both embrace social-democratic policies, the NPP follows a more liberal tradition, while the NDC follows a more social or even populist orientation (Brierley, 2012; Osei, 2012). Each party has a regional as well as an ethnic stronghold. NPP tends to perform better in the Southern regions of the country and within the Akan ethnic group. On the other hand, NDC performs better in Northern regions of the country and in non-Akan areas (Osei, 2012). However, while the salience of ethnicity is substantial, government performance is a critical determinant of vote choice (Harding, 2015; Weghorst and Lindberg, 2013). For the duration of the period, political institutions have remained largely stable. The electoral system is first-past-the-post single-member district (SMD). However, the number of constituencies increased from 200 in 1992, to 230 in 2004, and to 275 in 2010 (Sanches and Dias, 2021). Both parties instituted elite primaries through which they select candidates for parliament. In both cases, the selectorate of those primaries used to be constituted by the branch executives, which numbered between 200-800. Due to the small number of voters, vote buying took place extensively (Ichino and Nathan, 2012). In 2016, NDC broadened the pool to all registered party members, which led to a more diverse set of candidates receiving the nomination and running in the national elections (Ichino and Nathan, 2022). As it can be inferred by the absolute dominance of the NDC and NPP on the Ghanaian political scene, the country’s two-party system is very well institutionalized. They have a strong presence and distinct networks nationwide (Osei, 2012). The identities of the parties are linked to the way 34 they view the Rawlings Revolution of 1981. While NDC portrays it as a pivotal moment that led to the founding of a stronger democratic state, NPP views it as a significant step backward in the country’s democratization process. Candidates for office are recruited from within party networks, and they follow each party’s “founding mythology." These characteristics make the defection of MPs or candidates a less viable proposition than in most other countries in the region (Brierley, 2012). Overall, the Ghanaian parliament tends to act in unison with the executive, but it is fairly well institutionalized. Early on, the parliament’s funding depended on the executive. However, during the 4th parliament –which was in office for the period 2008-2012– individual committee budgets were introduced (Brierley, 2012). In addition, the parliament created a legislation drafting office, which is independent from that of the president in order to better retain control of the development of Bills (Hansard, 3/3/2011). These developments facilitated much greater independence within the legislative branch. The Ghanaian parliament holds three sessions per year for about twelve weeks each. This allows a substantial amount of time for the chamber to conduct business. The strong party identities lead to MPs generally acting as cohesive blocks when it comes to voting on legislation. In addition, the structure of the debates is such that party leaders and committee chairs are given considerable sway in deciding the items on the daily agenda. During debates, senior MPs like party leaders and committee chairs are allowed more time on the floor and are often given priority. Backbencher MPs have access to the floor independently of their parties. In addition, the formal rules are such that the back-and-forth between ministers and MPs, as well as between MPs, is encouraged (Sanches and Dias, 2021). 3.2.2 Botswana Botswana is considered one of the most democratic countries in Africa (Leith, 2005). The country has continuously held elections since 1965. While Botswana’s elections are considered free and fair, the Botswana Democratic Party (BDP) has been continuously in power since the inaugural elections took place. Unlike in many other cases, the British actively helped the BDP as it was being founded and facilitated the transfer of power after it won the 1965 elections (Good and 35 Taylor, 2008). Following this early victory, the BDP has remained in power and has consistently held overwhelming majorities in parliament. The most prominent opposition party has been the Botswana National Front (BNF), which joined with other opposition parties ahead of the 2014 elections to create the Umbrella for Democratic Change (UDC). In terms of ideology, both the BDP and the BNF have social democratic leanings, but BDP has had a more neo-liberal agenda (Molomo, 2000) and has strong ties with the private sector (Good, 2005). BDP has also had much broader access to resources. The party benefited from the state’s vast diamond wealth, donations from the private sector, and a near monopoly of access to state media. On the other hand, the opposition did not have any of these perks (Good, 2008). Overall, ethnicity has not been a major cleavage (Elischer, 2013). Botswana has a parliament that is directly elected in a first-past-the-post SMD system. This system is a major reason for the opposition’s seeming inability to compete with BDP (Mogalakwe, 2006). To present an extreme but indicative example, in the 2009 elections, BNF received 22% of the vote and got 10% of the seats, and another opposition party, the Botswana Congress Party (BCP), received 19% of the vote but 5% of the seats. On the other hand, BDP received 79% of the seats with 53% of the vote. The number of seats is relatively small, but it has grown significantly in the period since independence. There were thirty-five seats in 1965, thirty-eight seats in 1984, forty-four in 1994, sixty-three in 2004, and sixty-five in 2019. Since the inaugural elections about 10% of the seats are filled by appointment from the president. The president, while powerful, is not directly elected. Instead, following the parliamentary elections, the elected MPs vote for the presidency. In effect, this means that the leader of the largest party becomes president (Molomo, 2000). BDP started conducting non-binding constituency elite primaries in 1984, while both BDP and BNF started conducting mass primaries before the 2004 elections (Warren, 2022). Throughout the period, BDP has had a large advantage in terms of party institutionalization. It has a strong presence in the entire country, and its candidates are well-funded by the party. As a result, they are able to campaign much more effectively than opposition candidates, whose parties have a much harder time developing their networks in the more rural parts of the country (Good 36 and Taylor, 2008; Taylor, 2003). The introduction of mass primaries allowed BDP to optimize its candidate selection and to proceed to a “controlled turnover" of its legislators in parliament (Warren, 2022). Botswana’s parliament is weak vis-a-vis the executive. The president, while not directly elected, has the power to suspend or dissolve the parliament. On the other hand, the parliament cannot impeach or suspend the president. In case of a vote of no confidence in the executive, the parliament is also automatically dissolved (Mogalakwe, 2006). Furthermore, the parliament does not have its own independent budget. Finally, cabinet ministers are drawn from the ranks of the MPs, which means that a substantial section of the chamber is given ministerial portfolios. As a result of these institutional arrangements, it is very difficult for the parliament to check the executive (Mogalakwe, 2008). Motions passed in parliament aiming to strengthen it as an institution are often ignored by the executive, which allows the president to conserve the existing balance of power (Mogalakwe, 2006). Nonetheless, debates in parliament are active, and MPs get access to the floor with minimal constraints. 3.2.3 Malawi Malawi has held multiparty elections since 1994, following thirty years of one-party rule from the Malawi Congress Party (MCP). In the inaugural 1994 elections MCP’s Hastings Kamuzu Banda, who had been in power since 1964, was defeated by Bakili Muluzi, who was the candidate for the United Democratic Front (UDF). This early alternation in power set the tone for a tumultuous political environment. Since then, the balance of power has shifted multiple times, and two more executive turnovers have occurred. The party system is volatile and has been shaped by splits in mainstream parties, the rise of new parties, and the fall of once-predominant ones. While MCP has remained a significant force throughout the period, UDF was seriously weakened when Muluzi’s successor, Bingu Wa Mutharika, left the party to create the Democratic Progressive Party (DPP) in 2005 after he won the 2004 elections. DPP became a dominant force after the 2009 elections, but Bingu Wa Mutharika died in office in 2012. As a result, the estranged vice president, Joyce Banda, who had in the meantime left DPP and created the People’s Party (PP), became president. In the 37 next elections, Banda lost and Bingu’s brother, Peter Mutharika, became president as the head of DPP. Finally, after the annulment of the controversial 2019 elections by the Supreme Court, MCP’s candidate, Lazarus Chakwera, became president. As in much of Africa, Malawian parties do not neatly fit on the traditional left-right continuum. In most cases, party support is highly regionally concentrated. Based on some accounts, ethnicity was salient in Malawi politics in the early period after the introduction of multipartyism (Elischer, 2013). Other scholars have argued that regionalism rather than ethnicity has been the dominant cleavage (Kalipeni, 1997). In recent elections, regionalism has been indeed a major deciding factor. The Central region has traditionally been the MCP’s stronghold, while the Southern region has backed UDF and then DPP. Finally, the Northern region used to be the home base of the Alliance for Democracy (AFORD), but since 2004 the region’s constituencies have been swinging between the main parties (Ferree and Horowitz, 2010). This dynamic leads to some differentiation in policy preferences. For example, smaller parties and especially those based in the Northern region are proponents of decentralization and have been skeptical of initiatives that could lead to development at the possible cost of their marginalization (Frantzeskakis et al., 2021). For the duration of the period, political institutions have remained largely intact. Since the 1999 election, the country has had 193 parliamentary constituencies. The electoral system is First- past-the-post Single Member District (SMD) (Patel and Wahman, 2015). This system puts MPs in the spotlight in their individual constituencies and it encourages the development of a personal following. From the viewpoint of many MPs, this means that there is increased pressure on them to deliver clientelistic goods and constituency development, and less attention to whether they engage in the legislative process or on monitoring the executive (Wahman and Seeberg, 2022). The levels of party institutionalization are very low. Parties generally fail in enhancing their capacity, and elite defections are widespread. In order to expand from their traditional heartlands, parties try to co-opt local elites by offering them nominations to run for parliament. However, these alliances often do not last long-term. Local elites are prone to defecting if the party’s fortunes seem to worsen, and parties may switch their support to different locally prominent actors if their original 38 nominee’s local support seems to waiver. These dynamics play out in party-organized primaries, which are becoming notorious for high levels of manipulation and party-elite interference (Patel and Mpesi, 2010; Seeberg et al., 2018). Overall, the Malawian parliament is relatively weak. For the period covered in this study, its funding depended on the executive. Due to limited funds, the parliament tends to be in session only 70-100 days per year. By restricting the parliament’s funding and controlling the sitting calendar, the executive has historically been able to control it. A notable exception was the parliament’s successful intervention that blocked President Baliki Muluzi’s bid to change the constitution and run for a third term (Dulani and van Donge, 2005). Nonetheless, within parliament, the rules encourage MPs’ participation. Access to the floor is independent of seniority, and it is minimally constrained by the party elites. The institutional setting of the legislative debates in general facilitates a back-and-forth between MPs from different parties as well as between ministers and MPs (Frantzeskakis et al., 2021). 3.2.4 Zambia Zambia has held multiparty elections since 1991, after twenty-seven years of one-party rule under Kenneth Kaunda. In the inaugural elections, Kaunda’s party, the United Independence Party (UNIP), got crushed by the opposing Movement for Multi-Party Democracy (MMD), under Frederick Chiluba. Building on this early success, the MMD won convincingly in the 1996 elections. Prior to the 2001 elections, disputes over who would succeed Chiluba as the party’s presidential nominee led to major splits within the MMD’s leadership. This led to the creation of the United Party for National Development (UPND) and the Patriotic Front (PF). Despite this, MMD managed to retain control of the presidency in the 2001 elections, with a 2% lead over UPND’s candidate. In 2006, MMD won the presidency again, with a comfortable lead over PF, which came second. The results were much closer in the 2008 elections, which took place after the death of the incumbent president. In 2011, PF managed to win the presidency, with MMD coming second and UPND third. PF stayed in power after the 2016 elections. In both cases, the main opposition party was UPND. 39 As this brief overview indicates, MMD dominated Zambia’s political scene after the transition to multipartyism. In addition, both of the remaining two major parties were created as a result of elite splits within MMD. What started as a dominant party system has now developed into a competitive (Rakner and SvÂsand, 2004), fairly stable two-party system. Over the course of three decades, the main parties have been multi-ethnic. At the same time, the major ethnic groups have split their support among different parties. In the earlier period, while parties had a clear regional basis (Boone et al., 2022), they vied for support from farmers, the country’s largest employment population. The party that was better able to mobilize support within the group was able to win the elections (Kim, 2017). During the two-party phase, regionalism has played an even more pronounced role, with the East voting for PF and the West for UPND (Goldring and Wahman, 2018). During this period, MPs have been elected in a first-past-the-post single-member district system. Until 2016, the country was divided in 150 constituencies. After 2016, the number of constituen- cies increased to 156. Electoral contests are often plagued by poor organization and systematic manipulation (Goldring and Wahman, 2016). Since 2006, party nominations take place through a hybrid system where local party representatives are consulted, but the central party committee makes the final decisions. Nomination violence has been documented during these procedures (Goldring and Wahman, 2018). Party institutionalization in Zambia was initially very low. During the MMD era, party frag- mentation and defections were extremely commonplace (Rakner and SvÂsand, 2004). Since then, the parties’ organizational capacity has expanded markedly towards the periphery. However, in ar- eas where their networks are less developed, they rely on recruiting local elites in order to represent them in parliament(Goldring and Wahman, 2018). Such arrangements are tenuous, as both sides are keen on moving on to a new partnership if it seems to improve their individual prospects. Overall, the Zambian parliament is generally weak compared to the executive, with no indepen- dent budget. The executive also often restricted the time that the parliament was in session. The president was able to control the agenda using ministerial appointments as currency. Nonetheless, 40 the parliament was successful in blocking MMD’s Frederick Chiluba from running for a third term. By invoking the Speaker’s powers, Legislators called the parliament to session outright and with the threat of impeachment forced the president to renounce his intentions to run for a third term (Opalo, 2019). Access to the floor is independent of seniority, and it is minimally constrained by the party elites. As constituted by the Standing Orders, the institutional setting of the legislative debates in general facilitates a back-and-forth between MPs from different parties as well as between ministers and MPs (also see Wegmann and Evequoz, 2019). 3.2.5 Zimbabwe Zimbabwe is considered one of the more repressive states in sub-Saharan Africa. Despite having undergone six electoral cycles since 1990, there has been no change in government. Robert Mugabe, who ruled the country since 1980, easily won the inaugural electoral contest as the head of the Zimbabwe African National Union-Patriotic Front (ZANU-PF). In 1996 Mugabe got reelected with virtually no opposition. In the 2002 election, Mugabe faced stiff competition from Morgan Tsvangirai, the presidential candidate for the Movement for Democratic Change (MDC) (LeBas, 2006). While the outcome was decided in Mugabe’s favor, the accusations that the election was stolen were widespread both nationally and internationally. Regardless, Mugabe went on to win the elections in 2008 and 2013 against MDC-T, the stronger opposition party that emerged after the MDC split. Following a party coup, Mugabe was forced to resign and was replaced by Emmerson Mnangagwa, who went on to win the 2018 elections against the MDC Alliance. ZANU-PF has been a dominant force in Zimbabwe since the institutionalization of multiparty elections. It retained control of the presidency through the systematic and unapologetic use of violent tactics (Frantzeskakis and Park, 2022; Ruhanya, 2020). Nonetheless, the outcome of electoral contests was in several cases far from predetermined, especially at the sub-national level (Levitsky and Way, 2010). For example, after the 2008 elections, the opposition had control of parliament (Azevedo, 2009). In the post-2000 period, ZANU-PF drew support from the countryside while most of MDC’s strongholds were located in urban areas (Friesen, 2023). For the entirety of the period, MPs have been elected in a first-past-the-post single-member 41 district system. Until 2007, there were 120 constituencies. Since 2007, the number of constituencies increased to 210. In addition, 60 women are elected in reserved seats. The reserved seats are attributed following a proportional system from ten six-member constituencies that correspond to the provinces of the country. Aside from the 270 members of the National Assembly, since 2005, Zimbabwe also has had a Senate of eighty members, sixty of which are elected following the same PR system. From the remaining twenty seats, eighteen are reserved for chiefs, and two for people with disabilities. Since 2005, the ZANU-PF has nominated candidates based on the results of party primaries open to all of its members. On the other hand, MDC holds mass primaries only in constituencies where incumbents are not confirmed by the constituency’s party elites, and for constituencies that are not held by the party (Warren, 2022). During these processes, intra-party violence is quite pronounced (Reeder and Seeberg, 2018). Party institutionalization has been fairly high in Zimbabwe. ZANU-PF had a head start in terms of developing local structures during the one-party era. Its networks throughout the countryside are some of the most extensive in the region (Herbst, 2000; Levitsky and Way, 2010). Opposition parties that lacked such structures in the earlier period were not able to compete with ZANU-PF in the period immediately after the introduction of multi-party elections (LeBas, 2006). This high bar of entry forced the opposition forces that founded MDC to create a solid institutional foundation very early on after its creation (Riedl, 2014). The Zimbabwean parliament is quite constrained by the executive. The president is able to dictate policy, and the legislature does not have the power to enforce full transparency and accountability (Doro and Kufakurinani, 2018). When MPs try to resist the executive, they tend to quickly be whipped into line. However, backbencher MPs are active in debates that do not go against main government lines (Kriger, 2006), and minister question time tends to be the most active part of the debates (Compagnon, 2011). 3.2.6 Uganda Uganda is the most authoritarian of the states included in the sample. The first contested elections in the country took place in 1996. An important distinction between Uganda and the 42 other countries in the region is that in Uganda a party ban was in place during the first presidential elections. As a result, all candidates ran as independents. Yoweri Museveni, who had been in power since 1986, won the election in a dominant fashion. The results of the 2001 elections were equally lopsided. In 2005, the party ban was lifted. Following that, presidential candidates ran in the 2006 election as party nominees. At the head of the National Resistance Movement (NRM), Museveni easily won the 2006 elections. He also won the elections that followed in 2011, 2016, and 2021. Museveni’s worst outing came in the last elections when he received 58.4% of the vote. His main opponent in every election from 2001 until 2016 was Kizza Besigye. Starting from 2006, Besigye represented the Forum for Democratic Change (FDC). In the 2021 election, Museveni’s main challenger was Robert Kyagulanyi Ssentamu – also known as Bobi Wine – who was the nominee for the National Unity Platform (NUP). As the party backing President Museveni in the legislature, the NRM has been dominating the political scene in Uganda. It was originally founded as a liberation movement. When it took power in 1986, it abolished parties with the preface that they would lead to sectarianism and they would entrench ethnic politics. As a result, presidential nominees and hopeful MPs ran as independents. While the NRM is dominant in most of the country, opposition support tends to be concentrated in the urban centers and is especially strong among younger voters. Nonetheless, the opposition remains fragmented, and efforts to present a united front have been unsuccessful (Abrahamsen and Bareebe, 2016). All main parties emphasize development, but the main cleavage is pro vs against the incumbent regime (Abrahamsen and Bareebe, 2016). Despite the Movement system, ethnicity also remained an important political cleavage (Conroy-Krutz, 2013). Since the introduction of elections, the electoral system has remained first-past-the-post with single-member districts (SMD). However, the Ugandan parliament is very large – and constantly expanding – compared to that of other countries in the region. In addition, it includes a number of seats that are filled by indirect election or by appointment. The country is divided into districts, which are themselves divided into constituencies. While each constituency elects one MP, each district also elects one Woman Representative. In addition, seats are reserved for five Worker’s 43 Representatives elected from the Labour Unions, five Representatives for People with Disabilities, and five Youth Representatives elected at the national level. Finally, ten Army Representatives are chosen by the Army council. Overall, there were 293 seats in 1996, 305 in 2001, 319 in 2006, 375 in 2011, 426 in 2016, and 529 in 2021. The NRM instituted primaries starting before the 2006 elections. In the first cycle, the selectorate was limited to the local party executives. Starting in 2011, all party members could vote. Regardless of format, fraud has been widespread (Collord, 2016; Kjær and Katusiimeh, 2021). The party ban inhibited the institutionalization of opposition parties in Uganda but also of the NRM itself (Izama and Wilkerson, 2011). This legacy made it difficult for parties to develop their organizational power even in the post-2005 period (Collord, 2016). Candidates for the most part fund their own campaigns, both for the primaries and the general elections. In NRM strongholds, candidates compete for the nomination, but those who lose are likely to run in the general election as NRM-leaning independents (Collord, 2016). This dynamic creates a parliamentary body that is loosely controlled by party elites. Overall, Uganda’s parliament is more active than might be expected given the levels of democ- racy in the country. During the no-party period, MPs developed a sense of independence from the executive. This legacy survived the introduction of multi-partyism. In a parliament that is routinely dominated by the NRM in terms of sheer numbers, MPs are known to challenge the executive both in terms of policy initiatives and during ministerial questions time. Museveni is usually able to rein in the parliamentarians, but frequently only after at least, some concessions are made (Collord, 2016, 2021). Nonetheless, the parliament was not able to block Museveni from amending the constitution in order to extend his time in office. 3.3 Conclusion In this chapter, I developed the rationale based on which cases were selected for the observational analysis, the results of which are discussed in Chapters 4 and 6, and for the experimental part of the analysis, which is presented in Chapter 5. The main criteria considered for inclusion in the observational dataset were primarily related to concerns about the generalizability of the findings 44 and of data availability. Similarly, Malawi was chosen as the setting for the experiment because it exhibits the characteristics of a typical case among the set of cases included in the observational dataset. In addition, the profile of each country included in the dataset was discussed. Each profile provided information on five key axes: recent political history, institutional setup, parties and party system institutionalization, and strength of parliament. Each axis offers important insights related to the theoretical story developed in the previous chapter. Taken together, these axes showcase the diversity of the set of countries included in the sample. Having laid the theoretical foundation, and after providing some background information for the context under study, I proceed to empirically test the hypotheses stated in Chapter 2. 45 CHAPTER 4 LOCAL REPRESENTATION AND REELECTION In Chapter 2, I argued that voters will favor MPs that are active in representing their constituencies in parliament. Citizens expect that their representatives will take notice of their concerns, take them to the government, and help address them. In contexts of low state capacity and non-nationalized political systems, representation expectations are highly concentrated on the local level. Voters expect their MP to represent them by focusing on the constituency’s particular needs rather than on national-level issues. In this chapter, I test hypothesis 1, which states that “MPs who focus on constituency representation in parliament are more likely to be reelected." I do this using an original cross-national dataset covering the parliamentary speeches of MPs in six countries for an extended period of time. Using these cross-national data, I examine the aggregate-level implications of the theory discussed in the previous chapter. The first such implication is that MPs who provide more local representation should be more likely to be reelected. At this stage, it is not possible to test how the provision of local representation affects vote choice. This is examined in more detail in Chapter Four with the use of an experimental design. I begin by laying out the research design for this part of the analysis. First, I discuss the operationalization of local representation. Then, I go through the steps of identifying constituency- oriented speeches, and I briefly discuss the computational methods involved in creating the dataset. Overall, the dataset covers parliamentary debates in Botswana (2014-2019), Ghana (2005-2021), Malawi (1999-2019), Uganda (2011-2021), Zambia (2001-2021), and Zimbabwe (2008-2018), and it includes about 960,000 speeches made by 3,800 MPs. In the following sections, I provide an overview of the dataset, and I go over the model specification and the control variables. Having laid this groundwork, I then present the results of the main analysis. Following that, I report the results of additional robustness tests. Finally, I summarize the conclusions. 46 4.1 Research Design 4.1.1 Measuring Local Representation As defined in Chapter 2, representation is conceptualized as realized in action. Representation is achieved when representatives make representative claims in the name of the represented. This is key in understanding the importance of parliamentary speeches as an act of representation. Given the complicated nature of the concept, measuring empirically the degree to which MPs conduct local representation is challenging. A variety of different methods have been utilized in the literature. A first set of studies have utilized observational methods (e.g., Fenno, 1986), elite interviews (e.g., Searing, 1994), and surveys (e.g, Farrell and Scully, 2007; Katz, 1997; Scully and Farrell, 2003). While these approaches have merit, they do not directly measure local representation in parliament. Instead, they focus mostly on second-hand accounts regarding an MP’s role orientation. More recently, the literature started to rely on parliamentary questions as a measure of con- stituency focus. While the value of parliamentary questions is up for debate (MacCarthaigh, 2005), they offer a concrete way to identify an MP’s role orientation (Franklin et al., 1993; Martin and Rozenberg, 2012). Conducting content analysis on the questions fielded by an MP offers a solid indicator of whether they prioritize general policy issues or local constituency concerns (Martin, 2011). A major benefit of using parliamentary questions to assess role orientation is that the as- sessment is based on a primary source of data. This source is reliable, as parliamentary questions is a tool available in virtually every parliament (Franklin et al., 1993, p. 1), and they are fielded independently of the party leadership (Martin, 2011). Due to these key advantages, several studies have used this approach to study MP role orientation in a variety of contexts including Europe (Bailer et al., 2022; Borghetto et al., 2020; Chiru, 2018; Kartalis, 2021; Russo, 2021; Zittel et al., 2019), Latin America (Alemán et al., 2018), and Africa (Wegmann and Evequoz, 2019). Despite these advantages, parliamentary questions are but a small segment of the parliamentary debates. Of the main forms of debate, which include discussions on bills, statements, motions, petitions, and committee reports, to name a few, parliamentary questions are but one. As a result, focusing exclusively on them may not be an accurate representation of the degree to which an MP 47 engages in local representation throughout the debates. For example, MPs who initiate legislation due to specific needs of their constituency are highly likely to conduct local representation during time devoted to Bill discussion (Bertelli and Dolan, 2009; Däubler, 2020). With the rise of text analysis as a viable data generation approach in the social sciences, a number of attempts have been made to leverage text-as-data in order to study role orientation in the entirety of the debates (e.g., Alemán et al., 2017; Fernandes et al., 2019; Laver and Benoit, 2002; Proksch and Slapin, 2010; Quinn et al., 2010; J. B. Slapin and Kirkland, 2020). In this manuscript, I follow this latter approach. I utilize state-of-the-art text analysis techniques to develop a dataset on parliamentary speeches where localism is coded at the paragraph level. Overall, the dataset covers six countries: Botswana (2014-2019), Ghana (2005-2021), Malawi (1999-2019), Uganda (2011-2021), Zambia (2001-2021), and Zimbabwe (2008-2018). In total this amounts to about 960,000 speeches made by 3,800 MPs. In the next two sub-sections, I explain how constituency-oriented speeches were identified and the steps I followed to assemble the dataset. 4.1.2 Identifying local representation in parliamentary speeches Parliamentary speeches can vary dramatically in length. Depending on the sort of contribution that the MP is making, they may speak for a few seconds up to more than thirty minutes. A common example of a shorter speech is when MPs make a brief interjection while another MP is formally holding the floor. On the other hand, the longest speeches are related to readings of Bills. For example, when the responsible minister – or MP – introduces the Bill for the first time they hold the floor for an extended period of time. An important implication of the difference in individual speech length is the number of items or topics that an MP can cover in their speech. A very short speech almost always addresses one specific topic. On the other hand, when allowed to take the floor for a greater length of time, MPs may touch on multiple topics even if they are all tangentially related. In order to account for such differences, I begin the analysis of the speeches at the paragraph level. At that level, I first code each speech based on the topic it focuses on. To categorize the speeches into categories, I utilize the widely used Comparative Agendas Project (CAP) coding 48 scheme1 (Baumgartner et al., 2006). A complete list of the set of categories is available in the Appendix (Table A.1). In addition to the original CAP major topics, I add a category for procedural speech. This category includes speech that cannot be coded in any of the substantive topics outlined by CAP. Second, and still at the paragraph level, I code whether, in each speech, MPs engage in constituency representation. The original coding rules come from Martin (2011). However, Martin’s rules were developed to facilitate the coding of questions to the minister. As a result, some fine-tuning was necessary in order to code instances of constituency representation in the speeches made during different kinds of debates. In practical terms, a speech is considered constituency- oriented or “localist" when the MP discusses an issue in relation to their constituency. They may refer to the constituency by name, or the reference may be indirect and insinuated after a bigger part of the speech is taken into account.2 In addition, they may bring up a case or event related to their constituency. A key unifying element in these speeches is the sense of belonging or ownership that they exhibit. In the vast majority of speeches that ought to be coded as localist, the MP will say phrases like “my constituency...”, “in X constituency/city/village we...”, “my constituents ...". The fact that MPs speak in such inclusionary terms when talking about their own constituency is picked up by the neural network, and it helps differentiate true localist speeches from speeches where they speak about constituencies represented by other MPs. The outcome of this process is a binary measure of localism at the paragraph level. For reference, I provide two examples of what would be coded as constituency-oriented speech. Mr. Speaker, Fulani herdsmen are also on the increase. I am suffering from the activity of Fulani herdsmen in my constituency. If we go to Agogo, it happened there. So many areas, especially from Brong Ahafo to the north, east, and west of this country. They are causing the situation where we could confidently say that it would bring hunger 1I use the major CAP categories, which include 23 topics. These topics range from Macroeconomics, and Agriculture, to Public Lands Management, and Cultural Policy. 2This is a key reason for the selection of the paragraph rather than the sentence as the coding unit. 49 because they graze on farms and the farm crops are being eaten by these cattle. (Derek Oduro, NPP, Nkoranza North, Ghana, 5.26.2016) This segment was part of Hon. Oduro’s contribution to a statement made by another MP in the parliament of Ghana. The original statement touched on issues of personal safety, highway robberies, and grazing rights, and also contained elements of constituency representation for the speaker’s own constituency. Next, I provide a quote from a speech that was made in the parliament of Zambia. Does it take heavenly intervention for the Ministry of Transport, Works and Supply and Communication to put speed humps or something that can slow down motorists? For instance, every weekend we lose about eight lives at Zani Muone at 13 miles in Katuba. Do we have to come to your office with this information or do you already have the statistics at your fingertips?” (Jonas Shakafuswa, Katuba Central, UPND, Zambia, 06.16.2015) This quote comes from a follow-up question during ministerial questions time in the parliament of Zambia. The original question was made by another MP whose constituency also suffered from a large number of pedestrian fatalities. In this instance, as in the previous example, multiple MPs take advantage of an opportunity to provide constituency representation by bringing up important problems in their constituency. Notably, these opportunities were created on the spot. In both cases, other MPs made the initial statement and submitted the original question. Regardless, MPs who take such opportunities when they appear are lauded at home. The paragraph-level hand coding was done by a team of research assistants. Each Hansard was assigned to two coders. Each coder coded the entire Hansard independently. The team held weekly meetings during which we examined all disagreements on the coding of each Hansard. This process allowed us to identify potential pitfalls and, if needed, to establish more specialized and specific coding rules to be followed by all coders. Also, it ensured that all research assistants coded speeches in a uniform manner, and it ascertained the quality of the hand-coded data. At this stage, the format of the data is very fine-grained, as the dataset is constructed at the paragraph-MP-country level. In the following section, I discuss the computational methods involved 50 in leveraging the set of hand-coded speeches to code the entirety of available speeches across the six countries included in the sample. 4.1.3 Coding Parliamentary Speeches using a Neural Network Approach Given the volume of speeches, I used a neural network approach to facilitate the coding of the entire corpus. Since for both coding steps -i.e. topic and constituency orientation- the set of categories was pre-defined, a supervised learning approach was preferred. In recent years, there has been an uptick in the use of supervised machine learning techniques for text analysis tasks in political science applications. Examples of widely used approaches include the use of support vector machines (eg. D’Orazio et al., 2014) and random forests (Montgomery and Olivella, 2018). Artificial neural networks, however, are a rather novel addition that offers important advantages to other alternative approaches. Neural networks have been shown to be highly effective in Natural Language Processing (NLP) tasks (Popov, 2018) and are particularly well suited for text classification (Mirończuk and Protasiewicz, 2018). Neural networks are a form of machine learning in which a computer learns to perform some task by analyzing training examples. They are organized into layers of nodes with each node being connected to one or several nodes located in the previous and subsequent layers, hence creating the architecture of the network. Adding layers increases the depth of the network. The data pass through the layers in succession via the nodes. Each node assigns a number or “weight” to each of its incoming connections. When data are passed through these connections, it is multiplied by the associated weight. Upon arrival at the node, if the sum product of these weights surpasses a certain value, the data are fed forward to the outgoing connections at the next layer. Data are fed to the input layer and make their way to the output layer, where they arrive completely transformed. During training, the network is also fed with the desired label/output. As the data make their way through the network, the weights, which are initialized with random values, are adjusted in a way that data with the same labels consistently yield the same output (Haykin, 1999). While this field is expanding and developing with astonishing speed, all neural networks follow the logic described above. Recent contributions to this literature have demonstrated the usefulness of transfer learning. 51 This refers to the process of retraining and fine-tuning the weights of an already-trained network in order to perform a specialized task, like coding parliamentary speeches based on the Comparative Agendas Project typology. Transfer learning has been shown to drastically improve classification accuracy, especially when the available hand-coded training data are limited (Ballard et al., 2022; Howard and Ruder, 2018). Therefore, instead of training a neural network from scratch, I used hand-coded training data to fine-tune the pre-trained network BERT (Bidirectional Encoder Representations from Transform- ers), which is made available by Google (Devlin et al., 2018). In simple terms, BERT is capable of “comprehending” a text in terms of sentences rather than simply in terms of words. As a result, it is better equipped to identify words, phrases, or other kinds of sequences that are used most often when someone is speaking about a specific topic. By associating different speech characteristics with each topic, it can then classify new texts into one of these categories. In addition, due to the use of word embeddings, the network’s learning through the provided training data is not limited to the words used within these examples. Due to the existence of these word vectors that accompany each word, it generalizes the aforementioned associations to the words semantically close to the words it meets in the training examples. To summarize, due to its ability to read the text in terms of sentences, through the attention mechanism, and to generalize to a wider vocabulary, through the use of word embeddings, BERT is well equipped to handle complex NLP tasks like categorizing parliamentary speeches in pre-selected topic categories. In more technical terms, the model’s main innovation is that its architecture uses a multi-layer bidirectional Transformer encoder. This means that BERT is able to employ self-attention in a text sequence in a way that combines left-to-right and right-to-left training. In the process of learning, the network looks at the whole input before deciding how much attention to pay to each word, rather than only considering the words that came before the current word. This is done through the employment of a “masked language model” pre-training objective. Given a training text, the model masks a number of randomly chosen words, and the task is to predict their original value based on their context. Also, BERT is trained on “next sentence prediction.” Here the task is to predict 52 which sentence is more likely to follow a given sentence. This kind of pre-training makes BERT highly effective in several NLP tasks, one of which is text classification. BERT uses WordPiece (Wu et al., 2017), word embeddings with a 30,000-token vocabulary. BERT is currently being widely utilized in the social sciences for a wide array of text analysis applications. While its use in political science is a recent development, it can lead to the development of enormous data sets in areas of study that are underdeveloped due to a lack of data. For example, in a recent application, we followed this approach to code more than 110,000 speeches from the Malawian parliament (Wahman et al., 2021). I deploy BERT through the ktrain python library (Maiya, 2020). Through ktrain, I use the base version of BERT, which is 12 layers deep, with a hidden-layer size of 768 nodes, 12 self-attention heads, and 110 million total parameters. While using BERT offers tremendous advantages for classifying complex bodies of text, it requires a substantial amount of data in order to be re-trained to perform a specific task. To this end, I assembled a training set of tens of thousands of speeches from Ghana, Malawi, Zambia, and Zimbabwe. First, I used the hand-coded data from Ghana to train the network. To determine the viability of this approach, I employed 10-fold cross-validation. After being trained for three epochs, the classifier coded the left-out fold. On average, the network coded the policy topic of the test data with an average f1 score of 83.7% (precision: 84.1%, recall: 83.8%) and constituency focus with an f1 score of 95.3% (precision: 96%, recall: 96.1%). In both cases, the variation in predictive performance across folds was minimal. Performance statistics are provided in Tables A.3 and A.4. The number of hand-coded speeches per country and the period covered by each sub-sample is presented in Table A.2. Then, the data from the other three countries were used to validate the results of the training and to ensure that the coding quality was not compromised due to the neural network being trained on data from a specific country or year. Indeed, the network coded localism in Malawi with an f1 score of 90.5% (precision: 95.3%, recall: 92.8%), Zambia with an f1 score of 91.4% (precision: 93.7%, recall: 94.2%), and Zimbabwe with an f1 score of % 92.3 (precision: 92.6%, recall: 93%). The results of these tests demonstrate the generalizability of the training across years and countries. Having completed these steps, I used this neural net to 53 code the entire corpus. 4.1.4 Overview of the Dataset and Main Variables In this section, I begin by discussing the final steps of processing taken to prepare the dataset for analysis. This includes bringing it to the appropriate unit of analysis and incorporating additional variables necessary for the statistical analysis. Then, I present the main dependent and independent variables. In the paragraph-MP-country format, the dataset is not conducive to the type of statistical analysis necessary to test the stated hypotheses. In order to bring it to its final form for analysis, two more processing steps were performed. First, the dataset was collapsed into a speech-MP-country format. As a single speech, I code the sequence of paragraphs by the same speaker that is coded as being on the same policy topic – as opposed to the entire time that an MP had the floor uninterrupted. Therefore, an MP who took the floor for an extended period of time and talked about multiple topics is credited with multiple speeches. This is appropriate due to the large discrepancy in the length of speeches, the contributions that they make on the floor, and their ability to garner attention outside the chamber. At this stage, a speech is coded as constituency-oriented if a single paragraph contained in it was originally coded as constituency-oriented. In other words, the variable still has a binary form, with constituency-oriented speeches coded as 1 and non-constituency-oriented speeches coded as 0. In the final processing step, the dataset is further collapsed into a parliament-MP-country format, where each parliament –or parliamentary term – is the period between two electoral contests. At this stage, the measure of local representation is retained as a count variable, and it denotes the sum of constituency-oriented speeches made by an MP in the entirety of the parliamentary term. General information regarding the dataset’s coverage is provided in Table 4.1. At this point, additional election data are added to the dataset. These include information on the constituency in which each MP was elected, their party, whether they were elected for the first time, whether they were reelected in the following term, whether they are given a ministerial portfolio during the parliamentary term, and whether they are elected in some kind of reserved seat. Of those, the relevant control variables 54 are discussed at greater length in the next section. Table 4.1: Dataset summary Period Number of Number Country Covered Parliaments of MPs Botswana 2014-2019 1 63 Ghana 2005-2021 4 1027 Malawi 1999-2019 4 750 Uganda 2011-2021 2 829 Zambia 2001-2021 4 668 Zimbabwe 2008-2018 2 477 All Countries 1999-2021 17 3814 In order to evaluate Hypothesis 1, that MPs who focus on constituency representation in parlia- ment are more likely to be reelected, in this chapter I focus on the constituency-level implications. Using the variable of localist speeches, I operationalize the main independent variable as the aver- age number of localist speeches made by an MP per day that the parliament was in session. The reason for this additional processing step is that the number of days that the parliament is in session may vary wildly, both between the different countries in the sample and within countries over time. The dependent variable is reelection in the following parliamentary term. This variable is coded based on whether the MP is elected to return to parliament after their current term in office. The coding of the dependent variable is very conservative in that it does not include exceptions for non-reelection due to retirement, death, etc. In all such cases, the variable is coded as 0. This is a conservative coding decision because MPs who provided representation to a degree that should help their reelection bids fail to be reelected for reasons other than electoral defeat. This choice is justified because, in principle, politicians are keen on staying in power, which means that they should be running for reelection. If they fail to run in the general election, in most cases it will be due to unforeseen circumstances that are forced upon them, like being incapacitated due to health problems or death. Retirement could merit special attention, but in a region where most MPs fail to stay in parliament for more than a single term, it looks more likely that most “retirements” take place when the MPs can see early on that their chances of reelection are extremely slim. Since such 55 a development itself could be due to the fact that they have not been providing local representation, coding retirements differently could lead to potentially biased results. Data on MP reelection are collected from the official report of the election results published by each country’s electoral commission. Summary statistics for the main variables are provided in Table 4.2. Table 4.2: Descriptive Statistics for Main Variables Variable Number of Standard Mean Min Max Name Observations Deviation Localism (per Day) 3,814 0.16 0.25 0 3.35 MP Reelection 3814 0.33 0.47 0 1 4.1.5 Model Specification and Control Variables To test the hypothesis, since the dependent variable, reelection is binary, I employ a logit model. I cluster robust standard errors by MP to account for the heteroskedasticity of observations of the same MP in different years. Also, I add country-level fixed effects to account for time-invariant country-specific effects.3 To make the evaluation of the logit coefficients more straightforward, I discuss the results based on simulations that were conducted following the observed values approach (Hanmer and Kalkan, 2013). To account for possible alternative explanations of –non– reelection, I control for several individual-level characteristics of MPs. The descriptive statistics for all control variables are presented in Table 4.3.To begin with, I control for newcomer status. Newcomer MPs are not familiar with the ways of parliament, and their reelection strategies should not be fully developed, or at least they should be untested. This variable is coded based on the official election results published by each country’s electoral commission. In addition, I control for whether the MP was elected in some kind of reserved seat. While most MPs in the sample get elected in SMD seats, some of the countries have reserved seats for women, and in some cases for other groups like people with disabilities, unions, or even direct appointees 3Hausman tests indicate that the differences between the coefficients when using fixed effects and random effects are systematic. 56 of the president. MPs holding such reserved seats are likely to follow different kinds of strategies for their reelection. To account for this, I include a binary control. This variable is coded off of the official election results. I include another binary control for gender. It is widely documented that gender plays an important role in contesting a legislative seat (Schwindt-Bayer, 2005; Wang and Muriaas, 2019), as well as in retaining one (Wahman and Brooks, 2022; Yildirim et al., 2021). Furthermore, women MPs are known to participate less in the debates overall (Bäck and Debus, 2019; Clayton and Zetterberg, 2021). I coded for female MPs based on the names of the MPs. In case the gender could not be ascertained based on the name, further research was conducted on an individual basis. Furthermore, I control for independent status. Most MPs have a party affiliation, and in some countries, it is not even possible to run as an independent candidate. Nonetheless, in countries like Malawi and Uganda, there is a substantial number of independent candidates who contest and win seats in parliament. Since they have no official party affiliation, these MPs might be expected to focus even more strongly on their localist mandate On the other hand, they might prioritize getting into one of the existing parties. An MP was coded as an independent if, at the time of the election, they had no party affiliation. The variable was coded based on the official election results. Moreover, I control for government membership. MPs backing the executive are more likely to be judged based on the government’s performance. Thus, government MPs are expected to support government policy and refrain from actions that would undermine the executive. Therefore, they should be less likely to discuss local problems at length, so as not to be seen as blaming their own party. In addition, government MPs should be better able to navigate back channels and arrange private meetings with ministers to address problems and concerns out of the spotlight of the parliamentary floor. An MP was coded as a member of the government if, at the time of the election, their party affiliation was that of the party backing the executive. The variable was coded based on the official election results. Also, I include a control for ministerial appointments. Ministers are viewed differently than backbencher MPs. They are rightfully viewed by their constituents as more powerful and influential. 57 They are also tightly associated with the president, and they have overwhelmingly more access to resources compared to the average MP. Therefore, their reelection is much more likely to depend on the performance of the government, as well as on the degree to which they funneled resources to their home constituency. This is a binary indicator, and it is coded 1 if the MP served as a minister at any point during the term. The variable was coded based on WhoGov (Nyrup and Bramwell, 2020). Additionally, I control for the number of parliamentary seats in the chamber. The size of the parliament could be another bottleneck for MPs that would like to take the floor. Therefore, the opportunities for speaking should decrease as the number of MPs in the chamber becomes larger. This variable is coded as the number of MPs elected for a particular parliamentary term. Finally, I control for the number of speeches per day that an MP makes, regardless of focus. This accounts for the alternative explanation that speaking in parliament helps an MP’s reelection chances by simply bringing them in the spotlight. Based on this explanation, the contents of the speech are not as important as the fact that the MP receives public attention. As the MP garners name recognition, they could be a more compelling choice for voters at the polls. Table 4.3: Descriptive Statistics for Control Variables Variable Number of Standard Mean Min Max Name Observations Deviation Newcomer MP 3,814 0.65 0,48 0 1 Female MP 3,814 0.19 0.39 0 1 Reserved Seat 3,814 0.10 0.39 0 1 Government MP 3,814 0.53 0.5 0 1 Independent MP 3,814 0.06 0.24 0 1 Minister 3,814 0.10 0.30 0 1 Speeches (per day) 3,814 0.084 1.56 0 30.22 Parliamentary Seats 3,814 258 95 63 448 Parliament 3,814 2010 5 1999 2017 4.2 Results In this section, I analyze the relationship between constituency representation in parliament and MP reelection in a cross-national setting over an extended period of time. 58 Figure 4.1: Localist speeches and reelection Botswana Ghana Malawi 0.100 0.20 0.75 0.075 0.15 0.50 Number of speeches in a session 0.050 0.10 0.25 0.025 0.05 0.00 0.000 0.00 Uganda Zambia Zimbabwe 0.4 0.15 0.075 0.3 0.10 0.050 0.2 0.05 0.025 0.1 0.00 0.0 0.000 Not Reelected Not Reelected Not Reelected Reelected Reelected Reelected I begin the analysis by descriptively exploring the relationship between representation and reelection in Figure 4.1. In each panel, a bar plot summarizes the number of localist speeches per day made by MPs. The lighter column represents the averages of MPs who were reelected in the following elections, and the darker column represents the averages of MPs who were not reelected. Based on these plots, in all countries, MPs who were reelected gave on average considerably more localist speeches per day. Across the entire sample, MPs who were reelected made on average 0.20 constituency-oriented speeches per day during their tenure. On the contrary, MPs who were not reelected made 0.14 speeches. This difference of 0.6 speeches per day represents a 30% drop. 59 This descriptive finding indicates that providing local representation in parliament may indeed be an important predictor of reelection across countries in the region. I further explore this relationship in Table 4.4 (Model 1). The coefficient for localist speeches is positive and statistically significant, which supports the expectation that MPs who make more constituency-oriented speeches will be more likely to be reelected. Critically, this is despite controlling for the number of all speeches that the MPs made per day. This alleviates possible concerns that while representation might be important, voters may only have a sense of overall effort or that their MP is frequently in the spotlight. Contrary to such claims, local representation appears to have a much more pronounced effect compared to overall participation in the debates. Table 4.4: Constituency Representation and Reelection Model 1 Localism 1.18*** (0.22) Speeches 0.11** (0.04) Newcomer 0.47*** (0.08) Special Seat -0.10 (0.18) Female 0.00 (0.11) Minister 0.37** (0.13) Government -0.16+ (0.08) Independent -0.10 (0.18) Parliamentary Seats 0.00 (0.00) Num.Obs. 3814 BIC 4554.5 RMSE 0.44 + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 To assess whether these results are the artifact of outliers in the data, I calculate Studentized 60 residuals to detect outliers, hat values to detect leverage points, and Cook’s distances to detect influential data points. This sensitivity analysis showed that the results were not driven by influential observations (Figure A.1). In addition to parliamentary speaking, several other variables appear to affect the likelihood of reelection. Not surprisingly, MPs holding a ministerial portfolio are more likely to be reelected. Also, newcomer MPs are more likely to be reelected compared to veteran MPs. This may be due to the overall difficulty of achieving reelection in the countries in the sample. However, it also signifies that finding a formula for reelection is not straightforward. On the other hand, government MPs are less likely than opposition MPs to be reelected, albeit the relationship is only statistically significant at the 90% confidence intervals. This is consistent with accounts of government parties strategically shuffling the candidates they field (Demarest, 2021; Warren, 2019) as well as with accounts of increased turnover during primary elections for highly coveted seats (Ichino and Nathan, 2012). Finally, an MP’s gender and being elected in a special seat do not appear to affect the likelihood of reelection. To better gauge the substantive effect of localist speeches on the likelihood of reelection, I conduct simulations following the observed values approach. The top panel of Figure A.1 illustrates the results of the simulation. The rug plot at the bottom of the graph indicates the existence of observations in the sample that take the particular value of the main explanatory variable, i.e. localist speeches per day. According to the graph, representing one’s constituency on the floor of parliament strongly affects fortunes in the coming elections. Based on the simulation, an MP who makes 0 localist speeches per day has a 29% likelihood of being reelected. On the other hand, making 0.5 localist speeches per day increases the likelihood of reelection to 41%, and the few MPs who make 1 speech per day are 55% likely to be reelected. On the bottom panel of Figure A.1, I present the first differences with the comparison category being MPs who make no speeches representing their constituency in parliament. This graph allows for an easier understanding of the degree to which an MP’s likelihood of reelection improves with the number of localist speeches made. Based on the simulations, an MP who makes 0.1 speeches 61 with a localist focus per day is 2.3% more likely to be reelected compared to an MP who never takes the floor to represent their constituency (95% CIs: 1.4–3.3%). Figure 4.2: Representation focus and Reelection probability Probability of Reelection 0.6 0.5 0.4 0.3 0.00 0.25 0.50 0.75 1.00 Change in Probability of Reelection 0.3 0.2 0.1 0.0 0.25 0.50 0.75 1.00 Localist speeches per day To provide a better understanding of what the results mean in the context of the specific countries, I conduct additional simulations. I examine the effect of conducting local representation on the likelihood of reelection by altering the levels of localist speeches made. I run scenarios in which the variable takes the value of the 5th, the median, and the 95th percentile for government and for opposition MPs. While the median MP in the sample makes 0.08 localist speeches per day, MPs in the 5th percentile make 0 speeches per day, and those in the 95th percentile make 0.63 62 localist speeches per day that the parliament is in session. I run separate simulations for Ghana and Malawi. Ghanaian MPs have by far the highest rate of reelection compared to the other countries in the sample (50.2%). On the other hand, the reelection rate for Malawian MPs is on the lower end (27.6%). When calculating the changes in predicted probabilities of an MP getting reelected, binary control variables are held at their mode and continuous variables to their mean. As a result, the “model MP” in these simulations is male, newcomer, party-affiliated –i.e. not independent –, holds an SMD seat, has no ministerial portfolio, and makes 0.84 speeches per day. Table 4.5: Predicted Probability of Reelection Gov. Gov. Gov. Opp. Opp. Opp. 5% Median 95% 5% Median 95% Ghana 49.9% 52.2% 67.9 53.9% 56.2% 71.3% Malawi 21.4% 23% 36.7% 24.3% 26% 40.5% The results of these simulations are presented in Table 4.5. The first takeaway is that the results follow the theoretical expectations in that increasing the number of localist speeches substantially increases the likelihood of reelection in all scenarios. For example, a government MP moving from the 5th to the 50th percentile in the Malawian context increases their likelihood of reelection from 21.4% to 23%. Relative to their earlier odds, this is a 7.5% increase. While considerable, this change is dwarfed by the effect of moving further to the 95th percentile, which leads to a relative increase of an additional 59.6%. Similarly, a Ghanaian opposition MP who is in the 95th percentile is 15.1% more likely to be reelected in absolute terms when compared to his median colleague. In relative terms, this change represents a 26.9% increase. The second key takeaway is that opposition MPs are both more likely to be reelected, and they benefit more by focusing on local representation. For example, the change between a Malawian MP in the 5th percentile and the median MP is 11.1% if they are members of the opposition. This change is 3.6% greater than it would be for government MPs. While I did not explicitly theorize about different effects for government and opposition MPs, this finding fits within the broader theory. Constituency representation often means discussing the problems that are prevalent in the 63 constituency. Government MPs may prefer not to discuss local grievances so as not to paint the executive – and by association, their party and the party leadership – in a negative light. On the other hand, opposition MPs are much less likely to be hindered by such considerations. Furthermore, grievances are likely to be larger in opposition areas and voters will be particularly keen for their MP to articulate regional dissatisfaction with the central government. 4.2.1 Robustness Tests To test the stability of the findings, I ran additional models using the same specification as the main model but dropping from the sample one country at a time. The output of these models is presented in Table 4.6. In all iterations of the model, the main findings remain consistently positive and statistically significant. Regardless of which country is dropped from the sample, providing constituency representation remains an important predictor of reelection. That being said, in some of the models the magnitude of the effect changes substantially. In the models where Botswana, Uganda, or Zimbabwe are dropped the coefficient takes values around 1.2. When Ghana or Zambia are dropped, the coefficient decreases to 0.93 and 0.77 respectively. On the other hand, when Malawi is dropped, the coefficient increases to 1.98. Overall, these results signify that no single country drives the findings. They also indicate that providing constituency representation may be considerably more important for reelection in some countries compared to others. To further illustrate these findings, in Figure 4.3, I present the results of simulations of the first differences corresponding to each of the additional models. The baseline category is an MP making 0 localist speeches per day. 64 Table 4.6: Constituency Representation and Reelection with Country Dropout Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Drop Drop Drop Drop Drop Drop Botswana Ghana Malawi Uganda Zambia Zimbabwe Localism 1.27*** 0.93*** 1.98*** 1.24*** 0.77** 1.13*** (0.23) (0.24) (0.31) (0.23) (0.25) (0.22) Speeches 0.11** 0.18*** 0.03 0.12** 0.11** 0.11** (0.04) (0.05) (0.05) (0.04) (0.04) (0.04) Newcomer 0.47*** 0.15 0.46*** 0.65*** 0.44*** 0.55*** (0.08) (0.10) (0.09) (0.09) (0.09) (0.09) Special Seat -0.12 -0.09 -0.09 0.22 -0.03 -0.35+ (0.18) (0.19) (0.19) (0.27) (0.19) (0.21) Female 0.02 0.00 0.01 -0.03 0.01 -0.04 (0.11) (0.13) (0.13) (0.12) (0.13) (0.12) Minister 0.39** 0.13 0.31* 0.47*** 0.56*** 0.34* (0.13) (0.16) (0.14) (0.13) (0.15) (0.14) Government -0.17* 0.08 -0.11 -0.30*** -0.14 -0.26** (0.08) (0.10) (0.09) (0.09) (0.09) (0.09) Independent -0.10 0.08 -0.14 -0.14 -0.05 -0.17 (0.18) (0.19) (0.27) (0.24) (0.18) (0.19) Parliament -0.01 0.01 -0.02 -0.01 -0.02* -0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Botswana -15.65 26.88 18.82 40.73* 18.92 (21.41) (20.09) (17.70) (19.98) (17.36) Ghana 23.09 29.83 21.43 42.89* 21.36 (17.01) (20.06) (17.69) (19.97) (17.35) Malawi 21.78 -14.34 20.09 41.60* 20.06 (16.97) (21.34) (17.65) (19.91) (17.30) Uganda 21.45 -14.80 28.10 41.30* 19.79 (17.02) (21.41) (20.08) (19.98) (17.36) Zambia 21.83 -14.25 28.43 20.11 20.12 (16.98) (21.36) (20.03) (17.66) (17.32) Zimbabwe 22.19 -13.98 28.90 20.45 41.99* (17.01) (21.39) (20.05) (17.69) (19.96) Num.Obs. 3752 2787 3064 2984 3146 3337 BIC 4471.3 3165.6 3658.5 3737.2 3758.0 3950.6 + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 65 Figure 4.3: Representation focus and Reelection probability with Country Dropout Drop Botswana Drop Ghana 0.3 0.20 0.15 0.2 Change in Probability of Reelection 0.10 0.1 0.05 0.0 0.00 0.25 0.50 0.75 1.00 0.25 0.50 0.75 1.00 Drop Malawi Drop Uganda 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 0.25 0.50 0.75 1.00 0.25 0.50 0.75 1.00 Drop Zambia Drop Zimbabwe 0.3 0.20 0.15 0.2 0.10 0.1 0.05 0.00 0.0 0.25 0.50 0.75 1.00 0.25 0.50 0.75 1.00 Localist speeches per day Localist speeches per day 66 4.3 Conclusion Taken together, the results in this chapter show that MPs who focus more on constituency representation are more likely to be reelected. This trend is clear in the overall sample as well as in the individual countries. In addition, the effect of providing local representation in parliament is shown to be distinct from the effect of overall effort and participation in the debates, and much greater in magnitude. At this point, it is necessary to address concerns regarding two possible alternative explanations that could be driving the findings. The first alternative explanation is that MPs might be more likely to discuss local issues when credit-claiming for local public goods that have already been distributed or are soon to be distributed (Grimmer et al., 2012). In this case, both the speeches and the improved reelection prospects would be driven by the distribution of goods. Indeed, it is not unusual for MPs to take the floor to talk about goals that have been achieved, and, potentially, to thank the responsible minister, the relative parliamentary committee, or other people that contributed. However, it is rarely the case that they will take this step unless they have brought up the particular issue in parliament and requested that it is resolved. In addition, the utility of credit-claiming in such a way within parliament seems limited. Speaking about unsolved issues signals to the voters that the MP cares about them and tries to resolve them. In this instance, the key for the MP is to show that these issues are brought to the center. On the other hand, credit-claiming should be more effective when taking place in the constituency itself. MPs in most African countries are expected to spend much of their time in the constituency. Time, when they are among the voters or touring the constituency, should provide superior opportunities to claim credit for what they achieved. Therefore, it seems unlikely that the findings presented here are driven by speeches of credit claiming. A second alternative explanation is that MPs eschew discussing local issues when they believe they are in trouble. For example, it is possible that they reduce their attendance at parliamentary sessions when they are facing worse electoral prospects because they want to be visible at home. In this case, the low number of speeches itself could be due to the expectation of poor electoral 67 performance. This explanation is unlikely for two reasons. First, prolonged stays in the constituency can be very expensive for MPs. When MPs are present in their constituency, they are expected to be available to their constituents. Most constituents seeking the MP do so because they need some sort of assistance. As discussed in earlier chapters, most MPs are limited in the number of requests they can satisfy, which means that a number of petitioners would not be satisfied by the outcome of their meeting, Therefore, being visible in the constituency could further harm the chances of struggling MPs. Second, existing literature indicates that voters are not upset with MPs being in the capital when the parliament is in session, especially since in most countries in the region parliamentary sessions are relatively short. Voters become bitter when the MP is absent even when the parliament is not in session (e.g., Bratton et al., 2005). This issue also came up in some of my interviews. One deputy minister in particular explained that voters generally accepted that she needed to be in the capital during the parliamentary sessions. However, it was harder for them to accept that, due to her ministerial appointment, she would need to be in the capital much of the time even if the parliament was in recess. Other MPs talked about this in relation to parliamentary committees. Many of the committees kept meeting even when the parliament was not. However, committee meetings were not covered by the media nearly as much as the parliamentary debates, which in many cases led to voters being unable to justify the absence of the MP from the constituency. Therefore, even for struggling MPs it would make sense to remain in the capital during parliamentary debates, which would provide them the opportunity to take the floor and provide local representation, but they might feel pressured to skip participating in the committee meetings in order to be more visible in their constituencies. Nonetheless, an important shortcoming of the cross-national analysis presented in this chapter is that it does little to capture the individual voter-level processes that power the relationship between local representation and reelection. It does not test whether, as the theory suggests, voter demand for local representation drives individual vote choice. This will be the focus of Chapter 4. 68 CHAPTER 5 DEMAND FOR LOCAL REPRESENTATION In this chapter, I test the individual-level implications of hypothesis 1. Specifically, I explore whether a voter is more likely to vote for an MP if they are more active in representing the constituency at the floor of parliament. To accomplish this, I fielded a conjoint experiment in Malawi.1 In addition, I test whether the respondents were aware of the degree to which their sitting MP was providing local representation in parliament. Malawi offers a suitable but challenging context for this study. To begin with, the party system is very weak, the electoral system is SMD, and the candidate selection procedures are decentralized. These institutional characteristics do not inhibit the theoretical logic developed in the previous sections. On the contrary, they should entice MPs to cultivate a personal vote (Frantzeskakis et al., 2021) in order to increase their chances of reelection (Cain et al., 1987). As a result, Malawi provides an environment in which the accountability linkage discussed in earlier sections could be in effect. However, the cross-country analysis conducted in the previous chapter indicates that Malawi is certainly not the country driving the findings. Looking at the robustness tests, presented in Table 4.6, we see that while all models support the main hypothesis, the results are strongest in Model 4, the model in which Malawi is dropped from the sample. This is also evident in the results of the relevant simulations presented in Figure 4.3. Therefore, the relationship between local representation provision and reelection is not as strong in the case of Malawi as in other countries in the sample despite the conducive institutional environment. Turning to survey evidence, it is evident that Malawians have a rather negative view regarding their MP’s focus on representation. The Afrobarometer includes a question on whether, according to the respondent, their MP "[tries] their best to listen to what people like [the respondent] have to say." Looking at the results for Malawi in the 7th round of the survey, we see that 61% of the respondents said Never, 28% said Only Sometimes, 6% said Often, and only 4% said Always. These 1Both the experimental design and the empirical approach were pre-registered with EGAP. 69 numbers are considerably lower than the regional averages.2 Thinking back on the discussion on the concept of representation in Chapter 2, learning what the represented care about is an integral part of providing local representation. If MPs fail to note their constituents’ concerns, they are less likely to provide local representation in parliament. Also, MP reelection rates in the country are quite low as they tend to be under 30%. Overall, Malawi is a hard case to test how local representation provision affects individual vote choice. The chapter begins by laying out the research design for this part of the analysis. First, I describe in detail the experimental design, and the attributes included in the conjoint experiment. Then, I discuss the model specification, and I present the results of the experimental analysis. To close the chapter, I empirically test the degree to which survey respondents were aware of whether their current MP provided local representation in parliament from the time they were elected until the survey was fielded. Finally, I summarize the conclusions. 5.1 Research Design 5.1.1 Sampling Procedures The experiment was conducted in twelve rural constituencies between March and April 2022. The individual participants were selected after several levels of randomization. The study sample includes a total of six districts in the central and southern regions of Malawi (Ntchisi, Nkhotakota and Salima in the central region, Phalombe, Chiradzulu, and Nsanje in the southern region). In each district, two constituencies were selected, giving a total of twelve constituencies. Table 5.1, shows the complete list of constituencies from which the sample was drawn. Figure 5.1, shows the location of the sampled constituencies on the country’s map. Constituencies included in the sample are colored in orange while constituencies not included are colored in blue. The survey was conducted by the Institute of Public Opinion and Research (IPOR), a research firm which is based in the country. A minimum of 175 interviews were conducted in each constituency. A maximum of seven polling stations were randomly selected in each constituency, 2Looking at the entire sample, 46% of the respondents said Never, 31% said Only Sometimes, 10% said Often, and 6% said Always. 70 Table 5.1: List of Surveyed Constituencies Region District Constituency Central Ntchisi South North Nkhotakota South East North East Salima North Central Southern Phalombe Central South Chiradzulu East South Nsanje North Central Figure 5.1: Map of Sampled Constituencies Constituency in Sample Constituency not in Sample 71 and a minimum of twenty-five interviews were conducted in the catchment area of each polling station. The names of the villages in the catchment area of the polling station were documented and two starting points were decided. Each team had four enumerators, two of which were assigned each starting point and asked to go in opposite directions to start the data collection. This allowed for better coverage of the area. Employing a random walk approach, the enumerators followed a pre-determined skip pattern to identify households to be interviewed. Within selected households, the responsible interviewer employed simple random methods to select the household participant to be interviewed by listing the names of the eligible respondents by gender and age. Strict replacement rules were followed. The sample was stratified on gender. The interviews were conducted in the local languages and responses were recorded with tablets. This allowed for monitoring of the process and identification of potentially problematic entries. For example, interviews whose duration was deemed too short to be reliable were not included in the final sample.3 The final sample consists of over 2200 respondents, and it is representative at the constituency level. Each respondent was asked to complete three voting tasks. Power calculations that were conducted based on the procedures described by Stefanelli and Lukac (2020) indicate that the sample is sufficient to achieve power greater than the traditional threshold of 80% assuming an effect size of 0.04 (Stefanelli and Lukac, 2020). Balance statistics for the composition of the sample are presented in the Appendix, in Table A.5. 5.1.2 Conjoint Attributes and Experimental Treatment The forced-choice conjoint survey experiment was implemented as follows. Survey respon- dents were presented with a short biography of two hypothetical candidates, an incumbent, and a challenger. Note that in all scenarios an incumbent was facing a challenger. There are no scenarios where the respondents were asked to choose between two incumbents or two challengers. Respon- dents were read the following introduction at the beginning of the experiment: “Now, I will tell 3Over the course of the survey, two interviewers, in particular, were identified as completing the interviews much faster than expected. Upon further investigation, it was decided to drop the surveys conducted by them and replace them with new ones that were conducted by other members of the team. 72 you about different hypothetical candidates who could run for parliamentary elections and ask you some questions about them." Each respondent voted in three voting tasks in total, and each time they were asked to pick between an incumbent and a challenger. Aside from the candidate’s status as an incumbent MP or a challenger, the biographies of the candidates included six additional attributes. The attributes were selected on the basis of making the candidates more realistic by providing essential pieces of information respondents may consider critical when making real voting decisions. In addition, the levels of each attribute were designed to be viable and meaningful for the particular context the experiment took place in. In the following paragraphs, I briefly refer to each of them. Table 5.2 displays all attributes and their possible levels. Table 5.2: Attributes and Levels for Conjoint Survey Experiment Attribute Levels Status Incumbent / Challenger Gender Man / Woman Occupation Maize farmer / Major Business owner / Teacher Development Priority Boreholes / Roads / Education Party First party / Second party / Independent Assistance Known to rarely provide assistance to individual constituents / Known to frequently provide assistance to individual constituents Activity in Parliament No reference to parliamentary behavior / (for Incumbents only) Known to participate in debates in parliament / Known to talk about local issues and constituents’ concerns in parliament The first attribute is occupation, and the levels it could take include maize farmer, major business owner, and teacher. This attribute primarily gives two cues. The first is one of descriptive representation. As the majority of Malawians work in the agricultural sector, they might view a maize farmer candidate as being closer to them. On the other hand, since candidates often use their private resources, a candidate who owns a major business might be seen as more likely to disperse resources to constituents both during the campaign and while in office. Finally, teachers are regarded as well educated and respected individuals, but I do not have prior expectations regarding the kind of signal that this level of the attribute could send to the respondents. 73 The second attribute states the candidate’s development priority, which may be boreholes, roads, or education. All three options are prioritized by voters to different extents. As a result, they consist viable choices for candidates on the campaign trail. The third attribute regards the candidate’s party affiliation. The candidate may be running under the colors of the party that got the highest share of the vote the last election, the party that came second, or they may be an independent candidate. This attribute gives clear cues to voters that feel strongly about specific parties, and it may be especially important for voters in stronghold constituencies. The fourth attribute notes each candidate’s gender. However, an additional experiment that was conducted in an earlier part of the same survey administers a treatment related to the gender of candidates. In order to avoid possible contamination of the results due to this prior experimental treatment, the gender attribute was always kept constant within each incumbent/challenger pair. The fifth attribute informs respondents whether the candidate tends to provide assistance to individual constituents. It specifies that a candidate is either frequently or rarely known to provide assistance to individual constituents. This is a more direct cue to providing clientelistic benefits. While the cue is clear, the use of wording that refers directly to the transfer of resources or clientelism is avoided in order to minimize desirability bias in the responses. This attribute is included to control for clientelism and, to a degree, constituency service4 as an alternative explanation. The sixth attribute, which is the main attribute of interest for the purposes of this study, was only presented in the biography of the incumbent and it refers to their parliamentary activity. This attribute has three levels. At the first level, there was no mention of parliamentary activity. This serves as the baseline. At the second level, biographies mention that the MP is known to participate in parliamentary debates in general. This level is meant to signal general effort. Finally, the third- level biographies mention that the MP is known to actively take the floor to engage in constituency representation. This is the main level of interest in terms of studying the effect of representation 4In Malawi constituency service often takes the form of financial assistance directly used to access services like healthcare or education. For example, in the current parliamentary term, the conditions for the use of the CDF were altered to allow MPs to use a specific percentage of it to pay school fees for secondary education. 74 on vote choice. The distinction between the second and third levels of this attribute is a key part of the design. It helps control for the effects of general effort, which voters should like to see and to further isolate the effect of providing local representation. Balance statistics for the attributes are presented in the Appendix, in Table A.6. In the context of a real race, the first four attributes contain information that should be widely available and relatively easy to access. On the other hand, the last two provide information that only someone with significant local knowledge could have. This is critical in terms of determining the baseline category. My choices of baseline category for each of the two attributes are driven by the fact that, ontologically, it is much easier to establish that something happens than to establish for a fact that it does not happen. I expect that voters will have a sufficient level of knowledge to conclude that, to the degree they are concerned, a candidate only rarely assists constituents. For someone to know whether a candidate frequently or rarely assists constituents means that they are in contact with other people in the constituency and that they learn of the behavior of a candidate. While on the campaign trail, candidates try to be as visible as possible. During that time, people reach out to them for help and resources. Word of a candidate’s conduct in such situations spreads easily within the area. Therefore, for the Assistance attribute, I set knowledge of non-assistance as the baseline, and I provide the information that a candidate only rarely assists constituents. On the other hand, I expect that voters are less confident about their knowledge of what takes place in parliament. I argue that when an MP conducts constituency representation in parliament, their constituents take notice, especially when it happens frequently. This kind of information dissemination can happen through various channels as discussed in Chapter 2. However, it would be much more difficult for voters to establish that their MP does not provide constituency represen- tation. Therefore, for the Activity in Parliament attribute, the baseline category is “no reference to parliamentary behavior," which allows respondents to retain their preexisting assumptions regard- ing the average MP’s activity in parliament. As discussed at the beginning of the chapter, in the case of Malawi, the preexisting assumption is likely to be that the MP is unlikely to be focusing on constituency representation. 75 Table 5.3: Example of Candidate Bios Candidate 1 Candidate 2 Incumbent status: Challenger Incumbent status: Incumbent Gender: Man Gender: Man Education: University Education Education: Secondary Education Occupation: Teacher Occupation: Major Business Development priority: Roads Development priority: Boreholes Assistance: known to frequently Assistance: known to rarely provide assistance to individual constituents provide assistance to individual constituents Parliament: - (does not appear) Parliament: known to talk about local issues and constituents’ concerns in parliament Table 5.3 shows an example of a possible scenario that could be presented to a respondent. Each respondent was presented with and asked to “vote" in three different voting tasks. Upon being presented with a pair of biographies, respondents were asked to pick the candidate that they would vote for. They were also asked to pick which candidate they thought others in their constituency would vote for. Including this additional question achieves two goals. The first is that it captures the drivers of vote intentions for the wider community through the eyes of the respondent. The respondent might offer a different response to it if, for example, they think that they are an outlier in the way they think about politics. Second, it offers the respondent the option to give a more honest response that is less likely to be affected by desirability bias. The exact phrasing of the questions and possible answers are presented in Table 5.4. 76 Table 5.4: Voting Task Questions Which of these candidates would you vote for as your MP? a) Candidate 1 b) Candidate 2 c) Do not know [Do not read] Which of these candidates do you think others in your constituency would vote for as their MP? a) Candidate 1 b) Candidate 2 c) Do not know [Do not read] 5.1.3 Dataset Structure and Model Specification The main outcome of interest is whether the respondent themselves would vote for the incumbent (coded as 1) or for the challenger (coded as 0) in each scenario. The level of analysis is the individual scenario. Since each respondent was called to vote in three scenarios, the dataset is comprised of three observations per respondent. To analyze the data, I first estimate the average marginal component effects (AMCE), as is standard for conjoint analysis. These AMCEs indicate the change in the probability that an incumbent receives a vote by the respondent based on an attribute level compared to the reference category for that attribute. The model equation is as follows: 𝑌𝑖 𝑗 = 𝛼 + 𝛽 × 𝑋𝑖 𝑗 + 𝛿 × 𝐹𝑖 + 𝜖 (5.1) In the above, Y represents the vote choice (0 or 1) for the individual i in scenario j. 𝑋𝑖 𝑗 represents the attribute levels for the specific scenario, and 𝛽 represents the vector of the coefficients for the attributes relative to the base category. Since there are three scenarios per individual, the standard errors are clustered by individual. Finally, 𝐹𝑖 represents enumerator and constituency fixed effects for individual i. Since all attributes were independently randomized, their AMCEs will be estimated simultaneously through OLS regression (Hainmueller et al., 2014). Given the binary nature of the main outcome variable, I will also fit a logit model in order to test the stability of the results. Furthermore, I will run additional models following the same 77 specifications but replacing the dependent variable with the response to the question regarding the vote choice of others in the constituency. I expect the direction and magnitude of the effects to remain consistent across all four models. 5.2 Results In Figure 5.2, I present a coefficient plot with the results of the main OLS model based on the data from the conjoint experiment. The results are also presented in table format in the Appendix (Table A.7, Model 1). Looking at the graph, the effects for each attribute are presented by level. Effects bigger than 0 indicate an increase in the likelihood of voting for the incumbent and effects smaller than 0 indicate a decrease. The effects of the attributes included in the incumbent’s bio are presented in the top part of the graph, and the effects of the attributes included in the challenger’s bio are presented in the bottom part of the graph. The results related to the main attribute, parliamentary activity, are located at the very top of the graph. Based on the results, a respondent was 8.4% more likely to vote for incumbents with a representation focus in parliament (95% CIs: 5.5–11.2%). Substantively, this is a very significant boost. For comparison, when asked to choose between an independent and the candidate standing for the party that carried the last election, voters are 5.9% more likely to choose the latter (95% CIs: 3.3–8.9%). The results also indicate that substantively local representation is of greater concern than general activity in parliament, which increases the likelihood of support for an incumbent that was active in parliament by 6% (95% CIs: 2.8–8.6%). A linear hypothesis test indicates that the effect of local representation provision is 2% stronger than that of general activity albeit only at the 90% CIs (0.2–5%). Two additional observations warrant further discussion. The first regards the size of the effect that providing high assistance seems to have to constituents. Based on the model, an incumbent who is known to assist constituents is 24% more likely to receive a vote than one known to avoid it (95% CIs: 21–25.3%). The size of the effect is way stronger than the effect of any other attribute. This is a strong indication that clientelism and constituency service affect vote choice strongly, as we would anticipate from the literature. However, it is worth considering what this means 78 Figure 5.2: Determinants of Voting for the Incumbent: Respondent’s Vote Choice Inc. Active in Parliament Inc. Representation Focus Woman Candidate Inc. 1st Party candidate (vs Independent) Inc. 2nd Party candidate (vs Independent) Inc. Education priority (vs Boreholes) Inc. Roads priority (vs Boreholes) Inc. High Assistance (vs Low) Inc. Major Business owner (vs Farmer) Inc. Teacher (vs Farmer) Cha. 1st Party candidate (vs Independent) Cha. 2nd Party candidate (vs Independent) Cha. Education priority (vs Boreholes) Cha. Roads priority (vs Boreholes) Cha. High Assistance (vs Low) Cha. Major Business owner (vs Farmer) Cha. Teacher (vs Farmer) −0.2 −0.1 0.0 0.1 0.2 AMCEs on Voting for the Incumbent given the existence of the material constraints that were discussed in Chapter 2. The candidate in question, as most real candidates, can only satisfy a proportion of the requests for assistance that they receive. As a result, we can expect that within the constituency, the same candidate will be known to some voters to “frequently provide assistance to individual constituents,” while to others they will be known as “rarely providing assistance to individual constituents.” In other words, in most real electoral contests, the same candidate will be viewed differently by different voters depending on whether the voter’s friends and family have been assisted or denied assistance by the candidate. Therefore, while high assistance leads to a huge boost in the likelihood of receiving a vote, it is virtually impossible to maintain such a reputation across the entire constituency. On the 79 other hand, the benefit that an incumbent receives from a representation focus, which is the second largest among the attributes in the model, is not subject to similar limitations and, thus, might be more reliably acquired and maintained across the constituency. The second observation is that all attributes increase the likelihood of support for the incumbent or the challenger in similar ways. For example, respondents were 5.2% more likely to vote for an incumbent who owns a big business (95% CIs: 2.7–8.2%) and 3.3% more likely to vote for a challenger with a similar occupation (95% CIs: -6.4– -1.1%). Similarly, an incumbent who is known to assist constituents is 24% more likely to receive a vote than one known to avoid it (95% CIs: 21–25.3%). A challenger gets a boost of 20% for having the same attribute (95% CIs: -23.3– -18.7%). Therefore, while all other positive and negative attributes may cancel out to an extent, an incumbent’s access to the floor of parliament can be a critical advantage. These results strongly support H1, that voters are more likely to vote for incumbents known to speak about local issues and constituents’ concerns in parliament. The results remain consistent when the model is specified as a logit instead (Appendix, TableA.7, Model 2). As presented in Figure 5.3, the estimated effects are very similar when the respondent is asked how they think that others in their constituency would vote (see also Appendix, Table A.8). Looking at the main attribute effects, a respondent was 7.7% more likely to say that the constituency would vote for incumbents with a representation focus in parliament (95% CIs: 4.8–11%). Again, the results show that in substantive terms local representation is of greater concern than general activity in parliament, which increases the likelihood of support by 4.7% (95% CIs: 1.8–7.6%). However, based on this model, the difference between general activity and local representation provision is estimated to be 3% and is statistically significant at the 95% CIs (0.4–6%). Overall, regardless of model specification, respondents are significantly more likely to choose the incumbent’s profile if the sitting MP has a local representation focus. 80 Figure 5.3: Determinants of Voting for the Incumbent: Respondent’s Estimation for the Constituency Inc. Active in Parliament Inc. Representation Focus Woman Candidate Inc. 1st Party candidate (vs Independent) Inc. 2nd Party candidate (vs Independent) Inc. Education priority (vs Boreholes) Inc. Roads priority (vs Boreholes) Inc. High Assistance (vs Low) Inc. Major Business owner (vs Farmer) Inc. Teacher (vs Farmer) Cha. 1st Party candidate (vs Independent) Cha. 2nd Party candidate (vs Independent) Cha. Education priority (vs Boreholes) Cha. Roads priority (vs Boreholes) Cha. High Assistance (vs Low) Cha. Major Business owner (vs Farmer) Cha. Teacher (vs Farmer) −0.2 −0.1 0.0 0.1 0.2 Effect on Voting for the Incumbent (AMCE) 5.3 Do Voters Know if their MP Provides Local Representation? A critical condition of the argument proposed in this study, that local representation in parliament is important for vote choice, is that voters are aware of the level of their MP’s focus on constituency representation. Even if voters care about parliamentary representation, they need to be aware of their MP’s behavior in parliament in order to factor this information in their retrospective evaluations. 81 However, there is not much research on this topic in the region, and literature based on data from other regions offers little to no support for such a statement. Therefore, it is important to put this assumption to the test. To facilitate testing this key condition, an additional descriptive component was included in the survey. Respondents were asked to evaluate on a four-point scale how active they believed their current MP has been representing the needs of the constituents in parliament. These responses were matched with the number of constituency-oriented speeches that their MP had made in parliament from the beginning of their term in 2020 until the survey was fielded in May 2022. From the newly compiled dataset, I drop survey responses from Ntchisi North as Jaqueline Chikuta, the MP originally elected there, died from COVID-19 in May 2021. Therefore, the voter knowledge assumption is tested on a sample from the eleven remaining constituencies. Given the structure of this version of the dataset, the unit of analysis is the individual respondent. To test whether more constituency-oriented speeches lead to higher respondent evaluations since the dependent variable is considered continuous, I employ OLS regression. The main independent variable is localist speeches per day. I add interviewer-level fixed effects and cluster the standard errors by constituency. I also control for a set of variables that could affect the respondent’s response to the question of interest. First, I control for co-partisanship between the incumbent and the respondent. When a respondent identifies with the MP’s party, they might be more inclined to say that the MP does a good job representing the constituency. Then, I control for government status and independent status to capture additional effects related to party affiliation or lack thereof. In addition, I control for the MP’s gender, as respondents could anticipate women MPs to be less assertive in the chamber. Finally, I control for overall satisfaction with the MP’s performance. This is measured in the survey with the question “How satisfied are you with your MP?" Responses to this questions were recorded again on a four-point scale. This is an important control as it helps isolate the effect of local representation from that of satisfaction due to other reasons, like the provision of clientelistic benefits. In a secondary model, I also control for the number of speeches per day made by the MP. 82 Descriptive statistics for all variables included in the models are presented in Table 5.5. Table 5.5: Descriptive Statistics for Variables Used to Test Voter Knowledge Mean SD Min Max Voter Evaluation 2.20 1.27 1.00 4.00 Localism 0.25 0.33 0.00 0.98 Speeches 0.39 0.55 0.00 1.64 Co-partisan 0.44 0.50 0.00 1.00 Government 0.28 0.45 0.00 1.00 Independent 0.09 0.28 0.00 1.00 Female MP 0.26 0.44 0.00 1.00 Satisfaction 2.34 1.27 1.00 4.00 5.3.1 Testing Voter Knowledge The results, which are presented in Table 5.6, indicate that voters are aware of their MP’s activity in parliament. Beginning with the results for Model 3, one constituency-oriented speech per day leads to a statistically significant increase of 0.54 points on the four-point scale. This boost is relatively small given how many speeches an MP would need to give to receive it. Nonetheless, it signifies that voters have an awareness of their MP’s behavior in parliament. Looking at the control variables, respondents perceive government party MPs as doing a considerably worse job in terms of constituency representation. This finding resonates with other results presented in Chapter 4. Government MPs tend to be less active in parliament and make fewer demands, possibly to avoid being seen as rebellious. On the other hand, respondents are more likely to say that their MP represents the constituency well when they are more satisfied with the MP’s general performance. In Model 4, the speeches-per-day control is also included. This results in a drastic increase in the effect size for the main independent variable. Based on this model, one localist speech per day leads to an increase of 3.67 points on the 4-point scale. However, this effect is only statistically significant at the 90% CIs. Nonetheless, this effect size maps very well on the data for the overall dataset and on the findings presented in Chapter 4. Only a small number of MPs make one localist speech per day, but those who do have a 55% chance of reelection, compared to a 30% average for 83 the entire dataset. On the other hand, the coefficient for the total number of speeches per day is negative, although not at a statistically significant level. This is an indication that voters consciously judge the content of the speeches made by their MP. Another possible explanation could be that this is the result of information filtering conducted by channels of information dissemination, like local radio stations and WhatsApp channels. Table 5.6: Effect of Parliamentary Representation on Respondents’ Estimations Model 3 Model 4 Localism 0.54*** 3.67+ (0.16) (2.03) Speeches -1.72 (1.12) Co-partisan 0.05 0.05 (0.08) (0.08) Government -0.65* -1.16** (0.25) (0.36) Independent 0.09 -0.18 (0.11) (0.15) Female MP -0.03 -0.30 (0.07) (0.19) Satisfied 0.75*** 0.74*** (0.02) (0.02) Num.Obs. 1912 1912 R2 0.906 0.906 RMSE 0.78 0.78 Std.Errors by: Constituency by: Constituency + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 Overall, the results show that voters are more likely to consider that their MP actively rep- resents the needs of the constituents when they focus more on constituency representation. This finding indicates that voters have a good understanding of the degree to which their MP takes the parliamentary floor to speak on constituency issues. 5.4 Conclusion A central claim of this manuscript is that voters desire their MP to provide local representation. In contexts where state penetration is low, local politics are very salient and voters want to see that 84 their representatives prioritize them when they go to the capital. Therefore, I argued that a citizen will be more likely to vote for a sitting MP who displays a representation focus during their tenure. The analyses presented in this chapter conclude the empirical tests evaluating Hypothesis 1, that MPs who focus on constituency representation in parliament are more likely to be reelected. In this chapter, I tested the individual-level implications of the hypothesis. Individual voters will be more likely to vote for incumbent MPs if they have demonstrated a focus on constituency representation during their tenure. Experimental testing established that this accountability linkage exists in the context of Malawi. Furthermore, by combining survey data with data on parliamentary speaking, I demonstrate that respondents were aware of the degree to which their current MP was taking the floor to discuss constituency matters. Taken together, Chapters 4 and 5 offer strong support for Hypothesis 1. Overall, they provide evidence that Africans care about constituency representation and that they are aware of whether their MP provides it. At the aggregate level, this individual-level dynamic leads to MPs who provide representation being more likely to be reelected. These findings are of critical importance for our understanding of voting behavior in the region. They indicate that voters pay attention to their MP’s behavior and that they are in a position to engage in retrospective evaluations of their performance. This political knowledge may function as the foundation of an accountability mechanism tied to a legislator’s performance, which goes beyond the transactional clientelism narrative that has been the focus of existing literature. However, if constituency representation is such an important determinant of vote choice, the question that remains is why MPs don’t utilize it further. The argument proposed here is that the existence of locally powerful parties complicates the straightforward story presented up to this point. Under circumstances of strong party control at the local level, MPs might prioritize catering to the preferences of party elites than to those of the voters. I proceed to empirically test this part of the argument in Chapter 6. 85 CHAPTER 6 SERVING AT WHOSE PLEASURE? PARTY DOMINANCE AND LOCAL REPRESENTATION IN AFRICAN LEGISLATURES Previous chapters studied the relationship between the provision of local representation and re- election. Individual-level data show that voters are more likely to vote for incumbents if they are known to provide local representation. In addition, there is evidence to suggest that voters have a sense of the degree to which their MP focuses on providing local representation. Furthermore, cross-country data offer substantial evidence regarding the effect of local representation provision on reelection. Across all countries in the sample, MPs are significantly more likely to be reelected when they make a greater number of localist speeches throughout their term in office. A natural next question then is, why do MPs not provide constituency representation to a greater extent? I have argued that while voters want their MP to provide local representation, the leadership of the MP’s party might have different priorities and expectations. While voters are expected to care primarily for their own constituency, party leaders are keen on maximizing the strength of the party at the national level, and also on retaining their position at the top of the party hierarchy. They will be keen on retaining – or increasing – their party’s seats, but they will also want to position close allies in favorable situations. In addition, they will want to make sure that backbenchers do not become so strong locally that they are in a position to threaten their authority. Finally, as floor time is limited, it might not be ideal for them if backbenchers continuously vie for it and compete with party leaders for it. Taking those dynamics into consideration, party leaders might prefer MPs representing constituencies in party strongholds to be less assertive in parliament than those who represent competitive constituencies. I argue that MPs take the potentially competing interests of voters and party leaders into account when deciding whether or not to take the floor to talk on constituency issues. I expect that the degree to which MPs engage in local representation will depend on the degree to which MPs depend on the party brand. I introduce the concept of party dominance, which captures the difference in relative support in the constituency for the MP vis-à-vis the party. If an MP enjoys more support 86 in the constituency than the party, they will be more likely to engage in local representation. On the other hand, if the party has more support, the MP will be more cautious of upsetting the party leadership and will as a result be less likely to provide constituency representation. Overall, this chapter tests Hypothesis 2, which states that in constituencies where party dominance is higher, MPs will be less likely to engage in constituency representation. The empirical tests are conducted using the original dataset introduced in Chapter 4. I present evidence that the level of party dominance has a strong effect on the degree to which MPs focus on providing local representation in parliament. When support for the MP in the constituency is larger than that of the party they represent, the MP makes a larger number of constituency-oriented speeches. On the contrary, when the MP is elected because they represent the locally strong party, but support for them is relatively weak, they are much less likely to focus on addressing local issues from the floor of parliament. Overall, the findings offer strong support for Hypothesis 2. The chapter begins by laying out the research design for this part of the analysis. First, I discuss the operationalization of party dominance. In the following sections, I provide an overview of the dataset, and I go over the model specification and the control variables. Then, I present the results of the main analysis. Following that, I report the results of additional robustness tests. Finally, I summarize the conclusions. 6.1 Research Design 6.1.1 Measuring Party Dominance The main independent variable in this part of the analysis is party dominance. The concept of party dominance aims to capture, for a given constituency, the degree to which the party would be electorally successful if it fielded a different MP candidate and how successful the MP candidate could have been if they ran with a different party. In other words, it intends to measure the degree to which the constituency-level support for a party is driven by the party or by the MP. In most cases, we might expect that the party label would be a critical cue for the voters. However, in some cases well-liked politicians may receive a large number of votes regardless of the party they run for. Party dominance does not literally refers to how dominant a party actually is in the constituency, it 87 is rather a measure of the party’s strength relative to that of the MP. I operationalize party dominance for a given constituency as the difference in vote share between the party’s presidential candidate – at that constituency – and the party’s candidate for the constituency’s legislative seat. As a result, the measure could technically take values from -1 to 1. For example, in a constituency where the presidential candidate got 67% of the vote and the candidate for the legislative seat got 37% of the vote, the party dominance score will be 0.3. If on the other hand, the presidential candidate received 48% and the incoming MP received 56% of the vote, the party dominance score will be -0.08. This score is calculated only for the party whose legislative candidate won in the general election and is that constituency’s incoming MP. Therefore, there is one observation per constituency per parliamentary term. Descriptive statistics for the party dominance measure are provided in Table 6.1. In regard to this measure, two additional points need to be discussed. The first regards attrition in the data. To begin with, the party dominance score cannot be calculated in the case of Botswana. This is because in Botswana the president is not directly elected by the voters but by the incoming MPs. Therefore, it is not possible to calculate the difference in party vs MP support based on the differences in vote shares. As a result, Botswana is excluded from the sample. In addition, the results of the presidential elections were not available at the constituency level, especially for some of the earlier electoral contests.1 To address this, I use instead as a proxy the results of the presidential race at the district level – one administrative level higher than the constituency – in the cases that those are available. The second point regards the coding for MPs elected as independents. From the countries in the sample, Malawi, Uganda, and Zimbabwe elect a significant number of independent MPs. Nonetheless, in the vast majority of cases, these MPs are independent in name only. Most of the time, they were candidates who ran in party primaries and lost. Subsequently, they ran in the general elections without the party colors but were widely seen as supporters of the party in whose 1The elections for which constituency-level results for the presidential race were not available are Malawi 1999, Malawi 2004, Uganda 2006, and Uganda 2011. 88 primaries they initially competed. As discussed in Chapter 2, primaries attract more numerous and more wealthy candidates in a party’s stronghold constituencies. In such contests, the outcomes are nominally decided by the voters. Nonetheless, local party politics may also affect the outcome. While only one candidate can win the nomination, others may go on to run as independents. These dynamics often lead to scenarios where the locally strong party officially nominates one candidate to run in the national elections, but other candidates, who compete in the national elections as independents, also make it clear that they would support the locally strong party if elected to office. In other words, in the context of stronghold constituencies, independent MPs often brand themselves as another option for the locally strong party. This is the case even if independent MPs think they lost in the primaries due to external interference by the party elites. As a result, the vote of party supporters is split between the official party nominee and multiple pretenders. This explains why in several cases the vote share for a party’s presidential candidate is so much greater than that of the party’s legislative candidate. Among the Malawian MPs I interviewed, two made a point of being elected as independents despite losing the primaries. Both MPs also emphasized that the voting base knew they were true supporters of “their" party. Taking the above into account, in the case of independent MPs, party dominance is calculated against the vote share of the presidential candidate for the party that won the greatest vote share in the constituency. Independent MPs often act as if they represent the locally strong party, they are generally seen as members of that party by the voters, and they tend an effort to officially join the party when possible. Even if an MP manages to be elected as an independent, there are important advantages to receiving the official nomination including campaign funds, and the direct backing of the local party structures. Therefore, I view this coding choice as more accurately modeling the situation on the ground. However, for robustness, I also run models in which independent MPs are dropped from the sample. 6.1.2 Variables and Model Specification The dependent variable used to explore Hypothesis 2 is constituency representation. This is operationalized as the average number of localist speeches made by an MP per day that the 89 parliament was in session. The variable in its final form is continuous and coded at the MP- parliament-country level. This measure served as the dependent variable in the analyses presented in Chapter 4, and its creation is discussed at length in subsection 4.1.4. Updated descriptive statistics based on the more limited sample used in this chapter are provided in Table 6.1. To test Hypothesis 2, as the dependent variable is continuous, I employ OLS regression. I cluster robust standard errors by MP to account for heteroskedasticity between observations of the same MP in different parliamentary sessions. The main independent variable is party dominance. Also, I add country-level fixed effects to account for time-invariant country-specific effects. The results are discussed based on the output of the model as well as based on the results of relevant simulations. To account for possible alternative explanations, I also include some important controls. The descriptive statistics for all variables are provided in Table 6.1. To begin with, I control for newcomer status. Newcomer MPs are less likely to have an established network of support in the constituency that will be independent of that of the party. More senior MPs would have more opportunities to develop such a network and build up a personal following. In addition, newcomer MPs should be less likely to participate in debates and conduct constituency representation, since they are not as familiar with parliamentary procedures. This is a binary variable, and it is coded based on the official election results published by each country’s electoral commission. I also include another binary control for gender. Female MPs are generally expected to be less assertive in parliament. They are also widely documented as participating less in the debates overall (Bäck and Debus, 2019; Clayton and Zetterberg, 2021). Gender was coded based on the names of the MPs. In case the gender could not be ascertained based on the name, further research was conducted on an individual basis. Moreover, I control for government membership. The government party tends to be the one that captured the largest number of seats in the general elections, and in some cases, they perform in dominating fashion. As a result, we might expect that in the case of government parties, party dominance will be more pronounced. In addition, MPs backing the executive are more likely to be 90 judged based on the government’s performance. Thus, government MPs are expected to support government policy and to refrain from actions that would undermine the executive. Therefore, they should be less likely to discuss local problems at length, so as not to be seen as blaming their own party. An MP was coded as a member of the government if at the time of the election, their party affiliation was that of the party backing the executive. The variable was coded based on the official election results. Additionally, I include a control for ministerial appointments. Ministers are viewed differently than backbencher MPs. They are rightfully viewed by their constituents as more powerful and influential. They are also tightly associated with the president, and they have overwhelmingly more access to resources compared to the average MP. As they tend to be high in the party hierarchy, the level of party dominance should affect them to a smaller degree. This is a binary indicator, and it is coded 1 if the MP served as a minister at any point during the term. The variable was coded based on WhoGov (Nyrup and Bramwell, 2020). Furthermore, I control for independent status. While for the purposes of creating the party dominance scores I largely treat independent MPs as belonging to their constituency’s strongest party, it is important to differentiate between them and MPs who officially won a party nomination. An MP was coded as an independent if, at the time of the election, they had no party affiliation. The variable was coded based on the official election results. Table 6.1: Descriptive Statistics for all Variables Variable Number of Standard Mean Min Max Name Observations Deviation Localism 2490 0.15 0.23 0 3.35 Party Dominance 2490 0.07 0.17 -0.74 0.93 Speeches 2490 0.79 1.40 0 20.28 Newcomer MP 2490 0.64 0.48 0 1 Female MP 2490 0.13 0.33 0 1 Minister 2490 0.13 0.33 0 1 Government MP 2490 0.57 0.49 0 1 Independent MP 2490 0.07 0.26 0 1 Parliament 2490 2010 4.66 2001 2017 91 Finally, I control for the number of speeches per day that an MP makes, regardless of focus. The overall number of speeches is an important determinant of the number of constituency-oriented speeches that an MP can make. It also shows whether MPs prioritize making non-constituency- oriented speeches. 6.2 Results In this section, I analyze the relationship between party dominance and constituency represen- tation in parliament using a cross-national sample over an extended period of time. To begin with, I investigate H2 descriptively in Figure 6.1. The figure presents scatterplots with the dependent variable, the number of localist speeches per day, on the y-axis, and the main independent variable, party dominance, on the x-axis. A fitted line is also included in blue color. The data are presented separately by country. The theoretical expectation would be a negative relationship between party dominance and the number of localist speeches. At first glance, the data points give little indication of the existence of such a negative relationship. However, when looking at the fitted line, it becomes clear that in most countries the association is indeed negative. The only exception is Ghana, where the fitted line appears to be flat. These findings indicate that the level of party dominance may be an important predictor of representation in parliament in the region. However, it is clear that further analysis is necessary. To formally test H2, I run an OLS regression model where the dependent variable is the number of localist speeches that an MP makes in parliament. The results are presented in Table 6.2 (Model 1). Based on the results of the analysis, an increase of 1 point in the party dominance measure leads to a decrease of 0.07 localist speeches per day. When compared to the average number of localist speeches made by an MP per day, the effect represents a 46.7% change. The relationship is highly statistically significant. The results remain consistent in Model 2, where I restrict the sample to MPs that belong to a specific party and I drop those elected as independents. In substantive terms, the effect appears to be enormous. However, to better understand its magnitude it is important to take into consideration that Party Dominance ranges from -1 to 1, with the vast majority of cases falling between -0.3 and 0.3. 92 Figure 6.1: Localist speeches and reelection Ghana Malawi Uganda 1.00 1.5 3 0.75 1.0 2 0.50 0.5 1 Number of Localist Speeches per Day 0.25 0.0 0 0.00 −0.2 0.0 0.2 0.4 −0.8 −0.4 0.0 0.4 −0.4 0.0 0.4 0.8 Zambia Zimbabwe 2.0 1.00 1.5 0.75 1.0 0.50 0.5 0.25 0.0 0.00 0.0 0.5 −0.6 −0.3 0.0 0.3 Party Dominance To assess whether these results are the artifact of outliers in the data, based on Model 1, I calculate Studentized residuals to detect outliers, hat values to detect leverage points, and Cook’s distances to detect influential data points. This sensitivity analysis showed that the results were not 93 driven by influential observations (Figure A.2). Table 6.2: The Effect of Party Dominance on Localist Speeches Model 1 Model 2 Model 3 Party Dominance -0.07*** -0.06* -0.06** (0.02) (0.02) (0.02) Speeches 0.10*** 0.10*** 0.15*** (0.00) (0.00) (0.00) Newcomer 0.00 0.00 (0.01) (0.01) Female 0.00 0.00 0.01 (0.01) (0.01) (0.01) Minister 0.03* 0.03* -0.03* (0.01) (0.01) (0.01) Government -0.03*** -0.03*** -0.01 (0.01) (0.01) (0.01) Independent 0.03+ 0.03* (0.02) (0.01) Parliament 0.00*** 0.00*** 0.00*** (0.00) (0.00) (0.00) Ghana -7.59*** -5.80*** -5.52*** (1.57) (1.60) (1.57) Malawi -7.48*** -5.70*** -5.46*** (1.57) (1.60) (1.56) Uganda -7.51*** -5.71*** -5.45*** (1.58) (1.60) (1.57) Zambia -7.39*** -5.61*** -5.38*** (1.57) (1.60) (1.56) Zimbabwe -7.56*** -5.77*** -5.49*** (1.57) (1.60) (1.56) Num.Obs. 2490 2311 1589 R2 0.645 0.651 0.720 BIC -1860.8 -1792.4 -1987.5 RMSE 0.16 0.16 0.13 + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 In addition to party dominance, government membership also has an effect on the provision of local representation that is consistent, statistically significant, and substantively considerable. Based on the models presented in Table 6.2, government MPs make 0.03 localist speeches per day fewer than opposition MPs. This represents a 20% drop from the sample mean. This is an 94 indication that government MPs are more willing to forego engaging in local representation. A plausible explanation for this is that government party MPs rely more on party success. In addition, government MPs tend to have wider access to resources. They might rely more on the distribution of such resources rather than on local representation for a possible reelection bid. Based on the theoretical claims that were made in Chapter 2 and tested in Chapters 4 and 5, I would expect MPs who make more localist speeches to be more likely to be reelected. This should also affect their vote choice, which could in turn affect the measure of party dominance for the next parliamentary term if they manage to be reelected. Such a cycle would be in line with the theoretical expectations made in this dissertation. Nonetheless, this creates the possibility of an endogenous relationship in which the party dominance score is low because, in a previous term, the MP made a large number of localist speeches, which led to an increased vote share. In order to address this issue, in Model 3, I conduct the analysis with only newcomer MPs in the sample. The results remain consistent in direction, magnitude, and statistical significance. To help put the main finding in perspective, in the top panel of Figure 6.2, I present the results of simulations that were conducted following the observed values approach (Hanmer and Kalkan, 2013). These simulations predict the number of localist speeches made across the entire range of party dominance. The first differences are presented in the right panel of the Figure, with the comparison category being a party dominance score of 0. Focusing on some plausible values, an MP that got a vote 20% smaller than that of their party’s presidential candidate, is expected to make 0.14 localist speeches per day (95% CIS: 0.13–0.15). On the other hand, an MP whose vote share was greater than that of their party’s presidential candidate by 20% is expected to make 0.17 locally oriented speeches per day (95% CIs: 0.15–0.18). Therefore, the latter MP makes 17.6% more locally oriented speeches per day. To get a better substantive understanding of these results in the context of specific countries, I conduct additional simulations. I examine the effect of party dominance on local representation by altering the degree to which a party is dominant in the constituency. I run scenarios in which the party dominance takes the value of the 5th, the 50th, and the 95th percentile for government and for 95 Figure 6.2: Party Dominance and Representation focus Number of Localist Speeches 0.2 0.1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 Party Dominance Change in Localist Speeches 0.10 0.05 0.00 −0.05 −0.10 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 Party Dominance opposition MPs. While the median party dominance is 0.04, the measure takes a value of -0.15 in the 5th percentile and 0.41 at the 95% percentile. I run separate simulations for Ghana and Malawi. In Ghana, party dominance scores are more clustered around 0 compared to other countries in the sample (SD: 0.07). On the other hand, in Malawi, there is a larger variation between MPs (SD: 0.29). When calculating the changes in predicted probabilities of the number of localist speeches made by an MP per day, binary control variables are held at their mode and continuous variables to their mean. Thus, the “model MP” in these simulations is male, newcomer, party-affiliated –i.e. 96 not independent –, has no ministerial portfolio, and makes 0.77 speeches per day. Table 6.3: Predicted Probability of Localist Speeches Gov. Gov. Gov. Opp. Opp. Opp. 5% Median 95% 5% Median 95% Malawi 0.18 0.17 0.13 0.21 0.20 0.17 Ghana 0.07 0.06 0.03 0.11 0.09 0.07 The results of the simulations are presented in Table 6.3. To begin with, the results are consistent with the theoretical expectations as stated in H2. When the MP’s vote share is greater than that of their party’s presidential candidate, the MP makes a larger number of localist speeches. As this balance turns in favor of the presidential candidate, the MP makes fewer constituency-oriented speeches. For example, in the context of Ghana, a government MP in the 5th percentile makes 0.07 localist speeches per day while their colleague in the 50th percentile makes 0.06 speeches per day. This represents a drop of 14.3%. Continuing this trend, a third MP who is in the 95th percentile will only make 0.03 localist speeches per day, which is an additional drop of 50%. Similarly, a Malawian opposition MP who is in the 95th percentile makes 0.03 speeches fewer than their colleague in the 50th percentile. This represents a 15% decrease compared to their median colleague. A second important observation is that opposition MPs make more localist speeches per day, and they are less affected by the level of party dominance. For example, the change between a Ghanaian MP in the 50th and the 95th percentiles is 22.2% if they are members of the opposition. This change is 27.8% smaller than it would be for government MPs. This finding fits well with the observation made in Chapter 4, that government MPs make fewer constituency-oriented speeches and are less likely to be reelected. The simulations presented here indicate that government MPs tend to also be under firmer control by their party. 6.2.1 Robustness Tests To test the stability of the findings, I ran additional models using the same specification as the main model but dropping one country at a time from the sample. The output of these models is presented in Table 6.4. In all iterations of the model, the main findings remain consistently 97 negative and statistically significant. Regardless of which country is dropped from the sample, party dominance remains an important predictor of local representation provision. Table 6.4: Vote Share and Constituency Representation with Country Dropout Model 2 Model 3 Model 4 Model 5 Model 6 Drop Drop Drop Drop Drop Ghana Malawi Uganda Zambia Zimbabwe Party Dominance -0.07** -0.05* -0.07** -0.08*** -0.07** (0.02) (0.02) (0.02) (0.02) (0.02) Speeches 0.16*** 0.10*** 0.09*** 0.07*** 0.10*** (0.00) (0.00) (0.00) (0.00) (0.00) Newcomer 0.01 0.00 -0.01 0.00 -0.01 (0.01) (0.01) (0.01) (0.01) (0.01) Female 0.00 -0.01 0.00 0.01 0.00 (0.01) (0.01) (0.01) (0.01) (0.01) Minister 0.01 0.02* 0.03* 0.02* 0.03* (0.01) (0.01) (0.01) (0.01) (0.01) Government -0.02* -0.02*** -0.03** -0.03*** -0.03*** (0.01) (0.01) (0.01) (0.01) (0.01) Independent 0.05* 0.01 0.04* 0.03+ 0.03+ (0.02) (0.02) (0.02) (0.01) (0.02) Parliament 0.01*** 0.00 0.00*** 0.00*** 0.00*** (0.00) (0.00) (0.00) (0.00) (0.00) Ghana -0.05 -7.24*** -8.56*** -7.91*** (1.40) (1.64) (1.80) (1.72) Malawi -17.55*** -7.14*** -8.45*** -7.80*** (2.23) (1.63) (1.79) (1.72) Uganda -17.58*** 0.05 -8.49*** -7.83*** (2.24) (1.40) (1.80) (1.72) Zambia -17.45*** 0.13 -7.05*** -7.71*** (2.23) (1.40) (1.63) (1.71) Zimbabwe -17.59*** -0.02 -7.21*** -8.54*** (2.24) (1.40) (1.64) (1.80) Num.Obs. 1562 2115 2213 1932 2138 R2 0.708 0.723 0.640 0.543 0.642 BIC -913.1 -2392.9 -1532.2 -1852.8 -1323.8 RMSE 0.18 0.13 0.17 0.15 0.17 + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 Furthermore, the magnitude of the effect remains relatively consistent. In models where Ghana, Uganda, Zambia, or Zimbabwe are dropped, the coefficient for party dominance takes 98 values between -0.07 and -0.08. The only substantial deviation is that, when Malawi is dropped from the sample, the coefficient decreases to -0.05. Overall, these results signify that no single country drives the findings. To further illustrate these findings, in Figure 6.3, I present the results of simulations of the first differences corresponding to each of the additional models. The baseline category is a party dominance score of 0. Figure 6.3: Party Dominance and Representation Focus with Country Dropout Drop Ghana Drop Malawi 0.10 0.10 0.05 Change in Number of Localist Speeches 0.05 0.00 0.00 −0.05 −0.05 −0.10 −0.10 −1.0 −0.5 0.0 0.5 1.0 −1.0 −0.5 0.0 0.5 1.0 Drop Uganda Drop Zambia 0.10 0.10 0.05 0.05 0.00 0.00 −0.05 −0.05 −0.10 −0.10 −1.0 −0.5 0.0 0.5 1.0 −1.0 −0.5 0.0 0.5 1.0 Party Dominance Drop Zimbabwe 0.10 0.05 0.00 −0.05 −0.10 −1.0 −0.5 0.0 0.5 1.0 Party Dominance 99 6.3 Conclusion In previous chapters, I explored the relationship between local representation and reelection. Chapter 4 provided evidence for the existence of this relationship at the constituency level. Chapter 5 provided additional support for this main theoretical premise by showing how providing local representation affects vote choice at the individual voter level. In this chapter, I tested empirically the second part of the theory, which incorporates the existence of parties into the theoretical story. In order to improve their chances of remaining in office, MPs need to please the voters but to also be in good standing with the party leadership. I argue that while voters want local representation, party leaders may have different priorities, as their focus is on making the party more competitive on the national level. Therefore, the degree to which MPs view the voters vs the party leaders as the more important principle will affect how they act in parliament. The results presented in this chapter show that party dominance is a strong predictor of local representation provision by MPs in parliament. We observe that when support for the MP in the constituency is larger than that of the party they represent, the MP makes a larger number of constituency-oriented speeches. On the contrary, when the MP is elected because they represent the locally strong party, but support for them is relatively weak, they are much less likely to focus on addressing local issues from the floor of parliament. This trend is clear in the entire sample as well as in individual countries. These findings are directly in line with the theoretical expectations stated in Hypothesis 2, that we should observe less engagement in constituency representation when party dominance is higher. 100 CHAPTER 7 CONCLUSIONS For a long time, the connection between clientelism and victory at the polls has been considered the key determinant of voting behavior in Africa. However, the fact that incumbent MPs are frequently unable to get reelected, despite the lucrative perks afforded to them due to their office, indicates that there is more to the story. While much literature on African democracy has been concentrated on elite-level politics, much less is known about the representation deficits that exist in many new African democracies. Going against much of the literature, I argue that representation, a concept severely understudied with regard to the particular region, is a critical determinant of future support at the polls. I theorize that in contexts where state penetration is low, addressing local issues is especially important for voters. Therefore, MPs who provide local representation in parliament are more likely to have their vote in the next elections, and, on aggregate, they will be more likely to be reelected. Moreover, in the second part of this dissertation, I study what impedes MPs from performing local representation to a greater extent. I highlight the role of local party competition at the constituency level as an important determining factor for the level of local representation that MPs provide from the floor of parliament. I have argued that MPs are held accountable to two principals, their party’s leadership and to the voters. While voters prioritize local representation, this is not necessarily true for the party leadership. Therefore, in constituencies with low levels of party dominance, incumbent MPs will prioritize the wishes of the constituencies, and provide more local representation. On the other hand, in constituencies with high levels of party dominance, MPs might refrain from taking the floor as a precaution against angering the party leadership. 7.1 Summary of Findings In Chapter 4, statistical analysis of the cross-national dataset on parliamentary speeches provides support for Hypothesis 1, stating that MPs who focus on constituency representation in parliament are more likely to be reelected. Specifically, I found that MPs who make more locally oriented speeches in parliament are significantly and substantively more likely to be reelected. While 101 this finding is strong in the six-country sample as a whole, the relationship is also clear within each individual country. Importantly, the effect of local representation is distinct and greater in magnitude than the effect of overall effort and participation in the debates. However, this analysis, which takes place at the constituency level, does not directly test the relationship between local representation and vote choice, which takes place at the individual voter level. Therefore, in Chapter 5, I test the effect of providing local representation on intended vote choice via a forced-choice conjoint experiment, which took place in twelve constituencies, in Malawi. Specifically, I test whether individual voters will be more likely to vote for incumbent MPs if they have demonstrated a focus on constituency representation during their tenure. The analysis results show that voters are much more likely to vote for incumbents over challengers if the incumbent has demonstrated a focus on representing the constituency during their time in parliament. In addition, by combining survey data with data on parliamentary speaking, I test whether respondents are aware of the degree to which their sitting MP provides local representation in parliament. I show that there is a strong and statistically significant positive correlation between the number of localist speeches made by the MP and the respondents’ responses to the relevant question. This is a strong indication that respondents were aware of the degree to which their current MP was taking the floor to discuss constituency matters. In Chapter 6, I explored what inhibits MPs from focusing more on constituency representation. I argued that it is a function of the level of party dominance. Specifically, in constituencies where support for the party is stronger than support for the MP, MPs will be less likely to engage in constituency representation, as they will prioritize pleasing the party leadership over the voters. I empirically explored this second hypothesis using the dataset on parliamentary speaking. The results of the analysis show that when the MP’s influence in the constituency is larger than that of the party they represent, the MP makes a larger number of constituency-oriented speeches. On the contrary, when the MP is elected because they represent the locally strong party, but support for them is relatively weak, they are much less likely to focus on local representation. As with the earlier analysis, the results are stable and substantively large both in the entire sample and within 102 individual countries. Overall, I demonstrate that African voters care about local representation, and they are informed about whether it is provided. Consequently, this affects their vote choice, which makes MPs who discuss matters of their constituency on the floor of parliament more likely to be reelected. In addition, I show that under conditions of high party dominance, MPs will be less likely to provide local representation. 7.2 Academic and Policy Implications The findings presented are important for our understanding of voting behavior in Africa, as they challenge the conventional wisdom in two important regards. First, they cast doubt on earlier claims that what happens in parliament has a limited impact on political outcomes like voting behavior. They call greater attention to how processes that take place in parliament may affect political behavior. For example, these findings can have important repercussions on how we approach nation-building. As has been argued throughout this manuscript, weak state penetration makes local issues more salient for voters. This is especially the case for voters located in the periphery. If these issues are not brought to the center by their representative, it is easy for them to feel cut off and neglected. Under such circumstances, they will be less likely to buy into larger nation-building efforts promoted by the government. If on the other hand, citizens feel that their problems and concerns are taken seriously and brought to the government, they will be more likely to buy into such efforts and support them. Second, they demonstrate the existence of an accountability mechanism tied to a legislator’s performance beyond the transactional clientelism narrative that has been the focus of existing literature. The development of such a linkage is impeded when MPs focus their efforts on winning over party elites rather than the voters. As a result, it might be worth revisiting how we think about democratization in Africa and the way we promote and support it. Strong parliaments are crucial when it comes to checking the executive and preventing democratic backsliding. For this reason, large amounts of financial aid have been invested in increasing the capacity of parliaments across the continent. Such programs include training for MPs as well as the provision of funds to be used 103 at their discretion and have led for example to the creation of Constituency Development Funds. However, for a parliament to be strong a degree of continuity is critical (Opalo, 2019; Thomas and Sissokho, 2005). Based on the findings of this study, emphasizing and promoting the aspect of local representation is essential in achieving this continuity. Representation is the very foundation of modern democracy and it is key for democratic inclusion between elections. This is especially important in countries where the state’s penetration in much of their territory is low. In addition, representation and inclusion are attributes highly related to democratic satisfaction and support for democracy. Therefore, as increased focus on representation leads to increased electoral support, we can expect more experienced cadres of MPs in parliament who will be better able to check the executive. In addition, continued popular support would mean that citizens are invested in their representatives and, thus, more likely to support the regime. Therefore, in order to support democratization in the continent, it might be beneficial to invest in broadening the avenues of communication between citizens and MPs. Better understanding of MPs’ strategies and citizens’ priorities would allow us to develop methods to improve the two-way flow of information which will enhance representation and accountability, and also strengthen parliaments. Overall, the project contributes to important theoretical debates on elite and citizen political behavior, representation, accountability, and legislative politics. It adds to a growing literature on legislative politics in Africa by adding new cross-national data. Also, this study adds to our understanding of democratic consolidation on the African continent. While legislatures are central institutions for executive oversight, this paper suggests that low levels of subnational competition, which empowers central party elites, have often hindered individual MPs from fostering strong links with constituents. 7.3 Questions for Future Research Although this study provides a very optimistic outlook regarding democratic representation and accountability, several questions remain unanswered. There are several ways that this work can be extended to enhance our understanding of the dynamics related to local representation and 104 reelection. At a cross-national level, it is important to study the influence of key institutions like regime type, and party institutionalization. At a within-country level, areas that entail further study include opposition party membership and differences across the urban/rural divide. In the following sections, I expand briefly on each of these directions. 7.3.1 Local Representation and Quality of Democracy The cross-national dataset used in this dissertation includes countries that range from democratic to authoritarian. Nonetheless, the level of democracy is not considered at any point in the theorizing or in the analysis. In a democratic political system, there is a principal-agent relationship between voters and MPs. The expectation of fair elections and the possibility of retrospective sanctions by the voters may provide an incentive to MPs, regardless of ability, to perform better (Ferejohn, 1986). The existence of such a dynamic allows MPs to associate historical rates of turnover with their performance. However, in the context of new democracies, and especially when the quality of democracy is low, this process is often distorted by irregularities like unfair electoral contests. This can subvert the aforementioned effect (Ofosu, 2019). MPs who can reap the rewards of office without fear of retrospective punishment at the polls might be more likely to focus on accumulating rents than representing their constituents. Furthermore, if unfair elections lead to frequent turnover, MPs should be less likely to associate their performance in parliament with future electoral success. When the quality of elections is higher, voters should be better able to hold their representatives accountable based on their perceived performance. 7.3.2 Local Representation and the Institutionalization of Parties and Party Systems The importance of parties becomes clear in the second part of the theory proposed in this manuscript when exploring how party dominance affects the provision of local representation. Therefore, indirectly, the institutionalization of individual parties and of party systems as a whole plays a role in the theory. However, its importance is not directly discussed, and no relevant hypotheses are formulated. In more institutionalized party systems, the major parties tend to be the same over time, and the rules of the game tend to be more regularized (Mainwaring and Scully, 1995). In more institutionalized parties, the party leadership has a tighter grip on who 105 takes the floor, and the party line on policy issues is readily enforced. As a result, being able to consistently take the floor may signify that the party leadership is positively disposed towards a particular MP and that their position within the party is stronger. Such a nod of approval may go a long way toward reelection in terms of lowering within-party competition. Also, institutionalized party systems are more conducive to information dissemination (Moser and Scheiner, 2012), which should allow voters to monitor whether their MPs prioritize constituency representation. On the contrary, low-party system institutionalization makes it harder for voters to monitor politicians and impede accountability at the polls (Schleiter and Voznaya, 2018). 7.3.3 Local Representation and Opposition Party Membership The results reported in Chapters 4 and 6 indicate that there are important differences between government and opposition MPs when it comes to providing local representation. Nonetheless, this relationship is not theorized or purposely tested in this manuscript. Individual opposition MPs should have more opportunity to take the floor, considering that they will be fewer in number and that ministers tend to occupy much of the government party’s speaking time. In addition, government MPs might have reason to be concerned that discussing local problems and concerns in parliament could make their party – the government party – look bad, which might in turn harm their standing with the party leadership. Opposition MPs should be less impeded by such considerations. On the contrary, local representation by opposition MPs could function as a form of territorial opposition to the center (see Boone et al., 2022). Furthermore, government MPs should have access to more resources to distribute in their constituencies (Brierley et al., 2020), and their claims should be more credible (Wantchekon, 2003). Therefore, we could expect that, overall, constituency representation would be a more valuable tool for opposition MPs. 7.3.4 Local Representation and Urban-Rural Divide Another important cleavage that is not considered in this manuscript is the urban-rural divide. Historically, in presidential elections, urban constituencies are more likely to support the opposition, while rural constituencies tend to side with the incumbent. In urban constituencies, voters tend to have a larger degree of agency in terms of vote choice (Koter, 2013; Wahman and Boone, 2018). In 106 addition, information dissemination can be expected to be wider and more effective in urban areas, where there is deeper penetration of radio, TV, and the internet, and where people tend to be more educated (Conroy-Krutz, 2018a). In these contexts, voters should be better able to monitor their MPs. Therefore, we might expect that the effect of providing constituency representation might be more pronounced in urban constituencies compared to constituencies in rural areas. 7.4 Conclusion While so many unanswered questions remain, my goal in this project was to question long-held assumptions regarding political behavior in developing democracies, like the ones in Africa, and to create the conditions for a larger debate that goes beyond our current understanding of these topics. I point out that, in such contexts, existing literature can be portrayed as accurately identifying regionalism as a particularly important cleavage. However, while the literature has predominantly focused on clientelism and constituency service as the main points of emphasis, I draw attention to the provision of local representation as an additional expression of this regional cleavage. The goal is that the research presented here will motivate such a discussion and that additional studies will build on the findings presented here. 107 BIBLIOGRAPHY Abrahamsen, R., & Bareebe, G. (2016). Uganda’s 2016 elections: Not even faking it anymore. African Affairs, 115(461), 751–765. Acheampong, M. (2020). Legislators’ pathway to power in Ghana: Intra-party competition, clientelism and legislator – constituents’ relationship. The Journal of Legislative Studies, 1–17. Alemán, E., Micozzi, J. P., & Ramírez, M. M. (2018). The hidden electoral connection: Analysing parliamentary questions in the Chilean Congress. The Journal of Legislative Studies, 24(2), 227–244. Alemán, E., Ramírez, M. M., & Slapin, J. B. (2017). Party Strategies, Constituency Links, and Legislative Speech. Legislative Studies Quarterly, 42(4), 637–659. Alesina, A., Easterly, W., & Matuszeski, J. (2011). Artificial States. Journal of the European Economic Association, 9(2), 246–277. Arriola, L. R. (2009). Patronage and Political Stability in Africa. Comparative Political Studies, 42(10), 1339–1362. Arriola, L. R. (2013). Capital and Opposition in Africa: Coalition Building in Multiethnic Societies. World Politics, 65(2), 233–272. Arriola, L. R., Choi, D. D., Davis, J. M., Phillips, M. L., & Rakner, L. (2021). Paying to party: Candidate resources and party switching in new democracies. Party Politics, 1354068821989563. Asunka, J., Brierley, S., Golden, M., Kramon, E., & Ofosu, G. (2019). Electoral Fraud or Violence: The Effect of Observers on Party Manipulation Strategies. British Journal of Political Science, 49(1), 129–151. Azevedo, E. (2009). Elected legislatures in sub-saharan Africa: Attitudes of citizens from 18 countries towards legislatures, with a particular focus on Mozambique, its assembly and parliamentarians. Bäck, H., & Debus, M. (2019). When Do Women Speak? A Comparative Analysis of the Role of Gender in Legislative Debates. Political Studies, 67(3), 576–596. Bailer, S., Breunig, C., Giger, N., & Wüst, A. M. (2022). The Diminishing Value of Representing the Disadvantaged: Between Group Representation and Individual Career Paths. British Journal of Political Science, 52(2), 535–552. 108 Ballard, A. O., DeTamble, R., Dorsey, S., Heseltine, M., & Johnson, M. (2022). Dynamics of Polarizing Rhetoric in Congressional Tweets. Legislative Studies Quarterly, n/a(n/a). Barkan, J. D. (2008). Legislatures on the Rise? Journal of Democracy, 19(2), 124–137. Barkan, J. D. (Ed.). (2009). Legislative power in emerging African democracies. Lynne Rienner Publishers. Barkan, J. D., & Mattes, R. (2014). Why CDFs in Africa?: Representation vs. constituency service (Working Paper). University of Cape Town. Barkan, J. D., & Ng’ethe, N. (1998). African Ambiguities: Kenya Tries Again. Journal of Democracy, 9(2), 32–48. Baskin, M., & Mezey, M. L. (Eds.). (2014). Distributive politics in developing countries: Almost pork. Lexington Books. Bates, R. H. (2008). State Failure. Annual Review of Political Science, 11(1), 1–12. Baumgartner, F. R., Green-Pedersen, C., & Jones, B. D. (2006). Comparative studies of policy agendas. Journal of European Public Policy, 13(7), 959–974. Bertelli, A. M., & Dolan, R. M. (2009). The Demand and Supply of Parliamentary Policy Advocacy: Evidence from UK Health Policy, 1997–2005. Government and Opposition, 44(3), 219– 242. Bhandari, A., Larreguy, H., & Marshall, J. (2021). Able and Mostly Willing: An Empirical Anatomy of Information’s Effect on Voter-Driven Accountability in Senegal. American Journal of Political Science. Bird, K. (2014). Ethnic quotas and ethnic representation worldwide. International Political Science Review, 35(1), 12–26. Bleck, J., & Walle, N. v. d. (2019). Electoral Politics in Africa since 1990: Continuity in Change. Cambridge University Press. Bogaards, M., & Elischer, S. (2016). Competitive authoritarianism in Africa revisited. Zeitschrift für Vergleichende Politikwissenschaft, 10(S1), 5–18. Boone, C. (2003). Political topographies of the African state: Territorial authority and institutional choice. Cambridge University Press. Boone, C., Wahman, M., Kyburz, S., & Linke, A. (2022). Regional cleavages in African politics: Persistent electoral blocs and territorial oppositions. Political Geography, 99, 102741. 109 Borghetto, E., Santana-Pereira, J., & Freire, A. (2020). Parliamentary Questions as an Instrument for Geographic Representation: The Hard Case of Portugal. Swiss Political Science Review, 26(1), 10–30. Bowler, S. (2010). Private Members’ Bills in the UK Parliament: Is There an ‘Electoral Connec- tion’? The Journal of Legislative Studies, 16(4), 476–494. Bratton, M., Mattes, R. B., & Gyimah-Boadi, E. (2005). Public opinion, democracy, and market reform in Africa. Cambridge University Press. Bratton, M., & Van de Walle, N. (1997). Democratic experiments in Africa: Regime transitions in comparative perspective. Cambridge University Press. Brierley, S. (2012). Party unity and presidential dominance: Parliamentary development in the fourth republic of Ghana. Journal of Contemporary African Studies, 30(3), 419–439. Brierley, S., Kramon, E., & Ofosu, G. K. (2020). The Moderating Effect of Debates on Political Attitudes. American Journal of Political Science, 64(1), 19–37. Bulut, A. T., & İlter, E. (2020). Understanding Legislative Speech in the Turkish Parliament: Reconsidering the Electoral Connection under Proportional Representation. Parliamentary Affairs, 73(1), 147–165. Burke, E. (1774). Speech to the Electors of Bristol at the Conclusion of the Poll. The Works of the Right Honourable Edmund Burke (pp. 446–448). Cain, B., Ferejohn, J., & Fiorina, M. (1987). The Personal Vote: Constituency Service and Electoral Independence. Harvard University Press. Cappelen, C., & Sorens, J. (2018). Pre-colonial centralisation, traditional indirect rule, and state capacity in Africa. Commonwealth & Comparative Politics, 56(2), 195–215. Carey, J. M. (1998). Term limits and legislative representation (1st pbk. ed). Cambridge University Press. Carey, J. M. (2007). Competing Principals, Political Institutions, and Party Unity in Legislative Voting. American Journal of Political Science, 51(1), 92–107. Carey, J. M. (2008). Legislative Voting and Accountability. Cambridge University Press. Carrier, N., & Kochore, H. H. (2014). Navigating ethnicity and electoral politics in northern Kenya: The case of the 2013 election. Journal of Eastern African Studies, 8(1), 135–152. Chiru, M. (2018). Cheap Talk or Proper Signaling? Styles of Campaigning and Engagement in Constituency Service. Social Science Quarterly, 99(1), 283–295. 110 Cho, W. (2012). Accountability or Representation? How Electoral Systems Promote Public Trust in African Legislatures. Governance, 25(4), 617–637. Clayton, A. (2021). How Do Electoral Gender Quotas Affect Policy? Annual Review of Political Science, 24(1), 235–252. Clayton, A., Josefsson, C., & Wang, V. (2014). Present without Presence? Gender, Quotas and Debate Recognition in the Ugandan Parliament. Representation, 50(3), 379–392. Clayton, A., & Zetterberg, P. (2021). Gender and Party Discipline: Evidence from Africa’s Emerging Party Systems. American Political Science Review, 115(3), 869–884. Collord, M. (2016). From the electoral battleground to the parliamentary arena: Understanding intra-elite bargaining in Uganda’s National Resistance Movement. Journal of Eastern African Studies, 10(4), 639–659. Collord, M. (2021). Pressuring MPs to act: Parliament, organized interests and policymaking in Uganda and Tanzania. Democratization, 28(4), 723–741. Compagnon, D. (2011). A Predictable Tragedy: Robert Mugabe and the Collapse of Zimbabwe. University of Pennsylvania Press. Conroy-Krutz, J. (2020). The Squeeze On African Media Freedom. Journal of Democracy, 31(2), 96–109. Conroy-Krutz, J. (2013). Information and Ethnic Politics in Africa. British Journal of Political Science, 43(2), 345–373. Conroy-Krutz, J. (2018a). Individual Autonomy and Local-Level Solidarity in Africa. Political Behavior, 40(3), 593–627. Conroy-Krutz, J. (2018b). Media exposure and political participation in a transitional African context. World Development, 110, 224–242. Crisp, B. F., Escobar-Lemmon, M. C., Jones, B. S., Jones, M. P., & Taylor-Robinson, M. M. (2004). Vote-Seeking Incentives and Legislative Representation in Six Presidential Democracies. The Journal of Politics, 66(3), 823–846. Dahl, R. A., & Shapiro, I. (2015). On democracy (2nd edition). Yale University Press. Däubler, T. (2020). National policy for local reasons: How MPs represent party and geographical constituency through initiatives on social security. Acta Politica, 55(3), 472–491. Demarest, L. (2021). Elite clientelism in Nigeria: The role of parties in weakening legislator-voter ties. Party Politics, 939–953. 111 Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018). BERT: Pre-training of Deep Bidirec- tional Transformers for Language Understanding, 16. Dockendorff, A. (2020). When do Legislators Respond to their Constituencies in Party Controlled Assemblies? Evidence from Chile. Parliamentary Affairs, 73(2), 408–428. D’Orazio, V., Landis, S. T., Palmer, G., & Schrodt, P. (2014). Separating the Wheat from the Chaff: Applications of Automated Document Classification Using Support Vector Machines. Political Analysis, 22(2), 224–242. Doro, E., & Kufakurinani, U. (2018). Resource Curse or Governance Deficit? The Role of Parliament in Uganda’s Oil and Zimbabwe’s Diamonds. Journal of Southern African Studies, 44(1), 43–57. Dulani, B., & van Donge, J. K. (2005). A Decade of Legislature-Executive Squabble in Malawi, 1994–2004. In M. A. M. Salih (Ed.), African Parliaments: Between Governance and Government (pp. 201–224). Palgrave Macmillan US. Elischer, S. (2013). Political Parties in Africa: Ethnicity and Party Formation. Cambridge University Press. Farrell, D. M., & Scully, R. (2007). Representing Europe’s citizens? electoral institutions and the failure of parliamentary representation. Oxford University Press. Fenno, R. F. (1977). U.S. House Members in Their Constituencies: An Exploration. The American Political Science Review, 71(3), 883–917. Fenno, R. F. (1986). Observation, Context, and Sequence in the Study of Politics. American Political Science Review, 80(1), 3–15. Fenno, R. F. (2003). Home style: House members in their districts. Longman. Ferejohn, J. (1986). Incumbent Performance and Electoral Control. Public Choice, 50(1/3), 5–25. Fernandes, J. M., Geese, L., & Schwemmer, C. (2019). The impact of candidate selection rules and electoral vulnerability on legislative behaviour in comparative perspective. European Journal of Political Research, 58(1), 270–291. Ferree, K., & Horowitz, J. (2010). Ties that bind? The rise and decline of ethno-regional partisanship in Malawi, 1994–2009. Democratization, 17(3), 534–563. Fiorina, M. P. (1981). Retrospective voting in American national elections. Yale University Press. Franklin, M. N., Norton, P., & of Parliament Group, S. (Eds.). (1993). Parliamentary questions. Clarendon Press. 112 Frantzeskakis, N., & Park, B. B. (2022). Armed and dangerous: Legacies of incumbent-military ties and electoral violence in Africa. Electoral Studies, 80, 102531. Frantzeskakis, N., & Seeberg, H. B. (2022). The Legislative Agenda in 13 African Countries: A Comprehensive Database. Legislative Studies Quarterly, lsq.12404. Frantzeskakis, N., Wahman, M., & Yildirim, T. M. (2021). Malawi: Parliamentary Debate under Executive Dominance. The Politics of Legislative Debate Around the World. Oxford University Press. Friesen, P. (2023). The partisan consequences of liberation mentality: Ruling party support in Zimbabwe across space and time. Party Politics, 29(2), 306–321. Fujimura, N. (2016). Re-election isn’t everything: Legislators’ goal-seeking and committee activity in Japan. The Journal of Legislative Studies, 22(2), 153–174. Gandhi, J., Noble, B., & Svolik, M. (2020). Legislatures and Legislative Politics Without Democ- racy. Comparative Political Studies, 0010414020919930. Geese, L., & Martínez-Cantó, J. (2022). Working as a team: Do legislators coordinate their geo- graphic representation efforts in party-centred environments? Party Politics, 13540688221098157. Gerring, J. (2016). Case Study Research: Principles and Practices (2nd ed.). Cambridge University Press. Goldring, E., & Wahman, M. (2016). Democracy in Reverse: The 2016 General Election in Zambia. Africa Spectrum, 51(3), 107–121. Goldring, E., & Wahman, M. (2018). Fighting for a name on the ballot: Constituency-level analysis of nomination violence in Zambia. Democratization, 25(6), 996–1015. Good, K. (2005). Resource dependency and its consequences: The costs of Botswana’s shining gems. Journal of Contemporary African Studies, 23(1), 27–50. Good, K. (2008). Diamonds, Dispossession & Democracy in Botswana. Boydell & Brewer Ltd. Good, K., & Taylor, I. (2008). Botswana: A Minimalist Democracy. Democratization, 15(4), 750–765. Grimmer, J., Messing, S., & Westwood, S. J. (2012). How Words and Money Cultivate a Personal Vote: The Effect of Legislator Credit Claiming on Constituent Credit Allocation. American Political Science Review, 106(4), 703–719. Grossman, G., Humphreys, M., & Sacramone-Lutz, G. (2014). "I wld like u WMP to extend electricity 2 our village": On Information Technology and Interest Articulation. The 113 American Political Science Review, 108(3), 688–705. Habyarimana, J., Humphreys, M., Posner, D. N., & Weinstein, J. M. (2007). Why Does Ethnic Diversity Undermine Public Goods Provision? The American Political Science Review, 101(4), 709–725. Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2014). Causal Inference in Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Experiments. Political Analysis, 22(1), 1–30. Hanmer, M. J., & Kalkan, K. O. (2013). Behind the Curve: Clarifying the Best Approach to Calculating Predicted Probabilities and Marginal Effects from Limited Dependent Variable Models. American Journal of Political Science, 57(1), 263–277. Harding, R. (2015). Attribution And Accountability: Voting for roads in Ghana. World Politics, 67(4), 656–689. Harding, R., & Stasavage, D. (2014). What Democracy Does (and Doesn’t Do) for Basic Services: School Fees, School Inputs, and African Elections. The Journal of Politics, 76(1), 229–245. Haykin, S. S. (1999). Neural networks: A comprehensive foundation (2nd ed). Prentice Hall. Heitshusen, V., Young, G., & Wood, D. M. (2005). Electoral Context and MP Constituency Focus in Australia, Canada, Ireland, New Zealand, and the United Kingdom. American Journal of Political Science, 49(1), 32–45. Herbst, J. I. (2000). States and power in Africa: Comparative lessons in authority and control. Princeton University Press. Herzog, A., & Benoit, K. (2015). The Most Unkindest Cuts: Speaker Selection and Expressed Government Dissent during Economic Crisis. The Journal of Politics, 77(4), 1157–1175. Horowitz, D. L. (1991). A Democratic South Africa?: Constitutional Engineering in a Divided Society. University of California Press. Horowitz, D. L. (2003). Electoral Systems: A Primer for Decision Makers. Journal of Democracy, 14(4), 115–127. Horowitz, D. L. (2008). Ethnic groups in conflict: With a new preface (2. ed., Nachdr.). Univ. of California Press. Horowitz, J., & Long, J. (2016). Strategic voting, information, and ethnicity in emerging democ- racies: Evidence from Kenya. Electoral Studies, 44, 351–361. 114 Howard, J., & Ruder, S. (2018). Universal Language Model Fine-tuning for Text Classification. arXiv:1801.06146 [cs, stat]. Humphreys, M., & Weinstein, J. (2010). Policing Politicians. Working paper. Ichino, N., & Nathan, N. L. (2012). Primaries on Demand? Intra-Party Politics and Nominations in Ghana. British Journal of Political Science, 42(4), 769–791. Ichino, N., & Nathan, N. L. (2013). Crossing the Line: Local Ethnic Geography and Voting in Ghana. American Political Science Review, 107(2), 344–361. Ichino, N., & Nathan, N. L. (2022). Democratizing the Party: The Effects of Primary Election Reforms in Ghana. British Journal of Political Science, 52(3), 1168–1185. Izama, A., & Wilkerson, M. (2011). Uganda: Museveni’s Triumph and Weakness. Journal of Democracy, 22(3), 64–78. Kalipeni, E. (1997). Regional Polarisation in Voting Pattern: Malawi’s 1994 Elections. African Journal of Political Science / Revue Africaine de Science Politique, 2(1), 152–167. Kartalis, Y. (2021). Parliamentary current questions’ constituency focus under times of crisis: An examination of the Greek case. The Journal of Legislative Studies, 0(0), 1–21. Katz, R. S. (1997). Representational roles. European Journal of Political Research, 32(2), 211– 226. Keefer, P. (2007). Clientelism, Credibility, and the Policy Choices of Young Democracies. Ameri- can Journal of Political Science, 51(4), 804–821. Keefer, P., & Khemani, S. (2009). When Do Legislators Pass on Pork? The Role of Political Parties in Determining Legislator Effort. American Political Science Review, 103(01), 99–112. Kim, E. K. (2017). Party Strategy in Multidimensional Competition in Africa: The Example of Zambia. Comparative Politics, 50(1), 21–43. Kjær, A. M., & Katusiimeh, M. W. (2021). Nomination Violence in Uganda’s National Resistance Movement. African Affairs, 120(479), 177–198. Koter, D. (2013). King Makers: Local Leaders and Ethnic Politics in Africa. World Politics, 65(2), 187–232. Kramon, E. (2016). Where is vote buying effective? Evidence from a list experiment in Kenya. Electoral Studies, 44, 397–408. 115 Kramon, E. (2018). Money for Votes: The causes and consequences of electoral clientelism in africa. Cambridge University Press. Kriger, N. (2006). From Patriotic Memories to ‘Patriotic History’ in Zimbabwe, 1990 – 2005. Third World Quarterly, 27(6), 1151–1169. Lancaster, T. D. (1986). Electoral Structures and Pork Barrel Politics. International Political Science Review, 7(1), 67–81. Lancaster, T. D., & Patterson, W. D. (1990). Comparative Pork Barrel Politics: Perceptions from the West German Bundestag. Comparative Political Studies, 22(4), 458–477. Laver, M. (2021). Analyzing the Politics of Legislative Debate. The Politics of Legislative Debate Around the World. Oxford University Press. Laver, M., & Benoit, K. (2002). Locating TDs in Policy Spaces: The Computational Text Analysis of Dáil Speeches. Irish Political Studies, 17(1), 59–73. LeBas, A. (2006). Polarization as Craft: Party Formation and State Violence in Zimbabwe. Comparative Politics, 38(4), 419–438. Leith, J. C. (2005). Why Botswana Prospered. McGill-Queen’s Press - MQUP. Levitsky, S., & Way, L. (2010). Competitive authoritarianism: Hybrid regimes after the Cold War. Cambridge University Press. Levitsky, S., & Ziblatt, D. (2018). How democracies die (First edition). Crown. Lijphart, A. (2004). Constitutional Design for Divided Societies. Journal of Democracy, 15(2), 96–109. Lijphart, A. (2012). Patterns of democracy: Government forms and performance in thirty-six countries (2nd ed). Yale University Press. Lindberg, S. I. (2006). The Surprising Significance of African Elections. Journal of Democracy, 17(1), 139–151. Lindberg, S. I. (2010). What accountability pressures do MPs in Africa face and how do they respond? Evidence from Ghana. The Journal of Modern African Studies, 48(1), 117–142. Lipset, S. M., & Rokkan, S. (1967). Party systems and voter alignments: Cross-national perspec- tives (Vol. 7). New York: Free Press. MacCarthaigh, M. (2005). Accountability in Irish parliamentary politics. Institute of Public Administration. 116 Mainwaring, S., & Scully, T. (Eds.). (1995). Building democratic institutions: Party systems in Latin America. Stanford University Press. Maiya, A. S. (2020). Ktrain: A Low-Code Library for Augmented Machine Learning. arXiv preprint arXiv:2004.10703. Malesky, E., & Schuler, P. (2010). Nodding or Needling: Analyzing Delegate Responsiveness in an Authoritarian Parliament. American Political Science Review, 104(3), 482–502. Mansbridge, J. (2003). Rethinking Representation. American Political Science Review, 97(4), 515–528. Mansbridge, J. (2009). A “Selection Model” of Political Representation*. Journal of Political Philosophy, 17(4), 369–398. Mansbridge, J. (2011). Clarifying the Concept of Representation. American Political Science Review, 105(3), 621–630. Marcinkiewicz, K., & Stegmaier, M. (2019). Speaking up to stay in parliament: The electoral importance of speeches and other parliamentary activities. The Journal of Legislative Studies, 25(4), 576–596. Martin, S. (2011). Parliamentary Questions, the Behaviour of Legislators, and the Function of Legislatures: An Introduction. The Journal of Legislative Studies, 17(3), 259–270. Martin, S., & Rozenberg, O. (Eds.). (2012). The Roles and Function of Parliamentary Questions. Routledge. Mattes, R., & Bratton, M. (2007). Learning about Democracy in Africa: Awareness, Performance, and Experience. American Journal of Political Science, 51(1), 192–217. Mattes, R., & Mozaffar, S. (2016). Legislatures and Democratic Development in Africa. African Studies Review, 59(3), 201–215. Mirończuk, M. M., & Protasiewicz, J. (2018). A recent overview of the state-of-the-art elements of text classification. Expert Systems with Applications, 106, 36–54. Mogalakwe, M. (2006). From pre-colony to post-colony : Continuities and discontinuities in political power relations and governance in Botswana. Journal of African Elections, 5(2), 5–20. Mogalakwe, M. (2008). Botswana : Exploding the myth of exceptionality. Africa Insight, 38(1), 105–117. 117 Molomo, M. G. (2000). Understanding government and opposition parties in Botswana. Common- wealth & Comparative Politics, 38(1), 65–92. Montgomery, J. M., & Olivella, S. (2018). Tree-Based Models for Political Science Data. American Journal of Political Science, 62(3), 729–744. Morgenstern, S. (2003). Patterns of Legislative Politics: Roll-Call Voting in Latin America and the United States. Cambridge University Press. Moser, R. G., & Scheiner, E. (2012). Electoral Systems and Political Context: How the Effects of Rules Vary Across New and Established Democracies. Cambridge University Press. Motolinia, L. (2021). Electoral Accountability and Particularistic Legislation: Evidence from an Electoral Reform in Mexico. American Political Science Review, 115(1), 97–113. Mozaffar, S., & Scarritt, J. R. (2005). The Puzzle of African Party Systems. Party Politics, 11(4), 399–421. Mutiga, M. (2016). Tanzania cuts live parliamentary coverage, ending vital news source for citizens. Mzungu, W. (2022). MCP lawmaker Ulemu Chilapondwa clarifies and apologizes over “cooking oil is luxury commodity for the elite” statement - Malawi Nyasa Times - News from Malawi about Malawi. Nyasa Times. Nijzink, L., Mozaffar, S., & Azevedo, E. (2006). Parliaments and the enhancement of democracy on the African continent: An analysis of institutional capacity and public perceptions. The Journal of Legislative Studies, 12(3-4), 311–335. Norris, P. (2014). Why electoral integrity matters. Cambridge University Press. Norton, P. (1993). Back from Westminster: British Members of Parliament and Their Constituents. University Press of Kentucky. Nyrup, J., & Bramwell, S. (2020). Who Governs? A New Global Dataset on Members of Cabinets. American Political Science Review, 114(4), 1366–1374. Ochieng’ Opalo, K. (2022). Formalizing clientelism in Kenya: From Harambee to the Constituency Development Fund. World Development, 152, 105794. Ofosu, G. K. (2019). Do Fairer Elections Increase the Responsiveness of Politicians? American Political Science Review, 113(4), 963–979. Opalo, K. O. (2017). Legislative Independence and Incumbent Electoral Advantage: Evidence from Parliamentary Elections in Kenya, 46. 118 Opalo, K. O. (2019). Legislative Development in Africa: Politics and Postcolonial Legacies (1st ed.). Cambridge University Press. Osei, A. (2012). Party-voter linkage in Africa: Ghana and Senegal in Comparative Perspective. Springer VS. Papp, Z. (2021). Does Issue Alignment Matter? The Electoral Cost and Reward of Agricultural Representation in Urban and Rural Areas. Political Studies Review, 19(3), 376–392. Patel, N., & Mpesi, A. (2010). Between choice and imposition: The politics of nominations. Democracy in Progress: Malawi’s 2009 Parliamentary and Presidential Elections (pp. 295– 316). Patel, N., & Wahman, M. (2015). The Presidential, Parliamentary and Local Elections in Malawi, May 2014. Africa Spectrum, 50(1), 79–92. Pitkin, H. (1967). The Concept of Representation. University of California Press. Popov, A. (2018). Neural Network Models for Word Sense Disambiguation: An Overview. Cybernetics and Information Technologies, 18(1), 139–151. Powell, G. B. (2000). Elections as instruments of democracy: Majoritarian and proportional visions. Yale University Press. Poyet, C. (2021). Personalised Representation in a Weak and Party-Controlled Legislature: Policy Responsiveness in the French Parliament. Political Studies Review, 19(3), 393–409. Poyet, C., & Raunio, T. (2021). Reconsidering the Electoral Connection of Speeches: The Impact of Electoral Vulnerability on Legislative Speechmaking in a Preferential Voting System. Legislative Studies Quarterly, 46(4), 1087–1112. Prempeh, K. ( (2008). Presidents Untamed. Journal of Democracy, 19(2), 109–123. Proksch, S.-O., & Slapin, J. B. (2010). Position Taking in European Parliament Speeches. British Journal of Political Science, 40(3), 587–611. Proksch, S.-O., & Slapin, J. B. (2012). Institutional Foundations of Legislative Speech. American Journal of Political Science, 56(3), 520–537. Przeworski, A., Stokes, S. C., & Manin, B. (Eds.). (1999). Democracy, accountability, and representation. Cambridge University Press. Quinn, K. M., Monroe, B. L., Colaresi, M., Crespin, M. H., & Radev, D. R. (2010). How to Analyze Political Attention with Minimal Assumptions and Costs. American Journal of Political Science, 54(1), 209–228. 119 Rakner, L., & SvÂsand, L. (2004). From Dominant to Competitive Party System: The Zambian Experience 1991–2001. Party Politics, 10(1), 49–68. Rakove, J. N. (2010). Original Meanings: Politics and Ideas in the Making of the Constitution. Knopf Doubleday Publishing Group. Reeder, B. W., & Seeberg, M. B. (2018). Fighting your friends? A study of intra-party violence in sub-Saharan Africa. Democratization, 25(6), 1033–1051. Rehfeld, A. (2005). The Concept of Constituency: Political Representation, Democratic Legiti- macy, and Institutional Design. Cambridge University Press. Rehfeld, A. (2009). Representation Rethought: On Trustees, Delegates, and Gyroscopes in the Study of Political Representation and Democracy. American Political Science Review, 103(2), 214–230. Rehfeld, A. (2011). The Concepts of Representation. American Political Science Review, 105(3), 631–641. Riedl, R. B. (2014). Authoritarian origins of democratic party systems in Africa. Cambridge University Press. Ruhanya, P. (2020). The Militarisation of State Institutions in Zimbabwe, 2002–2017. In S. J. Ndlovu-Gatsheni & P. Ruhanya (Eds.), The History and Political Transition of Zimbabwe (pp. 181–204). Springer International Publishing. Russo, F. (2021). Going Local: Parliamentary Questions as a Means of Territorial Representation in the Italian Parliament. Political Studies Review, 19(3), 410–427. Saalfeld, T. (2012). Parliamentary Questions as Instruments of Substantive Representation: Visible Minorities in the UK House of Commons, 2005–10. The Roles and Function of Parliamen- tary Questions. Routledge. Sanches, E. R., & Dias, A. L. (2021). Ghana: The Politics of Legislative Debates in a Hybrid Presidential Regime. The Politics of Legislative Debates (pp. 399–419). Oxford University Press. Saward, M. (2010). The Representative Claim. OUP Oxford. Saward, M. (2014). Shape-Shifting Representation. American Political Science Review, 108(4), 723–736. Schedler, A. (2002). The Menu of Manipulation. Journal of Democracy, 13(2), 36–50. 120 Schleiter, P., & Voznaya, A. (2018). Party System Institutionalization, Accountability and Govern- mental Corruption. British Journal of Political Science, 48(2), 315–342. Schürmann, L. (2023). Do competitive districts get more political attention? Strategic use of geographic representation during campaign and non-campaign periods. Electoral Studies, 81, 102575. Schürmann, L., & Stier, S. (2023). Who Represents the Constituency? Online Political Com- munication by Members of Parliament in the German Mixed-Member Electoral System. Legislative Studies Quarterly, 48(1), 219–234. Schwarz, D., Traber, D., & Benoit, K. (2017). Estimating Intra-Party Preferences: Comparing Speeches to Votes*. Political Science Research and Methods, 5(2), 379–396. Schwindt-Bayer, L. A. (2005). The incumbency disadvantage and women’s election to legislative office. Electoral Studies, 24(2), 227–244. Scully, R., & Farrell, D. M. (2003). MEPs as Representatives: Individual and Institutional Roles*. JCMS: Journal of Common Market Studies, 41(2), 269–288. Searing, D. (1994). Westminster’s world: Understanding political roles. Harvard University Press. Seeberg, M. B., Wahman, M., & Skaaning, S.-E. (2018). Candidate nomination, intra-party democracy, and election violence in Africa. Democratization, 25(6), 959–977. Siavelis, P. M., & Morgenstern, S. (2008). Candidate Recruitment and Selection in Latin America: A Framework for Analysis. Latin American Politics and Society, 50(4), 27–58. Slapin, B. J., & Proksch, S.-O. (2021). Theories and Empirical Applications of Legislative Debate. The Politics of Legislative Debate Around the World. Oxford University Press. Slapin, J. B., & Kirkland, J. H. (2020). The Sound of Rebellion: Voting Dissent and Legislative Speech in the UK House of Commons. Legislative Studies Quarterly, 45(2), 153–176. Soroka, S., Penner, E., & Blidook, K. (2009). Constituency Influence in Parliament. Canadian Journal of Political Science/Revue canadienne de science politique, 42(3), 563–591. Stefanelli, A., & Lukac, M. (2020). Subjects, Trials, and Levels: Statistical Power in Conjoint Experiments (tech. rep.). SocArXiv. Strøm, K. (1997). Rules, reasons and routines: Legislative roles in parliamentary democracies. The Journal of Legislative Studies, 3(1), 155–174. Sulkin, T., Testa, P., & Usry, K. (2015). What Gets Rewarded? Legislative Activity and Con- stituency Approval. Political Research Quarterly, 68(4), 690–702. 121 Sulley, C. R. (2022). Democracy within parties: Electoral consequences of candidate selection methods in Tanzania. Party Politics, 28(2), 261–271. Svolik, M. (2013). Learning to Love Democracy: Electoral Accountability and the Success of Democracy. American Journal of Political Science, 57(3), 685–702. Taylor, I. (2003). As Good as It Gets? Botswana’s ’Democratic Development’. Journal of Contemporary African Studies, 21(2), 215–231. Thomas, M. A., & Sissokho, O. (2005). Liaison legislature: The role of the National Assembly in Senegal. The Journal of Modern African Studies, 43(1), 97–117. Thomassen, J. (1994). Empirical research into political representation: Failing democracy or failing models. Elections at home and abroad: Essays in honor of Warren Miller (pp. 237– 65). University of Michigan Press Ann Arbor. van Ham, C., & Lindberg, S. I. (2015). From Sticks to Carrots: Electoral Manipulation in Africa, 1986–2012. Government and Opposition, 50(3), 521–548. Vliegenthart, R., Walgrave, S., & Meppelink, C. (2011). Inter-party Agenda-Setting in the Belgian Parliament: The Role of Party Characteristics and Competition. Political Studies, 59(2), 368–388. Wahman, M. (2017). Nationalized incumbents and regional challengers: Opposition- and incumbent- party nationalization in Africa. Party Politics, 23(3), 309–322. Wahman, M., & Boone, C. (2018). Captured Countryside? Stability and Change in Sub-National Support for African Incumbent Parties. Comparative Politics, 50(2), 189–216. Wahman, M., & Brooks, L. (2022). A statistical analysis of the 2019 Malawi presidential and parliamentary vote: Persistence, change, and electoral geography. Wahman, M., Frantzeskakis, N., & Yildirim, T. M. (2021). From Thin to Thick Representa- tion: How a Female President Shapes Female Parliamentary Behavior. American Political Science Review, 115(2), 360–378. Wahman, M., & Goldring, E. (2020). Pre-election violence and territorial control: Political domi- nance and subnational election violence in polarized African electoral systems. Journal of Peace Research, 57(1), 93–110. Wahman, M., & Seeberg, M. B. (2022). Paying to Play: How Parliamentary Candidates Fund Ruling Party Campaigns in Malawi. Comparative Politics, 55(1), 95–118. Wang, V., & Muriaas, R. L. (2019). Candidate selection and informal soft quotas for women: Gender imbalance in political recruitment in Zambia. Politics, Groups, and Identities, 7(2), 122 401–411. Wanjicu, R. (2008). Kenyan Parliament offers live coverage of proceedings and digitized transcripts to the public | IT World Canada News. Wantchekon, L. (2003). Clientelism and Voting Behavior: Evidence from a Field Experiment in Benin. World Politics, 55(3), 399–422. Wanyama, F. O., & Elklit, J. (2018). Electoral violence during party primaries in Kenya. Democ- ratization, 25(6), 1016–1032. Warren, S. S. (2022). Democratizing candidate selection: Controlled turnover in Botswana’s Bulela Ditswe primaries. Party Politics, 28(2), 248–260. Warren, S. S. (2019). Candidate Selection and Political Accountability in African Legislative Elections (Ph.D.). New York University. United States – New York. Weghorst, K. R., & Lindberg, S. I. (2013). What Drives the Swing Voter in Africa? American Journal of Political Science, 57(3), 717–734. Wegmann, S., & Evequoz, A. (2019). Legislative functions in newly democratised countries: The use of parliamentary questions in Kenya and Zambia. The Journal of Legislative Studies, 25(4), 443–465. Wessel Tromborg, M., & Schwindt-Bayer, L. A. (2019). Constituent Demand and District-Focused Legislative Representation. Legislative Studies Quarterly, 44(1), 35–64. Williamson, S., & Magaloni, B. (2020). Legislatures and Policy Making in Authoritarian Regimes. Comparative Political Studies, 0010414020912288. Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M., Macherey, W., Krikun, M., Cao, Y., Gao, Q., Macherey, K., Klingner, J., Shah, A., Johnson, M., Liu, X., Kaiser, Ł., Gouws, S., Kato, Y., Kudo, T., Kazawa, H., . . . Dean, J. (2017). Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation. arXiv:1609.08144 [cs]. Yildirim, T. M., Kocapınar, G., & Ecevit, Y. A. (2021). Status Incongruity and Backlash against Female Legislators: How Legislative Speechmaking Benefits Men, but Harms Women. Political Research Quarterly, 74(1), 35–45. Yildirim, T. M., Kocapınar, G., & Ecevit, Y. A. (2019). Staying active and focused? The effect of parliamentary performance on candidate renomination and promotion. Party Politics, 25(6), 794–804. Young, D. J. (2014). An initial look into party switching in Africa: Evidence from Malawi. Party Politics, 20(1), 105–115. 123 Zittel, T., Nyhuis, D., & Baumann, M. (2019). Geographic Representation in Party-Dominated Legislatures: A Quantitative Text Analysis of Parliamentary Questions in the German Bundestag. Legislative Studies Quarterly, 44(4), 681–711. 124 APPENDIX General Supporting Material Table A.1: Comparative Agendas Project, Major Topics Code CAP Topic 1 Macroeconimics 2 Civil Rights 3 Health 4 Agriculture 5 Labor 6 Education 7 Environment 8 Energy 9 Immigration 10 Transportation 12 Law anc Crime 13 Social Welfare 14 Housing 15 Domestic Commerce 16 Defense 17 Technology 18 Foreign Trade 19 International Affairs 20 Government Operations 21 Public Lands 23 Culture Table A.2: Breakdown of Training Set Number of Country Period Speeches Ghana 2005-2017 15,000 Malawi 2004-2015 2,500 Zambia 2001-2015 7,200 Zimbabwe 2009-2017 2,900 125 Supporting Material for Chapter 4 Table A.3: BERT 10-fold Cross-Validation scores, Policy Topic Fold Precision Recall F1 1 84.66 85.12 84.3 2 84.16 84.4 84.24 3 85.35 84.96 85 4 83.87 83.45 83.46 5 83.88 83.35 83.42 6 81.08 80.46 80.51 7 85.65 85.04 85.06 8 84.78 84.25 84.32 9 83.86 83.25 83.36 10 83.91 83.26 83.59 Mean 84.12 83.75 83.73 Table A.4: BERT 10-fold Cross Validation scores, Localism Fold Precision Recall F1 1 96.2 96.3 95.3 2 94.5 94.9 93.4 3 96.6 96.5 95.6 4 94.9 95.3 94 5 97.8 97.9 97.5 6 96.3 96.9 96.2 7 95.1 95.7 94.5 8 96.1 96.6 95.8 9 96.1 96.4 95.5 10 96 96.1 95 Mean 84.12 83.75 83.73 126 hat−values Studentized residuals Cook's distance 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 −3 −2 −1 0 1 2 0.00 0.01 0.02 0.03 0.04 0.05 0 28 1000 1022 1022 1403 127 Index 1737 1737 2000 3000 Figure A.1: Diagnostic Plots for Ch. 3 Main Model (Model 1) Supporting Material for Chapter 5 Table A.5: Balance Statistics for Experiment Sample, Respondent Characteristics Control- Control- Effort- Attribute Control Effort Localism Effort Localism Localism Education 0.732 0.737 0.726 1 1 0.998 0.443 0.440 0.446 Employment 0.152 0.132 0.135 0.820 0.938 1 0.359 0.339 0.342 Agriculture worker 0.064 0.054 0.055 1.000 1.000 1 0.245 0.227 0.228 Correct MP name 0.839 0.838 0.840 1 1 1 0.368 0.369 0.367 Correct MP party 0.790 0.785 0.781 1 1.000 1 0.407 0.411 0.414 Close to party 0.718 0.729 0.709 0.999 1.000 0.666 0.450 0.444 0.454 DPP preference 0.558 0.553 0.518 1 0.205 0.212 0.497 0.497 0.500 MCP preference 0.349 0.342 0.359 1 1.000 0.957 0.477 0.474 0.480 UTF preference 0.057 0.069 0.081 1.000 0.814 1.000 0.231 0.253 0.272 Voted last election 0.730 0.737 0.739 1 1.000 1 0.444 0.440 0.439 Owns radio 0.395 0.392 0.391 1 1 1 0.489 0.488 0.488 Owns TV 0.092 0.082 0.091 1.000 1 1.000 0.289 0.275 0.288 Owns motor vehicle 0.107 0.093 0.087 0.991 0.856 1 0.309 0.290 0.282 Owns mobile 0.653 0.657 0.662 1 1.000 1 0.476 0.475 0.473 Asset index 1.246 1.224 1.232 0.991 1 1.000 1.061 1.026 1.035 Chewa 0.435 0.426 0.413 1.000 0.722 0.990 0.496 0.495 0.493 Lomwe 0.277 0.289 0.290 0.997 0.997 1 0.448 0.454 0.454 Yao 0.053 0.048 0.059 1 1 0.998 0.224 0.213 0.235 128 Table A.6: Balance Statistics for Experiment Sample, Conjoint Attributes Control- Control- Effort- Attribute Control Effort Localism Effort Localism Localism # of party 0.985 1.001 1.021 0.999 0.875 0.988 (incumbent) 0.816 0.811 0.815 Development 1.003 1.018 1.009 0.958 1 0.998 (incumbent) 0.808 0.818 0.810 Assistance 0.488 0.506 0.495 0.884 1 0.997 (incumbent) 0.500 0.500 0.500 Profession 1.032 1.005 1.003 0.973 0.713 1.000 (incumbent) 0.818 0.817 0.828 # of party 1.026 1.002 1.016 0.934 0.999 0.754 (challenger) 0.813 0.807 0.822 Development 0.989 1.008 1.027 0.893 0.758 0.901 (challenger) 0.827 0.816 0.824 Assistance 0.497 0.495 0.503 1 1 1.000 (challenger) 0.500 0.500 0.500 Profession 1.033 1.013 1.015 0.946 0.995 1 (challenger) 0.824 0.816 0.819 129 Table A.7: AMCEs for Personal Vote Choice Model 1 Model 2 OLS Logit Incumbent Attributes Active in Parliament 0.060*** 0.280*** (0.015) (0.068) Representation Focus 0.084*** 0.396*** (0.015) (0.068) 1st Party candidate (vs Independent) 0.059*** 0.281*** (0.014) (0.067) 2nd Party candidate (vs Independent) -0.021 -0.095 (0.014) (0.066) Education priority (vs Boreholes) -0.042** -0.194** (0.014) (0.065) Roads priority (vs Boreholes) 0.057*** 0.269*** (0.014) (0.066) High Assistance (vs Low) 0.235*** 1.062*** (0.012) (0.058) Major Business owner (vs Farmer) 0.052*** 0.247*** (0.014) (0.066) Teacher (vs Farmer) -0.003 -0.012 (0.014) (0.064) Challenger Attributes 1st Party candidate (vs Independent) -0.045** -0.208** (0.014) (0.067) 2nd Party candidate (vs Independent) 0.040** 0.186** (0.014) (0.065) Education priority (vs Boreholes) 0.039** 0.182** (0.014) (0.064) Roads priority (vs Boreholes) -0.050*** -0.234*** (0.014) (0.066) High Assistance (vs Low) -0.198*** -0.907*** (0.012) (0.056) Major Business owner (vs Farmer) -0.033* -0.151* (0.014) (0.065) Teacher (vs Farmer) 0.003 0.016 (0.014) (0.063) Woman Candidate 0.001 0.003 (0.012) (0.054) Num.Obs. 6717 6717 Std.Errors by: Respondent by: Respondent + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 130 Table A.8: AMCEs for Constituency Choice Model 3 Model 4 OLS Logit Incumbent Attributes Active in Parliament 0.047** 0.220** (0.015) (0.069) Representation Focus 0.077*** 0.364*** (0.015) (0.069) 1st Party candidate (vs Independent) 0.071*** 0.339*** (0.014) (0.065) 2nd Party candidate (vs Independent) -0.006 -0.029 (0.014) (0.065) Education priority (vs Boreholes) -0.034* -0.160* (0.014) (0.064) Roads priority (vs Boreholes) 0.036** 0.169** (0.014) (0.065) High Assistance (vs Low) 0.245*** 1.110*** (0.012) (0.057) Major Business owner (vs Farmer) 0.044** 0.209** (0.014) (0.067) Teacher (vs Farmer) -0.003 -0.026 (0.014) (0.064) Challenger Attributes 1st Party candidate (vs Independent) -0.049*** -0.230*** (0.014) (0.066) 2nd Party candidate (vs Independent) 0.055*** 0.258*** (0.014) (0.065) Education priority (vs Boreholes) 0.029* 0.139* (0.014) (0.064) Roads priority (vs Boreholes) -0.031* -0.148* (0.014) (0.066) High Assistance (vs Low) -0.198*** -1.011*** (0.012) (0.057) Major Business owner (vs Farmer) -0.036** -0.171** (0.014) (0.065) Teacher (vs Farmer) 0.003 0.007 (0.014) (0.064) Woman Candidate 0.001 0.005 (0.012) (0.055) Num.Obs. 6717 6717 Std.Errors by: Respondent by: Respondent + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 131 Supporting Material for Chapter 6 Figure A.2: Diagnostic Plots for Ch. 6 Main Model (Model 1) 0.15 1737 Cook's distance 0.05 0.10 3325 0.00 15 1737 Studentized residuals 2772 5 10 0 1504 3024 0.015 hat−values 0.010 0.005 0 500 1000 1500 2000 2500 3000 3500 Index 132