An economic analysis of revolutions, voting behavior and voter turnout

Rihab Roger Sawaya An Economic Analysis of Revolutions, Voting Behavior and Voter Turnout September 2021 Rihab Roger Sawaya An Economic Analysis of Revolutions, Voting Behavior and Voter Turnout Minho | 2021U Universidade do MinhoEscola de Economia e Gestão Rihab Roger Sawaya An Economic Analysis of Revolutions, Voting Behavior and Voter Turnout Universidade do MinhoEscola de Economia e GestãoSeptember 2021 Doctoral Thesis Ph.D. in Economics Work developed under the supervision of: Professor Doctor Linda Gonçalves Veiga and Professor Doctor Luís Aguiar-Conraria DIREITOS DE AUTOR E CONDIÇÕES DE UTILIZAÇÃO DO TRABALHO POR TERCEIROS Este é um trabalho académico que pode ser utilizado por terceiros desde que respeitadas as regras e boas práticas internacionalmente aceites, no que concerne aos direitos de autor e direitos conexos. Assim, o presente trabalho pode ser utilizado nos termos previstos na licença abaixo indicada. Caso o utilizador necessite de permissão para poder fazer um uso do trabalho em condições não previstas no licenciamento indicado, deverá contactar o autor, através do RepositóriUM da Universidade do Minho. Atribuição-NãoComercial-SemDerivações CC BY-NC-ND https://creativecommons.org/licenses/by-nc-nd/4.0/ ii Acknowledgments I would like to express my deep and sincere gratitude to my supervisors, Professor Doctor Linda Gonçalves Veiga and Professor Doctor Luis Aguiar-Conraria for their invaluable guidance, help, and instructive comments throughout my Ph.D. journey. The completion of this Ph.D. dissertation would not have been possible without their great support. I also acknowledge the vice director of the Ph.D. program Professor Doctor Miguel Portela for his availability in answering my questions and for the financial support he provided me to attend a summer school at the university. I am highly appreciative to Mrs. Estela Vieira who accompanied me since the first day at the program and was always available to answer my questions and help me in all possible ways. Along this journey, I made some good friends and classmates who were of good help and company. In particular, I would like to thank Samer, Bassem, Maher, Marta, Ziad, Georges, Izzat and Moustapha. I acknowledge the administration of the Antonine University and the Antonine School of Business for their continuous support and flexibility. I am also grateful to my colleagues and friends who have supported me along the way. In particular, I would like to thank Sandra and Saïd for the discussions we had that allowed me to have a better understanding of the Lebanese political context. I owe a special thank you to my cousin Nabil for his help in translating the abstract into Portuguese. Above ground, I am indebted to my family, my mother Nejmeh, my late father Roger and my brother Nibal whose value to me only grows with age. I am extremely grateful for their love and sacrifices. Finally, I am very much thankful to my husband Nadim for his unconditional support, love and patience to complete this research work. iii In memory of my father Roger A. Sawaya My inspiration and pride; yesterday, today and forever iv Statement of Integrity I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho. v Uma Análise Económica de Revoluções, Comportamento Eleitoral e Participaçao Eleitoral Resumo Esta tese compreende três artigos na área da Economia Política. O primeiro analisa as determinantes de revoluções bem-sucedidas (definidas como processos que levam à saída de um líder do cargo, com base em grande participação popular (Goemans et al., 2016)) num grande número de países; o segundo e o terceiro estreitam a análise a um país em particular do Oriente Médio, o Líbano, onde o comportamento e a participação eleitoral são respectivamente estudados. O primeiro artigo examina o impacto de fatores relacionados com a economia, a política, os meios de comunicação, a governança e as tecnologias da informação e comunicação (TIC) sobre a probabilidade de revoluções bem-sucedidas, em mais de 150 países, no período de 1996 a 2015. Os resultados confirmaram que o rendimento per capita e o crescimento do PIB real, em 5 anos, diminuem a probabilidade de revoluções bem-sucedidas. Demonstrámos também que novos fatores que emergiram das mais recentes análises da Primavera Árabe tiveram impacto na ocorrência de revoluções bemsucedidas nos países Árabes. Menos censura dos meios de comunicação social e melhorias na governança e nas TIC diminuíram a probabilidade de revoluções bem-sucedidas. Também há evidência de que uma indústria petrolífera bem-sucedida gera um efeito estabilizador, particularmente em países não democráticos. O segundo e o terceiro artigo focam-se no caso Libanês. O propósito do segundo é estimar os efeitos de fatores religiosos, socioeconómicos e políticos nas escolhas partidárias dos eleitores Libaneses. O Líbano tem um sistema eleitoral multiconfessional e pluripartidário. Independentemente de sua afiliação religiosa, os eleitores votam para todos os cargos do distrito, num único boletim eleitoral. Podem escolher tantos candidatos quanto o número de cargos atribuídos ao distrito eleitoral e podem remover ou adicionar candidatos de uma lista pré-estabelecida — desde que o equilíbrio sectário da lista se mantenha. Sendo o Líbano um país multiconfessional, diverso e pluralista, não é surpresa que fatores religiosos sejam estatisticamente significativos. Não obstante, a economia continua a ser um fator relevante: distritos eleitorais com elevados níveis de rendimento tendem a votar mais no Hezbollah e no Movimento vi do Futuro e menos em Independentes; aumentos na taxa de desemprego penalizam o Partido Social Nacionalista Sírio. Os Independentes obtêm melhores resultados eleitorais quando o número de ministros Independentes é maior. Adicionalmente, como esperado, o número de bombardeamentos e assassinatos tem um impacto significativo nos resultados eleitorais. Os candidatos independentes são penalizados em distritos que testemunham mais bombardeamentos e assassinatos em anos eleitorais. Finalmente, o terceiro artigo estuda o impacto de fatores económicos, políticos, sociodemográficos e institucionais nas taxas de participação eleitoral, nos 26 distritos administrativos Libaneses, durante as eleições parlamentares que ocorreram entre 1996 e 2018. Estimando o sistema GMM, concluímos que quer um aumento da margem de vitória quer do desemprego diminuem a participação eleitoral. Palavras-chave: Votação económica; Participação eleitoral; Revoluções; Desempenho económico; Governança; Religião.. xiii B 1 Candidates and winners in the 1996 elections .......................................................................... 162 B 2 Candidates and winners in the 2000 elections .......................................................................... 165 B 3 Candidates and winners in the 2005 elections .......................................................................... 168 B 4 Candidates and winners in the 2009 elections .......................................................................... 172 B 5 Number of seats won (and their change) per election year ........................................................ 176 B 6 1996 parliamentary elections results per district ....................................................................... 178 B 7 2000 parliamentary elections results per district ....................................................................... 179 B 8 2005 parliamentary elections results per district ....................................................................... 180 B 9 2009 parliamentary elections results per district ....................................................................... 181 B 10 Political parties in office during election years ......................................................................... 182 xiv List of Figures Figure 1 Conditional marginal effects of democracy over log real GDP per capita ............................... 32 Figure 2 Conditional marginal effects of democracy over 5-years average rate of real GDP growth ...... 33 xv List of Abbreviations AMAL Amal Movement CAS Central Administration of Statistics CIA Central Intelligence Agency EU European Union FE Fixed Effects model FM Future Movement GDP Gross Domestic Product GMM Generalized Method of Moments HEZ Hezbollah ICT Information and Communications Technology IFES International Foundation for Electoral Systems IND Independents Inf. Int. Information International LADE Lebanese Association for Democratic Elections LCP Lebanese Communist Party LCPS Lebanese Center for Policy Studies LL Log-Likelihood LPOS Lebanon Public Opinion Survey M14 Independents of March 14 coalition M8 March 8 coalition MENA Middle East and North Africa MEPV Major Episodes of Political Violence MOIM Ministry of Interior and Municipalities MOSA Ministry of Social Affairs NDI National Democratic Institute OLS Ordinary Least Squares PC Principal Component PM Prime Minister POLS Pooled Ordinary Least Squares xvi POL Political PRIO Peace Research Institute Oslo PSP Progressive Socialist Party RE Random Effects model SD Standard Deviation SOCECO Socio-economic SSNP Syrian Social Nationalist Party SSR Sum of Squared Residuals UN United Nations UNDP United Nations Development Program UNICEF United Nations International Children’s Emergency Fund USD Urban Social Disorder VIF Variance Inflation Factor WDI World Development Indicators ZIP Zero Inflated Poisson ZINB Zero Inflated Negative Binomial 1 INTRODUCTION For some earlier scholars, the term, political economy, applied to the whole field of economics, and, in some cases, it overlapped with development thinking. Nevertheless, in the 1950s and 60s, two distinctive movements in political economy were taking place in the Virginia and Chicago schools, opening the floor to debates on a range of important topics, such as political institutions, distortions in the policy process, and government failure. In general, the critical issues in the field of political economy focus, among others, on the role of institutions in resolving conflicts of interest and promoting common interest, and on incentives and selection in institution design (Besley, 2016). Political economy as a research field embraces a wide range of foci, such as the development of laws and social relations in material production; the organization of a system of consumption and production; or the application of economic logic to political behavior (Yandle, 1990). However, in recent years, the interest in understanding how the policy process affects policy outcomes has been increasing. Today, political economy research ranges from collective decision-making mechanisms to elections, political business cycles, political parties, institutions and economic growth, institutions and distribution and redistribution, and their role in conflicts and violence. This research focuses on three main topics in political economy: revolutions and the economy, voting behavior and political parties, and voting turnout. First, we reevaluate the economy's role in successful revolutions before examining new determinants in an extensive range of countries, including the recent Arab uprisings in the Middle East. Then, we focus on a particular Arab country, Lebanon, where we investigate the power of economics in explaining the vote share of six main Lebanese political parties and independents, before analyzing the factors that explain turnout in the last five parliamentary elections that took place in Lebanon. The problem of identifying the preconditions of revolutions has long been a concern for political scientists. Plato and Aristotle argued that poverty might be a cause of political revolution, while Tocqueville and Brinton assert that revolutions are preceded by a significant economic development increase (Tanter & Midlarsky, 2013). A comprehensive explanation of these two approaches was advanced by Davies (1962), who explained that revolutions are preceded by steady long term increases in economic development, followed by a sharp reversal just before the revolution's outbreak. 2 For scholars of modernization, economic development increases the individual capacity to engage in political action and democracy (Lipset, 1959). Such modernization could lead to rising and unmet expectations, popular frustration, and violent revolutions in societies that lack the strong political institutions necessary to absorb the mobilized masses into politics (Huntington, 1968). In general, higherincome should increase demands for democratization, thus providing a fertile climate for revolutions in autocracies. Collier & Rohner (2008) found that income growth makes democracies safer, but it also makes autocracies more prone to violence. In recent years, almost all scholars have emphasized the structure of economic relations, the importance of existing grievances, and political ideologies' role in igniting violent conflicts (Boix, 2008). The economy features in the development of almost every type of social conflict; however, the relatively recent Arab uprisings, where large crowds demanded regime change and, in some cases, old regimes were eventually forced out, opened the space for new factors to be investigated. These factors are diverse. Some authors emphasize social media's role in triggering these uprisings (Mortada & Salem, 2012; McGarty et al., 2013; Soengas, 2013; Tudoroiu, 2014), as it gives people a new space of freedom to frame their grievances. Others point out how the lack of governance, the rule of law, corruption, and cronyism help sustain authoritarian regimes (Sika, 2012), increasing the latter's chances of facing revolutions (Tiruneh, 2014). These factors have never been considered concomitantly in a model explaining the outbreak of revolutions; thus, an opportunity has arisen to reevaluate the triggers previously advanced in explaining the causation of revolutions (mainly economic factors) and exploring new ones. This scientific investigation's relevance is that it elucidates the main causalities behind either the observation of the event – in the present context, the case of successful revolutions – or their absence. The first paper, which we develop in Chapter 1, "New Determinants of Successful Revolutions, " will seek to extend Collier and Rohner's (2008) work and test it across many countries for the period 1996-2015. This study contributes to the existing literature by considering various variables ranging from economic, political, media, and governance factors to information and communications technology (ICT). We define our dependent variable, ‘successful revolutions,' as processes leading to a leader’s exit from the office, based on sizeable popular participation (Goemans et al., 2016). The Random-Effects Probit method will be considered the primary estimation method to deal with our dichotomous dependent variable. Another technique, the ‘Poisson regression model,' will be used when estimating other forms of social unrest –– such as the total number of demonstrations and riots, attempted coups d’état, and individual events of 3 violence against the government –– that are observed as a count number. The chapter also discusses the social revolutions that break out in some Arab countries. Our investigation in Chapter 2 focuses on a particular Arab country, Lebanon, where we analyze which factors influence a party’s vote share at the district level in four parliamentary elections. Earlier literature on voting behavior shows that voters respond to economic conditions, mainly inflation and unemployment, where they assess the competency of the government with regards to economic conditions (Kramer, 1971; Goodman & Kramer, 1975; Tufte, 1975, 1978; Hibbs, 1982; Lewis-Beck & Rice, 1992; Kiewiet & Udell, 1998). This hypothesis's support has been mainly found in two-party system countries, while in multi-party system countries, voters evaluate political parties based on their policies (Swank & Eisinga, 1999). Other factors might affect a voter's choice, such as religious affiliation. Voters tend to choose a candidate from their own faith (Kane, Craig, & Kenneth, 2004; Botterman & Hooghe, 2009; Campbell et al., 2011). If there were a country where one would not be surprised to find the religious factor to be relevant, it would be Lebanon. It is one Arab country of particular interest to this study. It is the most diverse, multi-confessional, and pluralistic country in the Arab and Middle Eastern regions. It has more than 18 religious sects and more than 30 political parties. This diversity is reflected in its constitution and electoral and political systems, the structure of which stipulates that the president’s position is held by a Maronite Christian, the premiership by a Sunni Muslim, and the Speaker of the House by a Shiite Muslim. Parliamentary seats are held equally between the Muslim and Christian communities and are distributed among the confessional groups of both communities across Lebanon's different geographical regions. Another exciting feature of Lebanon is its political instability amid turmoil in the region. Despite these particularities, studies on Lebanon's economic voting are very scarce (Harik, 1980; Hourani & Sensenig-Dabbous, 2012; Lebanese Association for Democratic Elections, 2014). These studies, which focus on one parliamentary election without accounting for each political party's specificities in explaining their vote share, show that political and religious factors were predominant in affecting the voter’s choice. This makes Lebanon a fertile ground for political economy studies in general and voting behavior in particular. Therefore, we contribute to the empirical analysis of voting behavior by estimating the effects of economic, religious, and political factors on party choice. To our knowledge, this chapter is the first to empirically analyze the factors influencing political parties’ vote shares in the four parliamentary elections, covering the 26 administrative districts in Lebanon, that have taken place since the civil war – 1996, 2000, 2005, and 2009. Six major political parties and independents are considered: Amal, Hezbollah, Lebanese Communist Party, Progressive Socialist Party, Future Movement, Syrian Social Nationalist Party, Independent of March 14 coalition and Independents. We estimate a different model for each party 4 based on the most relevant variables. The pooled OLS method is used with robust standard errors. Given the small number of observations, a series of tests are performed to check for multicollinearity and assess the sample size problem's severity. The impact of economic, sociodemographic, political, and institutional factors on the turnout rates, in 26 administrative districts, over the Lebanese parliamentary elections between 1996 and 2018 is the theme of Chapter 3. The reasons explaining why people participate in elections have been largely explored. Early studies advanced factors related to partisanship, sense of duty (Fiorina, 1960; Riker & Ordeshook, 1968), and cost-benefit calculation, where the expected benefits of voting should outweigh the costs (Downs, 1957b). There is currently a widespread consensus that economic, sociodemographic, political, and institutional factors have more explanatory power in explaining turnout. In particular, turnout in national elections is explained more by campaign expenditures, election closeness, and registration requirements. In contrast, population size and composition, simultaneous elections, previous turnout levels, the electoral system, and economic development play a more critical role in subnational elections (Cancela & Geys, 2016). Economically more advanced countries have a higher turnout rate (Blais & Dobrzynska, 1998; Fornos et al., 2004). However, the discussion of factors justifying why people turn out to polls has been mainly conducted in well-established western democracies. Very few studies have been piloted in the Arab countries and even less in Lebanon (Harik, 1980; Alhaber, 2007; de Miguel et al . , 2015; Mourad & Garrote Sanchez, 2019; Garrote Sanchez, 2021). Factors such as personal and political ties between voters and politicians (Harik, 1980), patronage at the level of low-income citizens, on which leaders of confessional communities depend to get more votes (Miguel et al . , 2015), voter age, unemployment, political affiliation, and partisanship were among the most important determinants of turnout (Mourad & Garrote Sanchez, 2019). The very few studies focusing on Lebanon open the possibility further to explore the determinants of voter turnout in this country. This paper contributes to the literature on voter turnout by empirically studying, for the first time, all elections that took place at the end of the civil war (except for 1992). It differs from previous studies in several ways: by aggregating the analysis over five parliamentary elections and 26 administrative districts; by considering elections taking place under different electoral systems; and, by expanding the set of explanatory variables of voter turnout in Lebanon to variables such as political fragmentation and the winning margin of the difference in the percentage of votes between the most voted list and the second most voted list. System-GMM is the primary estimation method used in this paper. Fractional Probit and Beta regression were also used as alternative estimation methods. Several tests were implemented to check the robustness of the regressions. 5 The last part concludes this dissertation. It presents and discusses the main results of the previous three chapters. The conclusion also addresses the main limitations faced and suggests some directions for future research. 6 CHAPTER 1: NEW DETERMINANTS OF SUCCESSFUL REVOLUTIONS Abstract This paper examines the impact of a range of factors on the probability of successful revolutions in more than 150 countries over the 1996-2015 period; the factors selected for this study are drawn from economic, political, media, governance, and information and communications technology (ICT) domains. Results affirm the effect of income per capita and 5 years real GDP growth in lowering the probability of successful revolutions. It also shows that new factors emerging from recent analysis of the Arab uprisings, and occurring prior to the events in the Arab countries, had impacted successful revolutions. Less media censorship, and an increase in governance and ICT lower the probability of successful revolutions. There is also evidence that oil has a particularly stabilizing effect in non-democratic countries. Keywords: Successful revolutions; GDP; Media censorship, Oil; Governance; ICT. 1.1. Introduction At the close of the Cold War, the world observed a technological and economic advancement that would prove to be unsustainable alongside, and in addition to, the Eurozone debt crisis, the occurrence of a number of financial crises – the last being that of 2007. The political scene was also substantially marked by the rise of world terrorism and its retaliatory repercussions, such as the wars in Iraq and Afghanistan, and the more recent Arab uprisings. Today, political instability has been revealed as a pervasive global phenomenon, with an increasingly deteriorated global level of peace (Institute for Economics and Peace, 2018). The emergence of these confrontations necessitates new inquiries, not least of which is the need to determine the factors that trigger revolutions. According to the economic theory of revolutions, known as the ‘by-product theory’, self-interest is one of the prime causal factors motivating people to participate in a revolution (Tullock, 1971). Discontent about economic conditions increases hostility towards the government and therefore increases revolutionary participation. Such discontent might be triggered by one or more of the following: excessive taxation and inflation (Cartwright et al., 1985); a decrease in the rate of income growth (Breton, 1969; McGuire, 1981; Collier & Hoeffler, 2005); expenditure cuts (Ponticelli & Voth, 2011); and/or a high level of inequality (Acemoglu & Robinson, 2001). However, in democratic societies, the probability of violent revolutions tends to be smaller (Collier & Rohner, 2008; Tiruneh, 2014), especially when these societies benefit 13 protesters (Acemoglu et al . , 2014), building networks, organizing and publicizing social protests, and in presenting the uprisings and conflicts to the world (Khondker, 2011). However, social media was unable to create the social ties needed to take the steps that a revolution requires. It was only a tool of coordination and communication (Soengas, 2013). According to Tudoroiu (2014, p.15), despite all its importance, social media “would have been unable to bring social mobilization and political change in the absence of the structural socio-political causes of revolutions”. 1.3. Empirical model and data Our baseline empirical model is built upon that of Collier and Rohner (2008) regarding the choice of control variables, the method, and the estimation approach. The model is described as follows: , , 1 , 1 , 1 , 1 , 1 , Rev SOCECO POL MEDIA Governance PC ICT PC i t i t i t i t i t i t i t                    (1) i = 1… 193; t = 1996 to 2015 Where i represents a country, and t represents a year. SOCECO is a vector of socio-economic variables; POL is a vector of political variables; MEDIA is an indicator of media censorship; Governance PC is the principal component of governance indicators; ICT PC is the principal component of Information and Communications Technology indicators; and ,it  represents the error term. , θ ,  and σ are parameters to be estimated, and  and  are vectors of the parameters to be estimated. 1.3.1. Variables and data sources The dependent variable is the revolution indicator. Many operationalizations of revolution were used in the literature. Some empirical studies refer to the Polity dataset to measure revolution. Mainly, they use the change in the composite Polity score (Maoz, 1989; Maoz, 1996; Enterline, 1998; Mansfield & Snyder, 2005) and durable variable (Smith, 2004; Morrison, 2009). Such specifications focus more on democratization rather than on revolution, and in many cases, they fail to separate the concept of the “democratic-ness” and the continuity of a regime (Colgan, 2012). Cartwright et al . (1985) considered revolutionary activity as anti-regime type wars fought within the territories of a country against the government in power to overthrow it. They excluded from their analysis the tribal wars and border wars. Such a definition is close to what (Marshall et al . , 2017) considers as the “Revolutionary War”. The latter is defined as “episodes of violent conflict between governments and politically organized groups (political 14 challengers) that seek to overthrow the central government, to replace its leaders, or to seize power in one region” (Marshall et al . , 2017, p.5) Others have used the Cross-National Time-Series Data Set compiled by Arthur Banks (Fearon & Laitin (2003); Collier Rohner (2008); Ponticelli & Voth (2011); Knusten (2014). The latter considers revolutions as “any illegal or forced change in the top governmental elite, any attempt at such a change, or any successful or unsuccessful armed rebellion whose aim is independence from the central government” (Colgan, 2012, p.8). Such a definition includes events that are not accompanied by the transformation of the existing social, economic, and political relationships with the state. Another dataset on revolutions, the Archigos, compiled by Goemans et al. (2016b) and used recently by Knusten (2014), considers successful revolutions as processes leading to a leader’s exit from office, based on large-scale popular participation. We will refer to this dataset in this paper to identify successful revolutions that took place during the period under study (1996-2015). Our dependent variable will, therefore, be ‘successful revolutions’, extracted from the Archigos dataset (Goemans et al., 2016b). This dummy variable takes the value 1 in any of the following cases: leader lost power as a result of domestic popular protest; leader removed by domestic rebel forces; leader removed by other government actors; or leader removed by the military without foreign support. Forty-seven cases of successful revolutions due to the irregular exit of the leader from office are observed.1 However, Fiji witnessed two revolutions in the same year (2000), but since the dependent variable is a yearly one, only 46 successful revolutions will be counted (see Appendix A 1). The right side of equation 1 includes the following: a vector of socio-economic (SOCECO) and political (POL) variables; governance (Governance PC); information and communications technology principal component (ICT PC); and, a media censorship (MEDIA) indicator. The lagged vector of socio-economic variables SOCECO considers proxies of economic performance, inequality, human development, educational attainment, and the importance of oil production. Regarding economic performance, the following variables are considered: the logarithm of real GDP per capita using constant 2010 US$ ( Log real GDP per capita ); the average real GDP growth rate over the last five years ( 5-years average rate of real GDP growth );2 inflation rate (Inflation %) ; and, the unemployment rate ( Unemployment % ) . Data for 1 We excluded irregular exit forms with foreign support. 2 Real GDP per capita growth rate and real GDP growth rate were tested instead of 5-years average rate of real GDP growth rates, but they did not report any statistically significant results. 15 GDP inflation and unemployment is extracted from the World Development Indicators of the World Bank (World Bank, 2018). The Gini index 3 (extracted from Provcal Net of the World Bank) is used as a proxy for inequality. To measure human development, we use the Human Development Index (HDI) from the United Nations Development Programme (UNDP). The HDI considers achievements in three key dimensions: a long and healthy life, access to knowledge, and a decent standard of living. The HDI is the geometric mean of normalized indices for each of the three dimensions during year t-1 (UNDP, 2016). A high HDI indicates a high performance in the three aspects of human development. The level of educational attainment is proxied by the % of school enrollment in primary education, which is the cumulative percentage of the population over 25 years of age that has at least completed primary education. The data is extracted from the World Bank Development Indicators (World Bank, 2017a). The size of the population is proxied by Log population . A dummy variable to identify oil exporter countries is included ( Oil exporter ). This variable takes the value 1 if more than 33% of the country’s exports are derived from oil, and 0 otherwise. An alternative measure is also used: Fuel exports is the percentage of oil exports of the total merchandise in year t-1 . Data is extracted from the World Bank Development Indicators (World Bank, 2017a). Finally, the effect of globalization is captured by the revised version of the KOF globalization index , a composite index covering the economic, social, and political integration aspects which distinguish between de facto and de jure measures along the different dimensions of globalization (Gygli et al., 2019). KOF de facto globalization measures actual international flows and activities; KOF de jure globalization measures policies and conditions that, in principle, enable, facilitate and foster flows and activities. The overall KOF globalization index (KOF Global Index) combines de facto and de jure measures. The original version of KOF globalization index was first developed by Dreher (2006) and updated in Dreher et al . (2008). KOF globalization index was used to investigate the impact of globalization on ethnic violence (Bezemer & Jong-A-Pin, 2013), political stability (Bergh et al., 2014), terror attacks (Gassebner et al . , 2011), among other factors.4 The risk of rebellion and revolutionary participation may be increased by: lower income growth (Breton & Breton, 1969; McGuire, 1981; Collier & Rohner, 2008; Knusten, 2014); high unemployment percentage (Ansani & Daniele, 2012); excessive inflation (Cartwright et al . , 1985); high level of inequality (Acemoglu & Robinson, 2001); and, large population size (Collier & Hoeffler, 1998). As for education, the effect on the probability of revolution is ambiguous. While Lipset (1959) found that the occurrence of revolution is 3 The Gini index measures the extent to which the distribution of income of a country i in year t-1 (or, in some cases, consumption expenditure) among individuals or households within an economy deviates from a perfectly equal distribution. 4 For a revision of the literature of papers using previous versions of KOF Globalization index see (Potrafke, 2015) 16 negatively correlated with the level of educational attainment, Huntington (1968) argued that, when people become more educated and urbanized as a result of economic development, they will demand greater political participation and more civil rights. If such popular demands are not addressed, discontent will surface, and political violence, including revolution, is possible. The lagged vector of political variables (POL) includes two variables: the indicator of the regime type; and neighbors with violence. The indicator of the regime type refers mainly to the binary variable, democracy, which takes the value of 1 if the country is democratic (POLITY 2 score greater or equal to 6), and 0 if it is autocratic. Other indicators were used as alternative measures for the robustness check: Polity 2 is computed by subtracting the autocracy (AUTOC) score from the democracy (DEMOC) score; the resulting unified Polity scale ranges from +10 (strongly democratic) to -10 (strongly autocratic). In recent years, the Polity 2 variable has come under considerable criticism as it attributes similar scores in quite different situations (F. J. Veiga et al., 2017). A new six-fold classification of regimes was developed by Cheibub et al . (2010). They define democracy as regimes in which governmental offices are filled as a consequence of contested elections. The new six-fold classification variable, DD, takes the following values: 0 for parliamentary democracy, 1 for mixed (semi‐presidential) democracy, 2 for presidential democracy, 3 for civilian dictatorship, 4 for military dictatorship, and 5 for royal dictatorship. The data is available up to 2008. We only consider a binary variable DD that takes the value 1 for parliamentary democracy, for mixed (semi‐presidential) democracy, and for presidential democracy; and 0 for the remaining three cases. We expect less successful revolutions to occur in more democratic countries. Another political variable was used: Neighbors with violence. It captures the possible impact of conflict in a neighboring state on the outbreak of revolution in a country i. The security of any state is affected not only by its conflict dynamics and outbreaks of Major Episodes of Political Violence (MEPV) but, also, by those in immediate proximity, mainly in neighboring states –– that is, states sharing a contiguous land or water border of two miles width or less (Marshall, 2017a). The variable neighbors with violence measures the number of bordering countries with any type of violence (societal or interstate). The dataset is extracted from the Center for Systemic Peace and covers the period 1996-2012. Finally, the media indicator (media censorship) is the freedom of the press score calculated by Freedom House ( Freedom of the press ) . Scores range from zero to 100 with ‘free’ going from 0-30, ‘partially free’ from 31-60, and ‘not free’ from 61-100. These scores are extracted from Freedom House (2017). We expect more successful revolutions to increase with more media censorship. 17 Governance PC5 and ICT PC refer to the principal components of the indicators of the vectors of governance, and information and communications technology, respectively. A principal component analysis was conducted to find components that express maximum information from the above two categories of indicators. The eigenvectors displayed by the principal component analysis are returned in orthonormal form –– that is, uncorrelated and normalized. Governance indicators include measures for: voice and accountability; political stability and absence of violence; government effectiveness; regulatory quality; the rule of law; and, corruption. These indicators are extracted from the Worldwide Governance Indicators of the World Bank (World Bank, 2017b), and they all range from approximately -2.5 (weak performance) to 2.5 (strong governance performance). The following indicators are considered: - Political stability and absence of violence/terrorism measures perceptions of the likelihood of political instability and/or politically-motivated violence, including terrorism. - Government effectiveness reflects: perceptions of the quality of public services; the quality of the civil service and the degree of its independence from political pressures; the quality of policy formulation and implementation; and, the credibility of the government's commitment to such policies. - Regulatory quality reflects: perceptions of the ability of the government to formulate and implement sound policies; and, regulations that permit and promote private sector development. - Rule of law reflects perceptions of the extent to which agents have confidence in and abide by the rules of society, in particular: the quality of contract enforcement; property rights; the police; and the courts; and, as well, the likelihood of crime and violence. - Control of corruption reflects perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ‘capture’ of the state by elites and private interests. 5 PC refers to Principal Component 18 The indicators of the vector of information and communications technology include: fixed-broadband subscriptions; fixed telephone subscriptions; mobile cellular subscriptions; and, internet users for country i at year t-1 . Data is extracted from the World Bank.6 Tables 1 and 2 include, respectively, the correlation between the two principal components and each of the variables. Table 3 presents the descriptive statistics of the variables used in the empirical work. 6 The term, ‘fixed broadband subscriptions’ refers to “fixed subscriptions to high-speed access to the public Internet (a TCP/IP connection), at downstream speeds equal to, or greater than, 256 kbit/s. Fixed telephone subscriptions refers to “the sum of the active number of analogue-fixed telephone lines, voiceover-IP (VoIP) subscriptions, fixed wireless local loop (WLL) subscriptions, ISDN voice-channel equivalents and fixed public payphones” (World Bank, 2017a). Mobile cellular telephone subscriptions are “subscriptions to a public mobile telephone service that provide access to the PSTN using cellular technology. Internet users are individuals who have used the Internet (from any location) in the last 12 months (World Bank, 2017a). 19 Table 1 Correlation between the principal component of governance and each of the governance variables Voice and accountability Political stability and absence of violence Government effectiveness Regulatory quality Indicator of the agent’s confidence and abidance by the rules Indicator of control of corruption by the public authority Principal component of governance 0.8749 0.8094 0.9568 0.9329 0.9779 0.9537 Source: Author’s computations based on data extracted from the World Bank. Table 2 Correlation between the principal component of ICT and ICT variables % Fixed broadband subscriptions % Fixed telephone subscriptions % Mobile subscriptions % Individuals using the internet Principal component ICT 0.8587 0.7913 0.7590 0.9548 Source: Author’s computations based on data extracted from the World Bank. 20 Table 3 Descriptive statistics (1) (2) (3) (4) (5) VARIABLES N Mean Sd Min max Dependent variable Successful revolution 3,860 0.0119 0.109 0 1 Socioeconomic variables Log real GDP per capita 3,650 3.638 0.657 2.064 5.159 5-years average rate of real GDP growth 3,728 0.0408 0.0448 -0.524 0.568 Unemployment % 2,034 0.0929 0.0611 0.0048 0.700 Inflation % 3,343 0.108 0.915 -0.181 41.45 Log population 3,857 6.693 0.937 3.965 9.135 KOF globalization index de facto 3,670 0.537 0.155 0.198 0.915 KOF globalization index de jure 3,750 0.585 0.164 0.174 0.923 KOF globalization index 3,670 0.562 0.154 0.214 0.907 Fuel exports 2,882 0.168 0.275 0 0.999 Oil exporter 2,886 0.176 0.381 0 1 % of school enrollment in primary education 2,245 0.878 0.144 0.218 1 Gini index 1,114 0.387 0.0948 0.162 0.658 Human development indicator 3,409 0.654 0.166 0.230 0.948 ICT indicators % of mobile subscriptions 3,785 0.470 0.479 0 2.078 % of fixed telephone subscriptions 3,782 0.186 0.196 0 1.327 % of fixed broad band subscriptions 2,255 0.0720 0.104 0 0.467 % of individuals using the internet 3,622 0.201 0.253 0 0.982 Governance indicators Voice and accountability index 3,075 -0.0463 1.008 -2.313 1.801 Political stability and absence of violence 3,027 -0.0480 0.998 -3.315 1.760 Government effectiveness 3,006 -0.0646 0.990 -2.446 2.437 Regulatory quality 3,007 -0.0778 0.986 -2.645 2.233 Indicator of the agents confidence 3,073 -0.0666 0.993 -2.606 2.100 Indicator of the control of corruption by the public authorities 3,015 -0.0655 0.998 -1.869 2.470 Media indicator Media censorship 3,770 0.467 0.245 0.0500 1 Political variables Polity 2 score 3,203 3.324 6.506 -10 10 Polity score 3,215 0.332 16.98 -88 10 Democracy 3,203 0.534 0.499 0 1 DD 2,645 0.574 0.495 0 1 Number of neighboring states with any type of political violence 2,765 0.781 1.064 0 7 Sources: (Cheibub et al . , 2010; Goemans & Chiozza, 2016b; Marshall, 2017a, 2017b; Marshall et al., 2017; World Bank, 2017a; 2017b; 2018; Freedom House, 2017; Croicu & Sundberg, 2017; Gygli et al., 2019) 21 1.4. Methodology Since our measure of revolution ( successful revolutions ) is dichotomous, taking the value of 0 or 1, linear regression methods are inappropriate. This is because estimates of a linear model are unbounded. They can produce predictions less than 0 or greater than 1. It is possible to estimate a model with a dichotomous dependent variable by OLS, by arbitrarily constraining predictions points to either 0 or 1. Still, the error term cannot satisfy the assumption of homoscedasticity. Furthermore, the residual will not be normally distributed. This implies that inference in small samples cannot be based on the usual suite of normality-based distributions such as the t-test (Greene, 2010). To address the problems mentioned above, two nonlinear binary response models are commonly used in the literature: the logistic distribution, yielding the logit model, and the standard normal distribution, yielding the probit model. The logistic distribution is similar to the normal distribution except in the tails, which are heavier. Therefore, both can produce similar predicted probabilities except when the sample contains very few 1’s or 0’s, and when there is a very wide variation in an important independent variable (Greene, 2010).7 In his survey of qualitative response models, Amemiya (1981) noted that probit and logit models usually give similar results, and it is difficult to distinguish them statistically. He presented several scalar criteria such as the number of wrong predictions; the sum of squared residuals (SSR); SSR weighted by estimated probabilities; squared correlation coefficient; log-likelihood function; and, Akaike Information Criterion. Such criteria can be used to favor one model over the other. However, this question remains unresolved, and advice on how to choose is sparse (Greene, 2003). Logit and probit methods are commonly used to study the onset of revolutions or other types of social unrest. Fearon and Laitin (2003) and Collier and Rohner (2008) used the logit method to study the onset of civil war. The probit method was used by Collier and Hoeffler (1998; 2005) and Collier (2000) to study the onset of civil wars and coups. Knusten (2014) studied revolutionary attempts with a panel fixed-effects logit method and a random-effects probit method, and successful revolutions with a panel random-effects logit method. 7 Despite the fact that the sample contains considerably more zeros than ones, random-effects probit and logit methods generate similar results. For parsimonious reasons we only report the results for the random-effects probit method. 22 Since the majority of countries included in our dataset did not witness any successful revolution, the fixedeffects model will discard information from these countries. Therefore, we will refer to the random-effects probit model. 1.5. Empirical results Table 4 shows the results of the baseline model: four socioeconomic variables (log real GDP per capita, 5years average rate of real GDP growth, oil exporter, and log population); one political variable (democracy); and, the media censorship indicator. All variables are lagged one period. These variables were selected because they have been frequently used in previous studies of revolutions,8 and they maximize the size of our sample (2,505) and the number of observations, where the dependent variable equals 1 (33 cases of successful revolutions).9 To test for a possible high correlation between two or more predictors, several multicollinearity diagnostic measures were applied (VIF, tolerance, eigenvalues, condition index, and R-squared).10 The Mean Variance Inflation Factor (VIF) was 1.69, with the highest being equal to 2.84 less than the ‘rule of thumb’ 10, and the condition number11 was 3.1996 for the baseline model, which falls within the accepted ‘rule of thumb’ range for concern. The estimated coefficients of the RE probit model are expressed as odds ratios and are often difficult to interpret from a practical point of view. Usually, models with a binary dependent variable using probit regression report marginal effects instead of odds ratios. The marginal effects of an explanatory variable are calculated as the partial derivative of the dependent variable with respect to this explanatory variable. In the tables, we report the marginal effects at the sample mean. As for binary dependent variables where the derivative with respect to a small change might be inappropriate, the following marginal effect is used: 8 5-years average rate of real GDP growth will be used for the first time. Previous studies used average annual growth over 15, 20, and 25-year periods. Media censorship will also be used for the first time. 9 Bangladesh (2007), Cameroon (1997), Central African Republic (2003 and 2013), Côte d’Ivoire (1999 and 2000), Ecuador (2000), Egypt (2011 and 2013), Fiji (2006), Gabon (1999, 2003 and 2009), Georgia (2003 and 2007), Guinea (2009), Honduras (2009), Kyrgyz Republic (2005 and 2010), Lebanon (2008), Madagascar (2002 and 2009), Mali (2012), Mauritania (2008), Niger (1996, 1999 and 2010), Pakistan (1996 and 1999), Thailand (2006 and 2014), Tunisia (2011), Turkey (1997) 10 By using the Collin command of Stata and centering the data. 11 Computed without a constant. 29 Table 6 Estimating the baseline model with each of the World Governance Indicators (1) (2) (3) (4) (5) (6) (7) Successful Revolutions VARIABLES Voice and accountability Political stability Government effectiveness Regulatory quality Rule of law Control of corruption Control of corr. + political sta. Log Real GDP per capita -0.00515 -0.00267 -0.00110 -0.00662* -0.00376 0.000754 0.00134 (-1.517) (-0.854) (-0.249) (-1.666) (-1.026) (0.245) (0.435) 5 years average rate of Real GDP growth -0.0907 (-1.527) -0.0643 (-1.187) -0.0708 (-1.243) -0.0962 (-1.511) -0.0814 (-1.310) -0.0453 (-1.123) -0.0391 (-1.025) Oil exporter -0.00841 -0.00777 -0.00999* -0.00817 -0.00877 -0.00882* -0.00831* (-1.359) (-1.436) (-1.650) (-1.242) (-1.488) (-1.909) (-1.912) Log Population -0.00103 -0.00220 -0.000740 -0.00110 -0.00102 -0.000987 -0.00160 (-0.755) (-1.614) (-0.585) (-0.779) (-0.761) (-1.020) (-1.643) Democracy 0.00179 -0.00151 -0.000890 -0.000600 -0.000929 -0.00125 -0.00151 (0.340) (-0.376) (-0.200) (-0.121) (-0.203) (-0.384) (-0.497) Media Censorship 0.00944 0.0183 0.0219 0.0290* 0.0223 0.0128 0.0105 (0.458) (1.615) (1.617) (1.879) (1.599) (1.268) (1.131) Voice and accountability -0.00727 (-1.195) Political stability -0.00629*** -0.00353* (-2.679) (-1.794) Government Effectiveness -0.00636 (-1.634) Regulatory quality -0.00101 (-0.289) Rule of law -0.00442 (-1.406) Control of Corruption -0.00830*** -0.00602* 30 Table 6 (Continued) VARIABLES Voice and accountability Political stability Government effectiveness Regulatory quality Rule of law Control of corruption Control of corr. + political sta. (-2.643) (-1.955) Observations 2,036 2,036 2,035 2,036 2,036 2,036 2,036 Number of countries 151 151 151 151 151 151 151 Log pseudo-likelihood -130.75764 -127.54503 -130.15517 -131.3124 -130.72695 -127.47214 -125.6876 Pseudo R-squared 0.4448 0.4584 0.4473 0.4424 0.4449 0.4587 0.4663 Note: The dependent variable is successful revolution from Archigos dataset. The estimation method is panel probit random-effects. Only marginal effects are reported. In column (1) we add voice and accountability to the baseline model, political stability in column (2), government effectiveness in column (3), regulatory quality in column (4), rule of law in column (5), control of corruption in column (6), and both political stability and control of corruption in column (7). All explanatory variables are one period lagged. The model includes a constant term. Robust z-statistics in parentheses ***p < 0.01, **p < 0.05, *p < 0.1. 31 Results presented in Tables 4 and 5 did not show any impact of the regime type (measured by democracy) on the probability of successful revolutions. In contrast, the income level and the economic growth rate were statistically significant. Previous investigations in the literature tried to assess whether the effect of income level and growth on the probability of revolutions and other types of social unrest is contingent on the regime type. Collier and Rohner (2008) found that the direct effect of democracy and that of the interaction term (democracy and income level) are significant in explaining the incidence of rebellion, but with opposite signs. The direct effect is to increase the impact of rebellion. This is offset by a favorable interaction with income, creating a threshold level of income per capita at around USD 2,750, below which the net effect of democracy is to increase the incidence of rebellion. Knusten (2014) added to his study of revolutions an interactive term between the regime type (Polity) and income level (log real GDP per capita). His results suggested that higher income levels may reduce the probability of leaders being ousted through revolutions more in democracies than in dictatorships. To test whether our model is consistent with the abovementioned findings in the literature, we added, in Table 7, a multiplicative term between Democracy and Log real GDP per capita (columns 1 and 2). The effect of the income level per capita on the probability of successful revolutions only seems to matter in the case of democracies, where it has a negative effect. The marginal effects analysis (at the bottom of the table) shows that log real GDP per capita is only statistically significant in democracies. This result is in accordance with Knusten (2014). However, contrary to Collier and Rohner (2008), we found no support for the hypothesis that the effect of democracy on the probability of revolutions is contingent on the level of income per capita. Figure 1 shows the conditional marginal effects of democracy. 32 Figure 1 Conditional marginal effects of democracy over log real GDP per capita A similar analysis was conducted for the 5-year average rate of real GDP growth. Columns 3 and 4 of Table 7 show the estimation results of a regression, with an interaction term between Democracy and the 5-years average rate of real GDP growth. The analysis of the marginal effects (at the end of the table) suggests that the impact of 5-years average rate of real GDP growth on the probability of successful revolutions only matters in non-democratic countries, where it has a negative effect. As can be seen from Figure 2, there is also evidence that the presence of democratic government reduces the probability of successful revolutions only for very low levels of the 5-years average rate of real GDP growth. 33 Figure 2 Conditional marginal effects of democracy over 5-years average rate of real GDP growth 34 Table 7 Model 2: baseline model with an interaction term between democracy and log real GDP per capita, and democracy and 5-years average rate of real GDP growth Note: Dependent variable is successful revolution from Archigos dataset. The estimation method is panel probit random-effects method. We include in column (1) the interactive term between democracy and log real GDP per capita. The marginal effect is reported in column (2). Column (3) includes an interactive term between democracy and 5-years average rate of real GDP growth and the marginal effects figure in column (4). The marginal effects of the interaction term are reported at the bottom of the table. All variables are lagged one period. The model includes a constant term. Robust z-statistics in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 (1) (2) (3) (4) Interaction 1 - Democracy*log real GDP per capita Interaction 2 - Democracy * 5-years average rate of real GDP growth VARIABLES Probit coefficients Marginal effects Probit coefficients Marginal effects Log real GDP per capita -0.246 (-0.938) -0.00928*** (-2.751) -0.444** (-2.551) -0.00828** (-2.561) Average five years GDP growth -5.840* -0.102* -8.702*** -0.0478 (-1.954) (-1.766) (-2.728) (-0.893) Oil exporter -0.312 -0.00578 -0.252 -0.00470 (-0.960) (-0.954) (-0.769) (-0.786) Log population -0.0405 -0.000750 -0.0363 -0.000677 (-0.607) (-0.598) (-0.541) (-0.531) Democracy 1.482 -0.000365 -0.196 0.00349 (1.229) (-0.0518) (-0.578) (0.870) Media censorship 1.518* 0.0281** 1.607** 0.0300** (1.955) (1.971) (2.483) (2.420) Democracy * log real GDP per capita -0.405 (-1.017) Democracy* average five years GDP growth 9.763* (1.778) Observations 2,505 2,505 2,505 2,505 Number of countries 151 151 Log pseudo-likelihood -154.99871 -154.43435 Pseudo R-squared 0.3418 0.3442 Marginal effects of GDP over democracy Democracy=0 -0.00847 (-0.992) -0.263** (-2.380) Democracy=1 -0.00673*** (-2.570) 0.0129 (0.248) 35 Earlier analysis in Table 5 showed interesting results when adding Governance PC and ICT PC separately to the baseline model . Therefore, in the following Table 8, we study first the direct impact of these two variables on the probability of successful revolutions when both are added to the baseline model, and then their indirect impact through adding a multiplicative term with the regime type (columns 2 and 3). Results reported in Table 8 show a statistical significance for oil exporter, Governance PC, and ICT PC. The analysis of the marginal effects suggests that being an oil exporting country reduces the probability of successful revolutions by 0.945 percentage points. As for the governance indicator, a one standard deviation increase in Governance PC reduces the likelihood of successful revolutions by (0.207*2.25116=0.4659) 0.4659 percentage points. The result is significant at 10% level of significance. The analysis of the marginal effects also suggests that a one standard deviation increase in ICT PC decreases the probability of successful revolutions by (0.226*1.68857=0.3816) 0.3816 percentage points. The result is significant at 10% level of significance. Regarding log real GDP per capita, 5-years average rate of real GDP growth, and media censorship, they are no longer statistically significant. It should be noted that the number of observations dropped to 1,549, with only 17 cases of successful revolutions. In columns 2 and 3 of Table 8, we interact ICT PC with democracy and Governance PC with democracy, respectively. These two multiplicative terms enable us to check whether the impact on the probability of successful revolutions of more developed information and communications technology, and better governance, is contingent on the regime type. Nonetheless, the analysis of the marginal effects reported at the bottom of the table does not show any contingency of the regime type on the impact that both ICT PC and Governance PC can have on the probability of successful revolutions. 36 Table 8 Model 3: adding Governance PC and ICT PC and model 4: interacting ICT PC and Governance PC with democracy (1) (2) (3) VARIABLES Adding ICT PC and governance PC ICT PC*Democracy Governance PC*Democracy Log real GDP per capita 0.429 0.438 0.429 (1.023) (1.063) (1.026) 5-years average rate of real GDP growth -5.757 (-1.240) -5.799 (-1.224) -5.901 (-1.244) Oil exporter -1.223*** -1.247*** -1.218*** (-2.735) (-2.813) (-2.717) Log population -0.118 -0.116 -0.115 (-1.333) (-1.308) (-1.262) Democracy -0.253 -0.516 -0.317 (-0.755) (-1.089) (-0.712) Media censorship 0.480 0.420 0.435 (0.490) (0.436) (0.413) Governance PC -0.267** (-1.974) -0.260* (-1.908) -0.239 (-1.435) ICT PC -0.293** -0.191 -0.289** (-2.086) (-0.926) (-1.979) Democracy*ICT PC -0.217 (-0.892) Democracy*Governance PC -0.0571 (-0.297) Observations 1,549 1,549 1,549 Number of countries 144 144 144 Log pseudo likelihood -77.207794 -76.896316 -77.179971 Pseudo R-Squared 0.6721 0.6735 0.6723 Marginal effects ICT PC over Democracy Governance PC over Democracy Oil exporter -0.00945* (-1.800) Governance PC -0.00207* (-1.766) ICT PC -0.00226* (-1.768) Democracy =0 -0.00463 (-0.959) -0.00542 (-1.354) Democracy =1 -0.00114 (-1.162) -0.00114 (-1.421) Note: Dependent variable is successful revolution from Archigos dataset. The estimation method is panel probit random-effects method. In column (1), we add governance PC and ICT PC to the baseline model. In column (2), we add an interaction term to the baseline model between ICT PC with democracy, and in column (3), we interact governance PC with democracy. The respective marginal effects are reported at the bottom of the table. All variables are lagged one period. The model includes a constant term. Robust z-statistics in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 37 In the following Table 9, we investigate whether being an oil exporter impacts the probability of successful revolutions differently in democratic and non-democratic countries.17 Ross (2001) and De Mesquita and Smith (2010) showed that oil has a particularly stabilizing effect in relatively autocratic regimes. Knusten (2014) found similar results for attempted revolutions but not for successful ones, and Collier and Rohner (2008) found that oil reduces the probability of guerrilla warfare but not revolutions. The results of our analysis with Model 1 (Table 5) suggest a stabilizing effect of oil in general. However, to test whether the impact of oil on the probability of successful revolutions is contingent on the regime type, we split the full sample into two sub-samples –– democratic and non-democratic countries. The results of the analysis with the two subsamples are reported in Table 9. In the sub-sample of democratic countries (columns 1 and 2), the variable, oil exporter, has been omitted from the analysis, since there is no democracy with oil exporter equal to 1. As for the sub-sample of non-democratic countries, there is evidence of a pacifying effect of oil income. Being an oil exporting country reduces the probability of successful revolutions by 1.42 percentage points in nondemocratic countries. The result is statistically significant at 10% level of significance. 17 Recall that Oil exporter is a dummy variable that takes the value of 1if more than 33% of the country’s exports are derived from oil and 0 otherwise. 38 Table 9 Model 5: oil impact in democracies and non-democracies VARIABLES Democratic countries Non-democratic countries (1) Democracy =1 (2) Marginal effects (3) Democracy =0 (4) Marginal effects Log real GDP per capita 0.372 0.000510 0.179 0.00226 (0.356) (0.376) (0.351) (0.335) 5-years average rate of real GDP growth 7.938 (1.214) 0.0109 (0.699) -10.61** (-2.304) -0.134 (-1.610) Oil exporter - - -1.128** -0.0142* (-2.181) (-1.658) Log population 0.135 0.000185 -0.0780 -0.000985 (0.780) (0.483) (-0.633) (-0.667) Media censorship 0.196 0.000269 0.865 0.0109 (0.0568) (0.0563) (0.815) (0.748) Governance PC -0.442 -0.000607 -0.299** -0.00378* (-0.880) (-0.793) (-1.966) (-1.740) ICT PC -0.428* -0.000587 -0.220 -0.00278 (-1.837) (-1.029) (-1.233) (-1.150) Observations 1,139 1,139 738 738 Number of countries 110 88 Number of obs. with Dep.=1 6 11 Log pseudo-likelihood -29.110485 -47.210092 Pseudo R-squared 0.680 0.6796 Note: Dependent variable is successful revolution from Archigos dataset. The estimation method is panel probit random-effects method. We restrict the analysis in column (1) to the sample of democratic countries and to the sample of non-democratic countries in column (3). The marginal effect is reported in column (2) and (4), respectively. The model includes a constant term. All variables are one period lagged. Robust z-statistics in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 1.5.1. Globalization effect In the following Table 10, we study the effect of the three KOF globalization indices on the probability of successful revolutions (KOF global index, KOF de facto , and KOF de jure ). The analysis of the marginal effects shows that the impact of oil, ICT, and governance on the probability of successful revolutions is consistent throughout the different types of KOF indicators with the expected sign. However, the analysis of the marginal effects shows that KOF indicators do not have a statistically significant impact on the probability of successful revolutions. 45 1.7. Other forms of violent events Other specifications of events of social unrest replaced the dependent variable in the baseline model –– that is, individual violent events against the government, successful coups d’état, attempted coups d’état, the total number of riots and demonstrations (in the Middle East, North Africa, Latin America, Sub-Saharian Africa, and East/South East Asia.), revolutionary war, and civil violence). The total number of individual events of violence against the government of a country i during year t is defined as: “an incident where armed force was used by an organized actor against another organized actor, or against civilians, resulting in at least one direct death at a specific location and a specific date” (Croicu & Sundberg, 2017 p.2). We only considered cases where one party in the event is the government. Syria is excluded from the dataset. The analysis covers the period 1996 to 2015. Successful coups d’état in a country i at year t is coded 1, when effective authority is exercised by a new executive for at least one month and 0 otherwise. Data is extracted from the Centre for Systemic Peace (Marshall, 2017a). Attempted coups d’état in country i at year t are unsuccessful coups. Data is extracted from the Centre for Systemic Peace (Marshall, 2017a). The variable, total number of riots and demonstrations in a country i at year t is extracted from the Urban Social Disorder (USD) event dataset of the Peace Research Institute, Oslo (PRIO). The scope of the USD dataset is to collect events that occur in select individual cities, not to represent the situation of entire countries. It contains information about urban unrest in 89 countries in the Middle East, North Africa, Latin America, Sub-Saharian Africa, and East/South East Asia. The total number of riots includes the following: - Spontaneous violent riot : distinct, continuous, and uncoordinated action resulting from an originally non-violent protest and directed toward members of a distinct ‘other’ group or government authorities. 46 - Organized violent riot : distinct, continuous, and coordinated action staged by members of a singular political or identity group and directed toward members of a distinct ‘other’ group or government authorities. Regarding demonstrations, we considered the following: - Spontaneous demonstration: distinct, continuous, and uncoordinated, largely peaceful action directed toward members of a distinct ‘other’ group or government authorities. - Organized demonstration : distinct, continuous, and coordinated, largely peaceful action directed toward members of a distinct ‘other” group or government authorities (Bahgat et al., 2017). Revolutionary war is a binary variable taking the value 1 if the country i witnessed an outbreak of a revolutionary war at year t , and 0 otherwise. There are 20 cases of an outbreak of revolutionary war in our sample. It is calculated by the Center for Systemic Peace (see Appendix A 3 for description). Civil violence is a binary variable that takes the value of 1 if there is an episode of civil violence during the year t in country i , and 0 denotes no episodes of civil violence. It is extracted from the Centre for Systemic Peace ( Marshall, 2017a). There are 104 episodes of civil violence over the period 1996-2015 (see Appendix A 2 for description). The descriptive statistics of the different forms of political instability and social unrest appear in Table 13 below. On average, there are 0.00772 successful coups d’état, 0.0188 attempted coups d’état, 0.977 demonstrations and riots with some country/unit witnessing a maximum of 20 demonstrations and riots, 4.177 individual violent events against the government, with a maximum of 652 in some countries. As for revolutionary wars, there is an incidence of, on average, 0.00544 revolutionary war throughout the analysis. 47 Table 13 Descriptive statistics: other forms of violent events (1) (2) (3) (4) (5) VARIABLES N mean Sd Min max Successful coups d’état 3,238 0.00772 0.0875 0 1 Attempted coups d’état 3,238 0.0188 0.145 0 2 Total number of demonstrations and riots 1,555 0.977 1.877 0 20 Individual events of violence against the government 3,659 4.177 30.13 0 652 Occurrence of episodes of civil violence 3,254 0.0292 0.168 0 1 Revolutionary war 3,860 0.00544 0.0736 0 1 Sources: Urdal & Hoelscher (2012); Sundberg & Melander (2013); Marshall (2017a); Bahgat et al. (2017); Croicu & Sundberg (2017). Since the following variables –– successful coups d’état, the occurrence of episodes of civil violence and revolutionary war –– are binary variables, the panel probit random-effects method will be used as an estimation method. As for the total number of demonstrations of riots, attempted coups d’état, and individual events of violence against the government, standard linear regression is not an appropriate estimation method since the observed variable is a count number. It cannot take into account the constraint that the data and the model’s predictions can only take non-negative integer values. The Poisson regression model has been widely used to study such data and to take into consideration the preponderance of zeros and the small values and discrete nature of the dependent variable. Poisson regression and its generalization, the negative binomial regression, are usually used in this case (Greene, 2003, 2010). Collier and Hoeffler (2005) used the Poisson formulation with the dependent variable being the number of coups plotted during the year. Ponticelli and Voth (2011) used the Poisson quasi-maximum likelihood estimation method with fixed-effects to examine what leads to social instability and protests. The dependent variable used is ‘chaos’, which is defined as the sum of the number of assassinations, demonstrations, riots, general strikes, and attempted revolutions. Panel Poisson estimation methods will be applied when using attempted coups d’état, the total number of demonstrations and riots, and individual events of violence against the government. The results of the estimation of different forms of social unrest are reported in Table 14. In column 1, we estimate the baseline model with individual events of violence against the government as a dependent variable. The analysis of the results suggests that an increase in economic growth and being an oil exporter reduce the probability of such events against the government. In column (2), we also estimated attempted coups d’état with the panel Poisson regression method. Results suggest that increases in real GDP per capita decrease the probability of attempted coups. The resulting 48 estimation of the total number of demonstrations and riots in column 3 suggests that increases in 5-years average rate of real GDP growth decrease the probability of demonstrations and riots. Oil has a pacifying effect since being an oil exporter reduces the probability of demonstrations and riots. Increases in the size of the population are associated with a higher probability of demonstrations and riots. These demonstrations appear to occur more in non-democratic countries since being a democracy reduces the probability of demonstrations and riots. The remaining forms of social unrest (revolutionary war, the occurrence of episodes of civil violence, and successful coups d’état) were estimated using the panel probit method. The analysis of the marginal effects in column (4) suggests that increases in the 5-years average rate of real GDP growth reduce the probability of revolutionary war, and that increases in the size of the population increase the probability of revolutionary war. The analysis of the marginal effects in column 5 suggests that increases in the size of the population increase the probability of the occurrence of episodes of civil violence. An increase in the media censorship increases the probability of civil violence. As for successful coups d’état, the analysis of the marginal effects reported in column (6) suggests that increases in log real GDP per capita and the 5-years average rate of real GDP growth lower the probability of successful coups d’état. Finally, we limited the model estimated in Table 14 to Arab countries only. The results are reported in Table 15 for the different forms of social unrest except attempted coups d’état where there were no such attempts in our sample. Only the population size seems to explain the violent events that occurred against the government in the Arab countries, where an increase in the size of the population increases the probability of violence against the government. As for the demonstrations and riots, the analysis of the results in column 2 suggests that increases in the 5-years average rate of real GDP growth decreases the probability of demonstrations and riots. Regarding the remaining forms of social unrest, the results reported in Table 15 do not show any statistically significant effect for any of the variables considered in our model. 49 Table 14 Baseline model: other forms of violent events (1) (2) (3) (4) (5) (6) VARIABLES Violence against the Gov. Attempted Coups Demonstrations and Riots M. Effects rev. war M. Effects Civil Violence M. Effects Succ. Coups Log real GDP per capita 0.821 -1.872** 1.462 -0.00142 -0.00394 -0.00509** (0.516) (-2.519) (0.625) (-1.194) (-0.606) (-2.523) 5-years average rate of real GDP growth -8.989** (-2.375) -15.73 (-1.426) -9.472** (-2.124) -0.0435** (-2.015) -0.0311 (-0.516) -0.0494* (-1.817) Oil exporter -0.754* 0.304 -0.823** 0.000440 -0.00602 (-1.862) (0.192) (-2.233) (0.353) (-0.831) Log population 0.818 -0.146 0.419*** 0.000955** 0.00932* 0.000357 (1.315) (-0.887) (3.142) (2.334) (1.739) (0.585) Democracy 0.294 0.0350 -0.318* -0.00155 0.000786 -0.000122 (0.670) (0.0290) (-1.894) (-1.601) (0.111) (-0.0427) Media censorship -2.927 0.609 1.047 0.00266 0.0595* 0.00841 (-1.316) (0.380) (1.278) (0.891) (1.819) (1.028) Observations 2,501 2,495 1,204 2,505 2,503 2,051 Number of countries 146 151 74 151 151 131 Log pseudo-likelihood -6228.0886 -126.46987 -1740.3648 -65.178568 -151.83728 -74.732418 Pseudo R-squared 0.4441 0.5643 0.2189 0.4929 0.3242 0.4759 Note: The count dependent variables used in columns (1), (2), and (3) are respectively: Occurrence of violence events against the government, attempted coups d’état, and total number of demonstrations and riots. They are estimated by panel Poisson random-effects method. The dependent variable in column (4) is binary variable revolutionary war, in column (6) civil violence, and column (8) successful coups d’état. They are estimated by panel probit random-effects method. Only the respective marginal effects are reported. All models include a constant term. All variables are lagged one period. Robust z-statistics in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 50 Table 15 Baseline model: other forms of violent events in the Arab countries VARIABLES (1) Violence against the Gov. (2) Demonstration s and Riots (3) Rev. War (4) M. Effects rev. war (5) Civil Violence (6) M. Effects Civil Violence (7) Successful coups (8) M. Effects Succ. Coups Log real GDP per capita -11.89 6.714 -1.621*** -0.00110 0.0347 0.000249 -2.507 -0.0355 (-1.096) (0.693) (-2.985) (-0.749) (0.0210) (0.0205) (-1.642) (-0.861) 5-years average rate of real GDP growth 4.224 (1.308) -37.70*** (-2.917) -32.42*** (-5.224) -0.0221 (-0.702) -5.805** (-2.413) -0.0417 (-0.749) 30.98 (0.761) 0.439 (1.233) Oil exporter 0.169 -0.687 -0.0390 -2.65e-05 -0.718 -0.00515 (0.185) (-0.834) (-0.0633) (-0.0633) (-1.021) (-1.003) Log population 15.23*** 1.730 0.453*** 0.000308 1.550** 0.0111 0.479*** 0.00678 (2.975) (1.340) (3.776) (0.743) (2.568) (0.834) (3.027) (0.741) Democracy -0.361 -2.481 (-0.875) (-0.370) Media censorship 2.629 -2.348 -0.690 -0.000470 4.650 0.0334 -4.175* -0.0591 (0.860) (-1.287) (-0.506) (-0.416) (1.243) (1.080) (-1.835) (-0.753) Observations 269 183 252 252 250 250 82 82 Number of countries 18 12 17 17 17 17 8 8 Log pseudo-likelihood -925.48799 - 263.6642 -13.585382 -39.403737 -7.6348737 Pseudo R-squared 0.7228 0.3929 0.522 0.1678 0.6533 Note: The count dependent variables are used in columns (1) and (2) are: Occurrence of violence events against the government, and total number of demonstrations and riots. They are estimated by panel Poisson random-effects method. The dependent variable in column (3) is binary variable revolutionary war, in column (5) civil violence, and column (7) successful coups d’état. They are estimated by panel probit random-effects method with the respective marginal effects in columns (4), (6), and (8). Estimations include a constant term. All variables are one period lagged. Robust z-statistics in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 51 1.8. Conclusion Over the period 1996-2015, the world witnessed the outbreak of 47 successful revolutions (Archigos dataset). Previous studies focused on economic and political factors, mainly the regime type, to explain the reasons behind the outbreak of such an event. In this chapter, we shed light on other potential determinants of successful revolutions. Besides log real GDP per capita, where its increase lowers the probability of successful revolutions, 5-years average rate of real GDP growth was used for the first time to study successful revolutions. Other factors were considered in the baseline model: the size of the population, an indicator of oil production, media censorship, and an indicator of the regime type. Although the literature found that in more democratic countries, there is a lower probability of revolutions; however, the regime type did not exhibit any statistically significant effect on the likelihood of successful revolutions. This result was consistent when using the different available measures of the regime type (DD, Polity 2, and Polity). However, when interacted with democracy, we found that log real GDP per capita reduces the probability of successful revolutions only in democratic countries, and the effect of 5-years average rate of real GDP growth on the probability of successful revolutions matters only in non-democratic countries. Another statistically significant variable considered in our baseline model is media censorship. This variable was never used previously to study neither the onset nor the success of revolutions. Our results suggest that an increase in media censorship increases the probability of successful revolutions. Stronger censorship in a country witnessing social unrest might result in higher levels of violence ultimately increasing the probability that such violence turns out to be a successful revolution. This conclusion is in line with the findings of Casilli & Tubaro (2012, p. 14) on social media censorship. The authors argue that “in the absence of robust indicators as to the rebelliousness of a given society, the choice of not restricting social communication turns out to be a judicious one for avoiding to trade democratic values and freedom of expression for an illusory sense of security”. In the absence of censorship, agents express freely and peacefully their discontent about their governments’ policies. The stronger the censorship, the more violent the protests are, and the higher is the probability of a successful revolution. Other variables were added to the baseline model: Only Governance PC and ICT PC reported statistically significant results, where an increase in either of these two variables lowers the probability of successful 52 revolutions. Similar results were obtained when we replaced Governance PC with political stability and control of corruption. It is only recently, with the outbreak of the Arab spring, that research has started to focus on the importance of governance and the development of media and communication technologies in explaining the onset of revolutions. However, our results showed that these two factors played a role in explaining the success of revolutions occurring many years before the Arab spring. Good governance assures macroeconomic stability and security of person and property rights (Springborg, 2011). Inadequate governance inhibits industrialization and threatens to the country’s economic well-being which raises people’s discontent leading them to resort to political violence. As for ICT, they facilitate the spread of information and the freedom of expression. Their effect on the probability of successful revolutions corroborates with that of media censorship. With more advanced ICT, people do not need to resort to violence in order to frame their grievances. We also noted that when adding these two variables to the baseline model, oil starts to exhibit statistically significant results, particularly in non-democratic countries. Such a pacifying effect was absent in previous studies of successful revolutions. Our baseline model was robust in explaining the successful revolutions that occurred in our sample over the period 1996-2015; this, however, was not the case when we restricted our analysis to the Arab countries. In this paper, we reevaluated the triggers previously advanced in explaining the causation of successful revolutions –– mainly income level, the size of the population, oil, and the conditional effect that democracy can have on the probability of successful revolutions. Our main contribution to knowledge in this domain consists of introducing and exploring new factors such as media censorship, governance, and ICT into the analysis. These factors proved to have a statistically significant effect in most of the regressions that we conducted. The main limitation of this study is related to the time frame considered for our analysis, which reduced the number of successful revolutions. However, considering a more extended period would not have allowed us to study many of the new factors included in this paper. Another limitation was the missing data for several variables, which reduced the number of observations. 53 CHAPTER 2: VOTING IN A MUTLI-CONFESSIONAL MULTI-PARTY SYSTEM: THE POWER OF ECONOMICS Abstract The purpose of this paper is to estimate the effects of religious, socio-economic, and political factors on party choice in Lebanon. In this multi-confessional, diverse, and pluralistic country, it is not a surprise to find statistically significant results of the religious factors. Nevertheless, the economy is still a significant predictor of political parties’ vote shares. Districts with high levels of income vote more for Hezbollah and Future Movement, and less for Independents. An increase in the unemployment rate penalizes the Syrian Social Nationalist Party. Independents benefit when the number of independent ministers is large. Moreover, as expected, bombings and assassinations have a significant impact. Independent candidates are penalized in districts that witness more bombings and assassinations during election years. Keywords: Economic voting; Income; Religion; Unemployment; Time in office. 2.1. Introduction The literature on economic voting behavior is extensive and consistently finds that economic factors, particularly unemployment and inflation, impact on electoral results, and voting behavior.19 The relevance of party identification in voters’ behavior is also well established in the literature (Campbell et al., 1960). Moreover, it has been demonstrated that the social profiles of political candidates, namely their religious affiliation, significantly affect levels of partisan voting (Campbell, Green, & Layman, 2011). However, most of the existing literature focuses on Western democracies, where consolidated democracies with secular states prevail. Analyses of Arabic countries, where religion polarizes society, and violent events are frequent, are very scarce. This paper intends to contribute to filling this gap in the literature by analyzing the factors that shape the voting behavior of Lebanese citizens, based on the parliamentary elections of 1996, 2000, 2005, and 2009. Using data for the 26 administrative districts, we investigate which variables influence parties’ 19 See Lewis-Beck & Stegmaier (2019) for a recent survey on economic voting. 54 vote shares at the district level. Our variables range from political to economic and social factors, including religion. With more than 18 religious sects and more than 30 political parties, we raise the question as to whether religion is the main determinant of Lebanese voters’ choice. Additionally, the impact of economic performance on incumbents’ electoral results is also a question worthy of investigation. To the best of our knowledge, these questions remain unanswered. In this paper, we present a quantitative analysis, based on econometric estimations, of the factors underlying Lebanese voters’ choice in parliamentary elections. The paper is organized as follows: Section 2.2 reviews the literature on voting and popularity functions. Section 2.3 briefly presents the political and electoral context of Lebanon. Section 2.4 explains the empirical model and describes the data. Section 2.5 displays and discusses the empirical results. Finally, Section 2.6 concludes our investigation. 2.2. Literature review The responsibility hypothesis of Downs (1957) is the theoretical starting point of the literature on economic voting. The basic idea is that voters hold the government accountable for the evolution of the economy: They support the government when the economy is doing well and, otherwise, vote for the opposition. Many empirical studies found supportive evidence for this hypothesis (Lewis-Beck & Stegmaier, 2013, 2019). Most of the early national-level studies analyzed the relationship between macroeconomic variables and election outcomes in the United Kingdom (Goodhart & Bhansali, 1970; Frey & Schneider, 1978b; Pissarides, 1980) and the United States ( Kramer, 1971; Goodman & Kramer, 1975; Tufte, 1975, 1978; Hibbs, 1982). Results confirmed that voters’ behavior responds to the macroeconomic conditions, particularly to inflation and unemployment.20 Similar results were obtained by studies focusing on other countries.21 20 For more recent studies about the UK and the USA see, among others Kiewiet & Udell (1998), Lewis-Beck & Rice (1992), Hodgson & Maloney (2012) and LewisBeck et al., (2013). 21 Among others, see for Australia Schneider & Pommerehne (1980); for Denmark Madsen (1980), Paldam & Schneider (1980); for France Lafay (1985), Rosa & Amson (1976) and Rosa (1980); for Japan Inoguchi (1980); for New Zealand Ursprung (1983); for Norway Madsen (1980); for Portugal Veiga (1998) and Veiga & Veiga (2004), and for Sweden Jonung & Wadensjoe (1979). 61 Table 17 Geographical and confessional distribution of parliamentary seats Administrative District Seat/Confession Beirut 1 1 Orthodox 1 Catholic 1 Maronite 1 Armenian Catholic 1 Armenian Orthodox Beirut 2 1 Sunni 1 Shiite 2 Armenian Orthodox Beirut 3 5 Sunni 1 Shiite 1 Orthodox 1 Druze 1 Evangelist 1 Minorities Keserwan 5 Maronite Byblos 2 Maronite 1 Shiite Matn 2 Orthodox 4 Maronite 1 Armenian Orthodox 1 Catholic Baabda 3 Maronite 1 Shiite 1 Druze Aley 2 Druze 2 Maronite 1 Orthodox Chouf 2 Druze 3 Maronite 2 Sunni 1 Catholic Sidon 2 Sunni Sidon villages 1 Catholic 2 Shiite Tyre 4 Shiite Bint Jbeil 3 Shiite Jezzine 2 Maronite 1 Catholic Marjeyoun 2 Shiite 1 Orthodox 1 Sunni 1 Druze Nabatiyeh 3 Shiite Akkar 1 Alawite 1 Orthodox 3 Sunni 1 Maronite MinniehDonnieh 3 Sunni Bcharreh 2 Maronite Tripoli 1 Maronite 5 Sunni 1 Alawite 1 Orthodox Zgharta 3 Maronite Koura 3 Orthodox Batroun 2 Maronite Baablbak-Hermel 6 Shiite 2 Sunni 1 Catholic 1 Maronite Zahle 2 Catholic 1 Orthodox 1 Armenian Orthodox 1 Maronite 1 Shiite 1 Sunni West Bekaa-Rachaya 1 Orthodox 1 Maronite 2 Sunni 1 Druze 1 Shiite Source: Lebanese Parliament (2000) Regardless of their confessional affiliation, voters vote for all seats in the district in a single ballot paper. They are entitled to choose as many candidates as the number of seats assigned to the constituency (Harik, 1980). They have the right to remove or add candidates from pre-arranged electoral lists, as long as the sectarian balance of the list remains the same (Salloukh et al., 2015). Eligible voters should be Lebanese, above 21 years old, and having all their civil and political rights (1960 Electoral Law, 1960). They vote in their ancestral village rather than their residential district. 62 The quota system forces candidates to compete against contenders from their own sect because the number of seats to which a sect is entitled in a constituency is defined by law (Harik, 1980). Within each sect, the candidate receiving the highest number of votes ultimately represents that confession in Parliament. This makes the contest intra-, not inter-sectarian, and reduces the chances of sectarian clashes during the elections (Harik, 1980). Salloukh et al . (2015) consider that the primary determinant of electoral results is the sectarian demographic composition of the electoral districts. The electoral districting used until the 2009 elections to configure confessions is based on five large governorates ( Muhafazat ) that are subdivided into smaller constituencies ( qadas ) — see Table 18 (Ekmekji, 2012). 2.3.2.2. Elections The first post-war parliamentary election was held in 1992. The number of seats increased from 108 to 128 as per Law 154 of 1992. However, the increase in the number of seats, despite maintaining the Parliament divided equally between Christians and Muslims, was not allocated in such a way as to represent the actual increase in the number of voters belonging to each religion. For instance, the new Christian seats were allocated in regions where the balance of votes favored Muslims rather than Christian voters (Salloukh et al., 2015). Adding to this, the proportion of deputies per number of voters varied tremendously among electoral districts. Such malapportionment distorted the electoral results and favored economically developed districts at the expense of less developed ones (Salloukh et al., 2015). Gerrymandering was also used in order to channel votes for the favor of specific political/sectarian where the size of electoral districts was modified several times between 1992 and 2009 — see Table 18. Turnover of eligible voters in some Christian areas was low in the 1992 parliamentary elections because some voters objected to voting in the presence of Syrian troops (Youngblood-Coleman, 2016). 63 Table 18 Structure of constituencies (electoral districts)22 in Lebanon by election year Governorate Qadas /Administrative Districts 1996 2000 2005 2009 BEIRUT 1 Beirut I (Areas) 1 Beirut I Beirut I Beirut I 2 Beirut II (Areas) Beirut II Beirut II Beirut II 3 Beirut III (Areas) Beirut III Beirut III Beirut III MOUNT LEBANON 4 Keserwan 1 Mount I Mount I 1 5 Byblos 1 1 6 Almatn 1 Mount II Mount II 1 7 Baabda 1 Mount III Mount III 1 8 Alay 1 1 9 Chouf 1 Mount IV Mount IV 1 SOUTH & NABATIEH 10 Sidon 1 1 South I 1 11 Sidon Villages 1 12 Tyre 1 13 Bint Jbeil 1 14 Jezzine South II 1 15 Marjeyoun-Hasbaya 1 16 Nabatieh 1 NORTH 17 Akkar 1 North I North I 1 18 Bcharreh 1 19 Donnieh (Area) 1 Minieh (Area) North II North II 20 Tripoli 1 21 Zgharta 1 22 Koura 1 23 Batroun 1 BEKAA 24 Baalbek-Hermel 1 Bekaa I Bekaa I 1 25 Zahleh Bekaa II Bekaa II 1 26 West Bekaa-Rachaya Bekaa III Bekaa III 1 Total Constituencies 10 13 14 26 Election Laws Law 587 13/08/1 996 Law 171 06/01/2 000 Law 171 06/01/200 0 Law 25 08/10/200 8 RECTIFIED Law 59 27/12/200 8 Source: Lebanese Parliament (2000) 22 2000 & 2005: Beirut I: Mazraa, Achrafieh and Sayfeh areas; Beirut II: Msaytbeh, Bachoura and Rmeil areas; Beirut III: Ras Beirut, Ain Mreiseh, Marfaa, Meena Hosn, Zkak Blat and Medawar areas. 2009: Beirut I: Achrafieh, Sayfeh and Rmeil areas; Beirut II: Medawar, Marfaa and Bachoura; Beirut III: Mazraa, Msaytbeh, Ras Beirut, Ain Mreyseh, Meena Hosn and Zkak Blat areas. 64 Parliamentary elections that took place in 1996 were not much different from those of 1992, where many Christians continued to call for the boycotting of the elections. Sectarian consciousness remained high (Hudson, 1999). These elections took place according to electoral law number 587, which was based on amendments made to the 1960 electoral law and stipulated ten electoral districts (see Table 18). This meant that elections in Mount Lebanon were arranged separately for each local district ( qada ) and not at the level of the governorate Mohafaza, as in the remaining districts. Such separation of electoral districts serves sectarian, political, or personal interests (El Khazen, 1998)— see Appendix B 1 for candidates and winners in 1996 elections. El Khazen (1998) considered that this law violated the constitution with regards to the following points: the distribution of electoral districts was not the same in all districts; disparities existed at the level of seats and voters between one electoral district and another. Additionally, three different electoral patterns existed. In Beirut, the nomination of candidates and the voting followed the level of the Muhafaza as an electoral district. In the North, Bekaa and the South, the nomination of candidates was made at the level of qada , but the voting was at the Muhafaza level. Finally, in Mount Lebanon, the nomination of candidates and the voting followed the level of qada . The distribution of seats among the different confessions was the same as that in the 1992 elections. Parliamentary elections that took place in 2000 and 2005 were based on a new electoral law (number 171), which included amendments to the 1996 law. These amendments were mainly related to the size of electoral districts rather than the electoral law (El Khazen, 2002). Seats were allocated among 13 electoral districts (instead of 10 as per the 1996 elections): the eight qadas in the North formed two districts; the eight qadas in the south formed two districts; the six qadas in Mount Lebanon formed four districts (Law Number 171: Modification of Parliament, Municipal and Mukhtars Electoral Law (Arabic), 2000). These amendments were made in order to preserve the position of certain political parties and to marginalize others, especially in the parliamentary elections in 2000 (El Khazen, 2002)— see Appendix B 2 and B 3 for candidates and winners in 2000 and 2005 elections. Despite the fact that the 2005 parliamentary election was conducted under the same electoral law as that of 2000, the context that preceded these elections was totally different. Syrian troops had withdrawn from the 65 Lebanese territory, there was growing international pressure to disarm Hezbollah, and, in addition to the succeeding staggering events, PM Hariri was assassinated (Rizk, 2007). According to Corm (2012), between 1992 and 2005, Lebanon was under Syrian-Saudi dominance, where the constitutional practice reinforced the communitarianism. Most of the constitutional power was in the hands of the PM and not the cabinet members, as stipulated by the constitution. Islamic radicalism and the radicalism of the western states placed relations between Sunni and Shiite center stage in the problems facing Lebanon. As for the Christians, an important modification occurred that created a sort of democratic consociationalism – the alliance between the Free Patriotic Movement (FPM) and Hezbollah. Fall 2006 witnessed mass protests from pro-Syrian and Hezbollah groups demanding the resignation of PM Siniora. This group was known as ‘the forces of March 8’ in opposition to ‘the forces of March 14’. The Christians of March 14 were allied with the Sunni Future Movement, which was in symbiosis with the KSA and eventually with the USA. Two years after the assassination of PM Rafik Hariri, Lebanon was deeply divided along sectarian lines between Shiite Muslims, Sunni Muslims, and Maronite Christians. Lebanon was also politically fragmented between pro-Syrian and pro-Hezbollah factions on one end of the equation and pro-western enclaves on the other end (Youngblood-Coleman, 2016). Rival political leaders met on May 18, 2008, in Doha, Qatar, and reached the Doha agreement, which stipulated the formation of a new government and the election of a new President of the Republic. Effectively, a couple of days later, Michel Suleiman was elected President, and Fouad Siniora appointed as PM. In June 2009, parliamentary elections were held in Lebanon as per the 1960 electoral law. It consisted of 26 electoral districts, which are the same 26 administrative districts qadas — see Appendix B 4 for candidates and winners in 2009 elections. Saad Hariri was nominated as PM; however, his government collapsed in 2011 after the resignation of 10 ministers of Hezbollah and his allies who were not able to convene the government to oppose the UN special tribunal that led the investigation into the assassination of the late PM Rafik Hariri. Four months later, Najib Mikati, the new appointed PM, formed a new government in which the FM and the March 14 alliance members refused to take part (Youngblood-Coleman, 2018). 66 Meanwhile, the repercussions of the Syrian conflict started to slip over into the Lebanese scene. Several violent conflicts occurred in the capital city, Beirut, and in the northern parts of the country between proSyrian activists and Sunnis backing the Syrian opposition. The internal intelligence chief, Wissam Al Hassan, who supported the March 14 alliance, was assassinated in a car bombing in October 2012. A few months later, PM Mikati resigned, and Tammam Salam was appointed to form a new government. The parliamentary elections that were supposed to take place during June 2013 were postponed. During subsequent years, Lebanon witnessed several assassinations targeting pro-Hezbollah as well as proMarch 14 alliance. The Lebanese army and security forces, as well as the southern suburb of Beirut largely inhabited by pro-Hezbollah Shiite, were the targets of many suicide and car bombings. Tensions accelerated, particularly in border areas between Lebanon and Syria, where Syrian terrorist militants became active. In late 2014, the term of the President of the Republic, Michel Suleiman, came to an end. Two years passed without reaching a consensus among the different Lebanese factions on who would be the next president of the republic. Finally, on 31 October 2016, the former commander of the Lebanese army, General Michel Aoun, was elected President of the Republic, and PM Saad Hariri was appointed to form a new government. On the security level, the Lebanese army forces and intelligence security were able to ensure stability across all Lebanese territory and to eliminate any remaining terrorists in the border areas of Lebanon. In June 2017, the government passed a new electoral law that would govern the elections scheduled on May 6, 2018 (Daily Star, 2018). This electoral law is based on proportional representation. It divided the country into 15 electoral constituencies (comprising one or more electoral districts). The elections effectively took place on May 6, 2018, with a turnout of 49.7%. 2.4. Empirical model Since the end of the Lebanese civil war, more than 20 political parties obtained seats in the Lebanese parliament. However, most of them were small and concentrated in specific constituencies.23 The six major political parties are: Amal movement (AMAL), Hezbollah (HEZ), Future Movement (FM),24 Progressive Socialist 23 See Appendices B 6 to B 9 for election results. 24 FM was officially established in 2007 after the assassination of PM Hariri 2005. Before 2005, it existed under the name of the Hariri coalition. 67 Party (PSP), Lebanese Communist Party (LCP), Syrian Social Nationalist Party (SSNP), and Independents of March 14 coalition (M14), which appeared after the assassination of PM Hariri in 2005. In our analysis, we will consider only the six major parties and independent candidates25 (IND), that is, candidates that are not affiliated to any political party. To explain party choice, our model includes economic, political, and social variables. Our data set covers the 26 administrative districts of Lebanon over the parliamentary elections that took place in 1996, 2000, 2005, and 2009.26 The baseline model can be specified as follows: 𝑉𝑑,𝑡 𝑖= 𝑖+ 𝑖𝐿𝑑,𝑡 + 𝑖𝐶𝐸𝐶𝑑.𝑡 + 𝑖𝐸𝐿𝑑,𝑡 𝑖+ 𝑑,𝑡, 󰇛𝟑󰇜 for i =AMAL, HEZ, FM, PSP, LCP, SSNP, M14 and IND; t = 1996, 2000, 2005, and 2009; and d =1, …, 26 Where i represents the political party, t represents the election year and d the district. The dependent variable is the vote share of party i , in district d , in the election held in year t . 𝑑,𝑡 represents the error term,  is a parameter to be estimated and ,  and  are vectors of parameters to be estimated. The right-hand side of the equation includes vectors of explanatory variables that are associated with political, socio-economic and religious variables. 25 Christian and other political parties obtained seats in parliament; however, since they run in a limited number of districts/year, the number of observations was insufficient to run regressions and to include them in our analysis. 26 The 1996 parliamentary elections took place on : 18 August 1996 in Mount Lebanon Governorate (Keserwan, Byblos, Matn, Baabda, Alay, and Chouf); 25 August 1996 in the North Governorate (Akkar, Bcharreh, Donnieh, Minieh, Tripoli, Zgharta, Koura, and Batroun); 1 September 1996 in Beirut Governorate (Beirut I, Beirut II, and Beirut III); 8 September 1996 in South Governorate (Sidon, Sidon-Villages, Tyre, Bint-Jbeil, Jezzine , MarjeyounHasbaya, and Nabatiyeh); and 15 September 1996 in the Bekaa Governorate ((Baalbak-Hrermel, Zahle, and WestBekaaRachaya). The parliamentary elections of 2000 took place on: 27August 2000 in Mount Lebanon Governorate (Keserwan, Byblos, Matn, Baabda, Alay, and Chouf), and the North Governorate (Akkar, Bcharreh, Donnieh, Minieh, Tripoli, Zgharta, Koura, and Batroun); 3 September 2000 in Beirut Governorate (Beirut I, Beirut II, and Beirut III), the South (Sidon, Sidon-Villages, Tyre, Bint-Jbeil, Jezzine , MarjeyounHasbaya, and Nabatiyeh) and Bekaa Governorates (Baalbak-Hrermel, Zahle, and WestBekaaRachaya). The 2005 parliamentary elections took place on : 29 May 2005 in Beirut Governorate (Beirut I, Beirut II, and Beirut III); 12 June 2005 in Mount Lebanon Governorate (Keserwan, Byblos, Matn, Baabda, Alay, and Chouf); 5 June 2005 in the South and Nabatiyeh Governorates(Sidon, Sidon-Villages, Tyre, Bint-Jbeil, Jezzine , MarjeyounHasbaya, and Nabatiyeh); 19 June 2005 in the North Governorates (Akkar, Bcharreh, Donnieh, Minieh, Tripoli, Zgharta, Koura, and Batroun); and 12 June 2005 in the Bekaa Governorate (Baalbak-Hrermel, Zahle, and WestBekaaRachaya). The parliamentary elections of 2009 took place on 7 June 2009. 68 The vector of political variables includes the average number of ministers that a party has had in office since last election, and the total number of political assassinations and bombings that occurred in district d in the election year. The socio-economic performance of each electoral district is captured by the unemployment rate, real income per capita , and the percentage of households that have access to water in the election year. The vector of religious variables includes the share of voters belonging to the several religious groups . For consistency of the results, the same method was used to select the relevant variables for all party equations. We started by putting only one variable from each group. The variables initially selected were: the number of ministers, among the political variables; income per capita as a socio-economic variable; and the sect to which the political party is associated as a religious variable (i.e. Shiite for Amal and Hez, Druze for PSP, Sunni for FM). Other variables from each group were then added to the model, one at a time. Only the statistically significant variables were left in the model. We extract data for political variables from public administrations, international organizations, and local newspapers. For the 2009 and 2005 election results, the references used are those of the Ministry of Interior and Municipalities in Lebanon, the International Foundation for Electoral Systems, and Information International. Regarding the two previous elections (2000 and 1996), the data is from the archives of local newspapers. Data for the total number of assassinations and bombings was obtained from local newspapers (Al Shark, 2005 and 2006; L’Orient Le Jour, 2007; Annahar, 2008). Data on political parties in office comes from the Lebanese presidency of the Council of Ministers — see Appendix B 10 for political parties in office during election years. Regarding the unemployment rate, access to water, and income, the data is from the Central Administration of Statistics (CAS), country reports of the United Nations Development Program (UNDP), and the World Development Indicators (WDI). All variables have an annual frequency. We report the descriptive statistics for the variables used in the empirical model in Table 19. The party with the lowest share of votes is the LCP (3.58%), whereas political parties who were able to capture the highest share of votes are Independents with an average of 40%, followed by FM (37.7%), and AMAL (27.6%). In some districts, IND obtained a maximum of 100% of the votes. As for the average number of years for independents in office since the last election is 7.9 years. Regarding the variable total number of assassinations and 69 bombings in each district during the election year; results show that, on average, there have been 0.135 assassinations per district. A maximum of 5 bombings and assassinations occurred, whereas in other districts, there have been no assassinations or bombings. Regarding socio-economic variables, the unemployment rate is 8.16% on average in the election year. In some districts, the unemployment reached a maximum of 14.8%, but it did not go below 4.3%. The real per capita income is 4430$ on average. 78.1% of households have access to water on average in all the administrative districts. In some districts, access to water reached 98%, while in others, it did not exceed 30% of total households. When we take into consideration the religious sect of voters, we notice, on average, the following: 23.6% of voters belong to the Muslim Shiite sect; 23.3% are Muslim Sunni; 4.95% are Muslim Druze; 0.481% are Muslim Alawite; 28.3% are Christian Maronite; 5% are Greek Catholic; 8.71% are Greek Orthodox; 0.369% are Protestant Evangelical; 0.6% are Armenian Catholic; 2.61% are Armenian Orthodox; 1.49% are Christian minorities; and 0.5% belong to other religions like Jews. 70 Table 19 Descriptive statistics (1) (2) (3) (4) (5) VARIABLES N mean Sd min Max % votes of Amal Movement 30 0.276 0.197 0.0242 0.648 % votes of Hezbollah 29 0.223 0.130 0.0219 0.489 % votes of Future Movement 29 0.377 0.221 0.0754 0.849 % votes of Progressive Socialist Party 16 0.120 0.0423 0.0711 0.196 % votes of Lebanese Communist Party 29 0.0358 0.0307 3.09e-05 0.114 % votes of Syrian Social Nationalist Party 29 0.0813 0.0612 0.00746 0.205 % votes of Independents of 14 March coalition 31 0.232 0.106 0.0344 0.442 % votes of Independents 100 0.400 0.300 0.00143 1 The average number of ministers that Independents had in office since the last election 104 7.917 2.087 4.500 9.666 Total number of assassinations and bombings 104 0.135 0.654 0 5 Unemployment rate 104 0.0816 0.0203 0.043 0.148 Real income per capita (in thousands) 104 4.430 1.793 1.728 8.799 Percentage of households that have access to water 104 0.781 0.150 0.300 0.980 % of voters belonging to the Muslim Shiite sect 104 0.236 0.306 0 0.952 % of voters belonging to the Muslim Sunni sect 104 0.233 0.271 0.000409 0.851 % of voters belonging to the Muslim Druze sect 104 0.0495 0.117 0 0.528 % of voters belonging to the Muslim Alawite sect 104 0.00481 0.0156 0 0.0776 % of voters belonging to the Christian Maronite sect 104 0.283 0.296 0.0143 0.991 % of voters belonging to the Christian Greek Catholic sect 104 0.0509 0.0485 0.000586 0.201 % of voters belonging to the Christian Greek Orthodox sect 104 0.0871 0.122 3.24e-05 0.646 % of voters belonging to the Christian Protestant Evangelical sect 104 0.00369 0.00317 0 0.0129 % of voters belonging to the Christian Armenian Catholic sect 104 0.00620 0.0120 0 0.0592 % of voters belonging to the Christian Armenian Orthodox sect 104 0.0261 0.0545 0 0.268 % of voters belonging to the Christian Minorities 104 0.0149 0.0223 8.10e-06 0.107 % of voters belonging to Other religions 104 0.00510 0.00784 8.16e-06 0.049 Sources: Addiyar (1996); Alhayat (1996); Annahar (1996, 1996b, 1997, 2008); Al Shark (1996, 2005); CAS and UNICEF (2010); CAS (1997, 1996, 1998a, 1998b, 2000, 2005) (Central Administration of Statistics et al. , 2008) MOSA and UNDP (2007); UNDP (2008); Chambers (2009); LADE et al. (2010); MOIM (1996, 2005a, 2005b, 2005c, 2005d, 2005e, 2006, 2009, 2010); Nohlen, Grotz, and Hartmann (2003); El Khazen (1998); NDI (2009); EU (2009); IFES (2011, 2009b); L’Orient Le Jour (2000); Atallah (2000); Lebanonwire, n.d.; Information International (2009a, 2009b, 2009c, 2009d, 2009e); Nationl Information Agency (1996); World Bank (2017) and Center for Educational Research and Development (2017). 77 parliamentary elections that took place since 1992 but did not win a single seat. In 2005, a former secretary general of LCP, George Hawi, a recent critic of Syria, was assassinated in a car bombing. Allies of Hawi accused Syria of the assassination. Table 24 Election model for the Lebanese Communist Party LEBANESE COMMUNIST PARTY Pooled OLS Coefficients Std. Err. Druze -0.123*** 0.036 Alawite -0.635*** 0.144 Protestant Evangelical 3.64* 2.129 Constant 0.034*** 0.087 R-squared 0.29 Number of Observations 29 Note: The dependent variable is the vote share of LCP in district d in election year t. The estimation method is POLS method. Robust standard errors are reported in the last column. ***p < 0.01, **p < 0.05, *p < 0.1 In the case of the Lebanese Communist Party, we can see that the religious composition of the electorate is not of the utmost importance, although it seems that the presence of more Druze Alawite is associated with a lower vote share. Again, our tests do not show evidence that the small sample is a big issue. The condition index is 1.63 and the largest VIF is 1.26. We found one candidate to be an influential observation, but removing does not change the results. In this specific case, because the share of votes of the communist party is close to zero in several districts, we also estimated a fractional probit. The estimated marginal effects of each variable are almost the same as the coefficients of Table 24. 2.5.6. Syrian Social Nationalist Party The Syrian Social Nationalist Party was founded by Antoun Saadeh in 1932. It aims at: creating a Syrian Social Nationalist renaissance which guarantees the achievement of the party’s principles and brings back the Syrian nation (Lebanon, Syria, Palestine, Jordan, Iraq, Kuwait, and Cyprus) to strength and vitality; organizing a movement that leads to the complete independence of the Syrian nation and the vindication of its sovereignty; establishing a new order that ensures the nation’s interest and raises its standards of living, and working towards creating an Arab front (Saadeh, 1948). The SSNP took part of the Lebanese civil war as 78 a member of the Lebanese national movement. SSNP is a secular party whose members are present in different geographical areas of the country. Table 25 Election model for the Syrian Social Nationalist Party SYRIAN SOCIAL NATIONALIST PARTY Pooled OLS Coefficients Std. Err. Sunni 0.520*** 0.144 Shiite 0.637*** 0.141 Druze 0.670*** 0.131 Maronite 0.505*** 0.140 Greek Orthodox 0.769*** 0.129 Protestant Evangelical 5.63*** 1.925 Armenian Catholic 3.34*** 1.032 Unemployment -0.355* 0.174 Constant -0.474*** 0.137 R-squared 0.83 Number of Observations 29 Note: The dependent variable is the vote share of SSNP in district d in election year t. The estimation method is POLS method. Robust standard errors are reported in the last column. ***p < 0.01, **p < 0.05, *p < 0.1 In Table 25, we can see that, as with other parties, the vote share of SSNP depends mostly on the religious composition of the district. However, their vote share is not restricted to a particular religious sect which is expected as they are considered a secular party. Still, as most of the times before, economics also plays a role. A higher unemployment rate is associated with a lower vote share. This is the regression in which the small sample problems are stronger. The largest VIF is 74 and the condition index is 21. Removing the variable Shiite from the regression solves this problem, but the fit of the regression drops significantly. 2.5.7. Independents of March 14 coalition This coalition was the result of the Cedar Revolution – the popular demonstrations that took place in Lebanon following the assassination of former Prime Minister, Rafik Hariri on 14 February 2005. March 14 coalition31 was mainly formed of political parties and independents who were opposed to Syria and its allies in Lebanon 31 Future Movement, Progressive Socialist Party, Lebanese Forces Party, Lebanese Phalange Party, Social Democrat Henchokian Party, Armenian Democratic Liberal Party, Ramgavar Party, Democratic Left Movement, Independence Movement, Renewal Democratic Movement, Change Movement and Liberal Nationalist Party. 79 known as March 8 coalition. Political parties and Independents of March 14 coalition are mainly associated with Christian religion, Sunni and Druze sects. The goals of March 14 coalition were mainly to establish an international investigation commission to reveal the perpetrators of the assassination of the former PM Hariri and to ask the parliament for discussing the series of assassinations that followed and revealing the truth. They called for the departure of the Syrian troops from Lebanon and for the support of the Arab countries and the international community (14 March Organization, n.d.). Table 26 Election model for the Independents of March 14 coalition MARCH 14 COALITION Pooled OLS Coefficients Std. Err. Druze 0.386*** 0.092 Greek Catholic -1.196*** 0.262 Protestant Evangelical -29.567*** 4.357 Christian Minorities 2.780*** 0.682 Constant 0.338*** 0.030 R-squared 0.55 Number of Observations 31 Note: The dependent variable is the vote share of M14 in district d in election year t. The estimation method is POLS method. Robust standard errors are reported in the last column. ***p < 0.01, **p < 0.05, *p < 0.1 In Table 26, we can see that only religious variables seem to matter. Given that the Independents of March 14 coalition are not associated with specific religious sects, it is not surprising to see that the associated coefficient to Druze and Christian minorities is positive and significant. 2.5.8. Independents This is the most interesting case for several reasons. First, we have 100 observations. Second, we found a richer model, which includes variables such as assassinations and the average number of ministers in the government. 80 Table 27 Election model for Independents INDEPENDENTS Pooled OLS Coefficients Std. Err. Shiite -0.438*** 0.059 Greek Catholic 1.112*** 0.378 Assassinations and Bombings -0.052*** 0.017 Income per capita (in thousands) -0.082*** 0.011 Average numbers of ministers 0.051*** 0.008 Constant 0.415*** 0.099 R-squared 0.60 Number of Observations 100 Note: The dependent variable is the vote share of IND in district d in election year t. The estimation method is POLS method. Robust standard errors are reported in the last column. ***p < 0.01, **p < 0.05, *p < 0.1 As for parties, religious variables also play a role for independent candidates. Districts with a larger share of Greek Catholic tend to vote more for independents, while they find less support among the Shiites. More violence hurts Independents, probably because violence increases polarization around established parties. Districts with high income per capita vote less for Independents. The latter benefit more when the number of independent ministers is large. 2.6. Conclusion This chapter aimed at examining the effects of religious, socio-economic, and political variables on party choice in Lebanon. The POLS method was used to describe the link between the party choice on the one hand and three groups of independent variables on the other hand. Our empirical investigation faced limitations in the data used. However, despite these limitations, this paper is the first, to our knowledge, to empirically test the effect of religious, socio-economic, and political variables on party choice in Lebanon, over the four parliamentary elections under study. Our analysis was applied to six political parties and independent candidates in Lebanon. For each political party, only the most relevant explanatory variables were considered to avoid losing degrees of freedom. A series of tests were performed to make sure that the small number of observations was not a big issue. As one would expect, in a multi-confessional and diverse society such as Lebanon, the religious factors are strong predictors of parties’ vote shares. However, results also show that the state of the economy influences electoral results. 81 Religious variables, namely the percentage of voters belonging to specific religious sects, are particularly relevant for explaining the vote shares of the Amal Movement, the Hezbollah, the Progressive Socialist Party, and the Independents of the March 14 coalition. Being associated with the Shiite, it was not surprising to find that the vote share of the Amal Movement is positively and mostly associated with the percentage of voters belonging to the Muslim Shiite sect. A similar result was obtained for the vote shares of Hezbollah. A larger share of voters belonging to the Muslim Druze sect positively affects the vote shares of the Progressive Socialist Party and the Independents of March 14 coalition. The latter also benefits from a higher percentage of voters from Christian minorities. Furthermore, results suggest that the Lebanese Communist Party obtains more political support among Protestant Evangelical, and that Greek Catholics tend to support Independent candidates. Finally, religious variables also explained the vote share of the Syrian Socialist Nationalist Party, but are not restricted to a particular sect given the fact that it is a secular party with partisans having different religious affiliations. Despite the importance of religious factors, socio-economic variables also explain Lebanese parties’ vote shares. There is evidence that the state of the economy, namely income per capita, positively influences the vote share of the Hezbollah and the Future Movement and negatively that of Independents. The economy also plays a role in the vote shares of the Syrian Social Nationalist Party, where higher unemployment is associated with lower vote shares. Finally, more developed districts, with better access to water, vote more for the Progressive Socialist Party. The richer model in our analysis is that of the Independents. Not only does this model exhibit the highest number of observations, but it also includes new significant variables such as bombings and assassinations and the number of independent ministers in the government. Districts that witnessed more violence penalized independents as voters might be polarized toward established parties. The analysis of the results also suggests that Independents benefit when the number of ministers is large. 82 CHAPTER 3: VOTER TURNOUT AT THE LEBANESE PARLIAMENTARY ELECTIONS Abstract This paper studies the impact of economic, political, sociodemographic, and institutional factors on the Lebanese voter’s turnout. We estimate a dynamic turnout model using panel datasets covering 26 administrative districts over the five Lebanese parliamentary elections that occurred between 1996 and 2018. Although Lebanon is one of the few democracies in the Middle East, participation rates in the elections under study were significantly lower than those in Western democracies. In this multi-confessional country, it is not a surprise to find statistically significant results of the religious fragmentation factor. Nevertheless, other factors do matter as well in explaining turnout. Applying the system-GMM method, we found that an increase in the winning margin and unemployment lower participation in elections. Keywords: Elections; Unemployment, Log population; Turnout; Winning margin. 3.1. Introduction Participation in elections is a fundamental act of democracy. Political economists and political scientists have long been interested in investigating the reasons that explain voter turnout at the election polls. A vast and growing literature exists on participation in established and young democracies. From a theoretical perspective, Downs's ‘rational choice’ model explains people's decision to vote based on a cost-benefit analysis. However, it ignores an essential expressive feature, which is the moral obligation to vote (Fiorina, 1976; Blais et al., 2000). The group coordination process (Feddersen, 2004), the amount of available information about the candidate (Matsusaka, 1995), and the candidate's ethical approach (Goodin & Roberts, 1975) have also been considered theoretically by scholars. On the empirical level, studies have stressed economic, sociodemographic, political, and institutional factors. In a recent meta-analysis of national and subnational elections, Cancela and Geys (2016) found that campaign expenditures, election closeness, and registration requirements have more explanatory power in explaining voter turnout in national elections; in contrast, population size and its composition, simultaneous elections, and the electoral system play a more critical role in explaining participation in subnational elections. 83 Most of these studies were conducted in well-established western democracies; very few were carried out in young democracies, and fewer in authoritarian and Arab countries. Lebanon is one of the few democracies in the Middle East where political organizations with sectarian orientations play a crucial role in meeting the basic needs of the population (Cammett & Issar, 2010), and compete to mobilize as many of the voters to attend the polls. However, very few studies have been devoted to revealing the determinants of voter turnout in Lebanon (Harik, 1980; Alhaber, 2007; de Miguel et al., 2015; Mourad & Garrote Sanchez, 2019; Garrote Sanchez, 2021). Furthermore, the existing literature presents individual-level analysis focusing on just one election. In this paper, we analyze the factors that determined voter turnout in the 1996, 2000, 2005, 2009, and 2018 parliamentary elections. The analysis is implemented in 26 administrative districts and investigates which factors influenced the probability that a voter casts a vote at the district level. Different types of variables are used. They range from economic to sociodemographic, institutional, and political factors. To the best of our knowledge, this is the first aggregate-level study that investigates the above-mentioned factors over five Lebanese parliamentary elections. The paper is organized as follows: Section 3.2 reviews the literature on voter turnout; Section 3.3 briefly presents the Lebanese setting; Section 3.4 discusses the empirical model and defines the data. The empirical results are shown in Section 3.5, and Section 3.6 concludes the study. 3.2. Literature review Early theoretical models explaining voter turnout were developed by political scientists and were rooted in sociological and socio-psychological tradition. These models were based on expressive components of voting decisions, such as political interest, partisanship, sense of duty, and political efficacy. Voters are driven by the utility they get from expressing their political preferences, showing solidarity with their peer groups, and/or by contributing to the functioning of democracy and fulfilling of a civic duty (Fiorina, 1976). However, expressive voting ignores the economic aspect of political action and the rationality of voters. In his rational choice theory, Downs (1957) showed that the decision to vote is purely instrumental. One may decide to vote as the result of a personal cost-benefit calculation in which the expected benefits of voting should outweigh its costs. The anticipated benefits of voting are equal to the value placed on the difference 84 in the positions of candidates multiplied by the probability that one vote will influence the election. Voters go to the polls to help their preferred candidate win. However, Downs (1957) realized that, in mass elections, the probability that an individual's vote affects the outcome is minimal, such that the costs of voting will outweigh the expected benefits. So his model predicted abstention. Yet, rational people continue to vote, a phenomenon known as the ‘paradox of voting’ (Fiorina, 1976; Mueller, 1989). Fiorina (1976) and Riker and Ordeshook (1968) extended Downs’ rational choice model by adding a socialpsychological ingredient. The latter posit that citizens go to the polls because they may derive another kind of satisfaction, known as 'citizen duty', from voting itself. Such theorists contend that psychological satisfaction might outweigh the cost of voting. Their model includes both instrumental and expressive components of the voting decision.32 According to Blais et al. (2000), the rational choice model provides a useful but limited explanation of voting. In their study, based on a survey that was explicitly designed to test the rational choice model, the authors found that the moral obligation to vote, interpreted exogenously, is the most compelling motivation to go to the polls. Campbell et al. (1960) found that American voters with a strong sense of civic duty have a higher turnout rate than those with no sense of civic duty. In the same spirit, Verba et al. (1995), Tullock (2000), Clarke et al. (2004) and Blais and Galais (2016) suggest that civic duty is the essential motivation to vote. Besides instrumental and expressive components of voting, scholars have considered three different theoretical approaches explaining observed participation levels at elections: group-based, information-based and ethical approaches. Group-based theories emphasize the importance of the coordination process among voters to affect the election outcome. Voters might participate in elections because they are directly coordinated and rewarded by leaders (Feddersen, 2004). Filer et al. (1993) developed and empirically tested a model of rational choice that incorporates group behavior. In their model, voting may be rational, since the expected benefits can exceed the cost of voting at the group level. They examined the relationship between group-expected benefits and the cost of voting and income. Given that the government is the vehicle for income redistribution, elections are assumed to be the battlegrounds in the struggle over income distribution. If candidates differ over the 32 For a revision of theoretical literature on turnout see Dhillon and Peralta (2002) and Smets and Ham (2013). 85 progressivity of the redistribution, individuals with high income have more to gain or to lose from the outcome of the election and will exhibit higher turnout rates. On the other hand, individuals with a high level of income place a greater value on time and thus find it costlier to go to the polls and vote in comparison to those with a lower level of income. The information-based theories explain the impact of available information on an individual's decision to vote. Matsusaka (1995) developed an economic model of voter turnout, where he showed that voters with little information about the candidates will abstain from voting. Such behavior is in line with the cost-benefit calculation of the rational choice model, where a person's expected benefit from casting a decisive vote increases in line with his certainty that he is supporting the best candidate. As a result, voter turnout will increase as people become more confident and sure about the quality of candidates. Such behavior requires more information. The probability of turnout increases as long as voters can access information at a low cost. Similar results were attained by Feddersen and Pesendorfer (1999), who analyzed a model in which voters have asymmetric information and diverse preferences. They demonstrated how private information combined with preference diversity could result in two contradictory results: the turnout rate of more informed voters is higher than those with less information. However, turnout does not necessarily increase if the fraction of the informed electorate increases. This is due to the fact that, when the fraction of informed voters increases, the informativeness of the election result also increases and uninformed voters become more inclined to abstain. According to the ethical approach, voters will consider the utilities of others when voting. They have moral preferences that will be included in their welfare function. Political men will behave ethically whenever the losses they expect from doing so are low enough to be outweighed by rewards. Politicians try to gain support among voters by a "moralizing crusade rather than focusing on questions of who gets what, when, and how" (Goodin & Roberts, 1975, p. 928). On the empirical level, several explanatory variables were found to explain voter turnout. They are typically grouped into three categories: political, institutional, and socioeconomic variables (Geys, 2006b; Cancela & Geys, 2016). Regarding political factors, three variables are found to have a substantial impact on voter turnout: electoral closeness, campaign expenditures, and political fragmentation. Most of the empirical literature found a positive effect on the probability of voters participating in elections resulting from these three variables 86 (Cancela & Geys, 2016). The impact of closeness on turnout in the election goes back to the assumption of the instrumental behavior defined by Downs (1957), where a closer election is expected to increase participation since each vote can have a decisive impact on the output. Cox and Munger (1989) and Shachar and Nalebuff (1999) argue that the positive effect of closeness in elections is due to the mobilization effects of political parties and candidates to persuade unconvinced voters to cast their votes. In his review of aggregate-level research, Geys (2006a) found a positive association between turnout and election closeness in 206 out of 362 empirical tests and in 36 out of 52 studies. Similar results were obtained by Cancela and Geys (2016), who extended the meta-analysis of Geys by adding 102 studies. More recently, Dubois and Leprince (2017) found evidence that closer elections are expected to have a higher turnout rate in small French towns in Brittany. However, in their meta-analysis of individual-level research on voter turnout, Smets and Ham (2013) did not find a statistically significant impact of electoral closeness on turnout at the national level. Higher campaign expenditures appear to be associated with higher turnout, since this might increase awareness and information within the electorate and decrease the costs of information acquisition (Chapman & Palda, 1983). They reflect mobilization efforts by political parties (Cox & Munger, 1989; Fauvelle-Aymar & François, 2005). Empirical findings show that electoral spending is more significant in national, compared to subnational elections (Cancela & Geys, 2016). Regarding political fragmentation, the number of studies confirming a positive effect of fragmentation on turnout is decreasing with time. Geys (2006a) discussed inconclusive results, whereas Cancela and Geys (2016) considered that political fragmentation has little direct, independent relationship to voter turnout. The second category of covariates, which are regarded as the most potent determinants of turnout (Jackman, 1987; Jackman & Miller, 1995), are institutional variables such as the electoral system, compulsory voting, simultaneous elections, and registration requirements. Turnout is affected differently under different electoral systems. Proportional systems are considered to increase the probability of voters going to the polls in the belief that their vote might make a difference. In majoritarian systems, the disproportion between voters and seats leads the former to think that their vote is of no importance (Geys, 2006b). A proportional system increases competition at the district level (Blais & 93 Lebanese citizens elect their representatives in parliament by direct ballot in one round, according to the electoral law in effect (The Lebanese Constitution, 1995). The electoral system used during the first five parliamentary elections that took place following the end of the civil war (1992, 1996, 2000, 2005, and 2009) was a plurality system (First-Past-The-Post FPTP). A proportional system was adopted for the 2018 elections. Candidates run for elections in a multi-seat, multi-sect constituency, and the mandate is for four years. The only exception is the last election (2018), which took place nine years after the previous one (2009). According to the plurality system that was adopted until 2009, candidates are gathered in competing but open electoral lists balanced across sectarian lines (Salloukh et al., 2015). They can run either as independents or by the appointment of their political parties (1960 Electoral Law, 1960). The candidate contesting for a seat must be from the same confession to which the seat is allocated. Seats are won by whichever candidate from that confession gains the plurality (i.e., the highest number) of the votes cast. If there is more than one seat allocated to a confession, the seats are won by as many candidates as available seats. Voters vote in their ancestral village rather than their residential district. Regardless of their confessional affiliation, voters vote for all seats in the district in a single ballot paper. They are entitled to choose a number of candidates equal to the number of seats assigned to the constituency (Harik, 1980). They have the right to remove or add candidates from pre-arranged electoral lists, as long as the sectarian balance of the list remains the same (Salloukh et al., 2015). In 2017, a new electoral law based on proportional representation was enacted. This electoral law gave, for the first time, the opportunity for registered Lebanese expatriates the right to vote from their country of residence. Each voter votes for one competing list and is entitled to give his preferential vote to one candidate only in the respective administrative district. The competing list should obtain a minimum number of votes (threshold) to win seats in the parliament. This minimum is equal to the total number of voters divided by the number of seats in the district. Competing lists that do not get the threshold are excluded from the election. The threshold is counted again after excluding the total number of votes of the disqualified lists. Seats are distributed to eligible lists that gained the largest percentage of votes remaining from the first division. 94 Candidates in the eligible lists are sorted from first to last according to the percentage of preferential votes each candidate receives in the administrative district. The first seat is assigned to the candidate who gets the highest percentage of votes. The second-ranked candidate on the list will fill the second seat, and so on. The following two conditions should be respected when distributing seats: - The seat should be vacant as per the distribution of seats per sect in the respective administrative district. If the seat for a specific sect has already been filled by a winning candidate and is no longer vacant, candidates running for this seat are excluded from the competition. - If the distribution process reaches a candidate that belongs to a list that has fulfilled its quota of seats, the seat is passed on to the next eligible candidate (from another list) (Parliament Members Election Law, 2017). Throughout all parliamentary elections, seats remained equally distributed between Christian and Muslim communities, proportionally among the confessional groups within each of the two religions, and proportionally among geographic regions. 3.3.2. Turnout from 1992-2018 Lebanon's first postwar parliamentary election took place in 1992 amid a significant opposition from many political and religious leaders calling to postpone it. The turnout rate was 30.34%, the lowest since independence. Byblos constituency witnessed the lowest turnout of 6.52% while Baalbek-Hermel the highest with 51.77%. Most of the electorate was either unconcerned or opposed to the elections. Areas with a majority of Christian electorate boycotted the elections. They objected to voting in the presence of Syrian troops (El Khazen, 1998). The turnout in the 1996 parliamentary elections was higher than that of 1992, due to boycott weakening and the direct participation of Rafik Hariri35 with an important electoral list in Beirut. Hariri was able to attract voters, especially the low-income residents of Beirut, by delivering handouts (Sleiman, 1998). Several factors 35 Rafik Hariri was a Lebanese businessman who made most of his career and wealth in Saudi Arabia. He was the prime minister for 5 cabinets from 1992-1998 and from 2000-2004. Hariri participated in constructing the Taif agreement and in reconstructing the city center of Beirut at the end of the Lebanese civil war. His economic plan was mainly based on borrowing to reconstruct leading to a significant increase in the country’s debt. His tenure resulted from a geopolitical agreement mainly between Saudi Arabia, Syria and the USA. His relationship with Syria deteriorated a couple of years before his assassination in February 2005. 95 played a role in weakening the boycott of Christians. On the one hand, some Christian leaders (Albert Moukahyber and Jean Hawwat), who were previously opposing the elections, called for mass participation in the 1996 elections. On the other hand, there was an increasing demand from the USA, France, and the Vatican to participate and not to repeat the experience of the 1992 election (Salem, 1998). On a local level, the government interfered directly and indirectly by dumping the constituencies, mainly in North Lebanon, with electoral lists and candidates (Nassif, 1998). Despite the decrease in turnout rate from 46% in 1996 to around 43% in 2000, the latter was highly competitive. Many regional and local events happened during this period: the withdrawal of Israeli troops from the South of Lebanon, the death of the Syrian president, Hafez El Assad, and the return of former president of the Republic, Amin Gemayel from exile. Although other Christian leaders were still calling for boycotting elections, Christian voters were no longer convinced of abstaining (El Khazen, 2002). In an opinion survey conducted by Statistics Lebanon 75.52% considered voting as a national duty, while 41.97% participated in supporting a particular candidate (Shaoul, 2002). Religious and sectarian mobilization was the common factor that characterized the 2005 parliamentary elections (Sleiman, 2007). The assassination of PM, Rafik Hariri, and the other staggering events divided the country into two camps: 14 March and 8 March (Rizk, 2007). The turnout in Beirut was low mainly because the results were known way ahead of the election day. No one doubted that the electoral list headed by the son of the late PM, Rafik Hariri, would win all seats in the district. Adding to this, nine candidates won by walkover (Gebara, 2007). In the South of Lebanon, turnout was close to 50% favoring 8 March (Atrissi, 2007). A high turnout rate was achieved in Mount Lebanon, where it varied between 62.84% (first electoral district) to 51.21% (second electoral district) (Kosseifi, 2007). In the northern district, the Sunni were highly mobilized by 14 March, while Christians, Shiite, and the Alawite voted massively and unconditionally in favor of 8 March lists except in Bcharreh, Batroun, and Koura (Doueihi, 2007). The 2009 parliamentary elections took place after an 18-month political crisis that ended with the signing of the Doha agreement36 in May 2008. The elections took place according to the middle size electoral districts 36 Two years after the assassination of PM Rafik Hariri, Lebanon was deeply divided along sectarian lines between Shiite Muslims, Sunni Muslims, and Maronite Christians. Lebanon was also politically fragmented between pro-Syrian and pro-Hezbollah factions on one end of the equation and pro-western enclaves on the other end. Rival political leaders met on May 18, 2008, in Doha, Qatar, and reached the Doha agreement, which stipulated the formation of a new government and the election of a new President of the Republic. Effectively, a couple of days later, Michel Suleiman was elected President, and Fouad Siniora appointed as PM. 96 of the 1960 law, which liberate Christian candidates from the Muslim vote in many constituencies. Around 120,000 expatriates returned to Lebanon to vote with their travel expenses covered by political parties (approximately 80,000 were brought by 14 March and another 30,000 by 8 March) (Salloukh et al., 2015). On 6 May 2018, Lebanon held its first parliamentary elections in nine years. A new electoral law was passed in 2017, which was based on proportional representation and allowed expatriates to vote for the first time in their country of residence. The turnout rate was 49.7% for residents and 56% for registered voters among the diaspora. The percentage of voters who are expatriates is 2.51% (46,799 out of a total 1,861,203) (United Nations Development Programme, 2018). The highest turnout rates were seen in Keserwan, Byblos, and Hermel, while the lowest were in Beirut, Tripoli, Bcharre, Zgharta, Koura, Bint Jbeil, Hasbaya, and Rachaya. Regarding the participation rate per religious affiliation, the highest turnout rate was for Shiite voters (54%) and the lowest for Armenian Catholics (26%) (Mourad & Garrote Sanchez, 2019). 97 Table 29 Turnout rates per districts 1992-2018 DISTRICTS / CAZAS SEATS 1992 1996 2000 2005 2009 2018 Voters %Share 1 Beirut 1 19 16.28 37.64 40.35 31.28 45.18 42.93 2 Beirut 2 34.22 35.67 30.52 38.86 40.79 3 Beirut 3 31.53 32.51 21.31 29.25 34.80 4 Keserwan 5 2.22 54.48 57.37 63.97 67.62 67.09 5 Byblos 3 59.80 60.14 60.95 66.72 65.94 6 Matn 8 45.85 47.09 51.11 56.66 51.38 7 Baabda 6 44.22 42.61 53.27 55.77 48.18 8 Alay 5 48.39 48.67 56.22 51.45 50.39 9 Chouf 8 53.16 52.02 49.01 50.37 53.89 10 Sidon 2 38.08 61.65 49.99 42.23 67.58 56.71 11 Sidon Villages 3 49.74 52.19 52.02 53.64 51.88 12 Tyre 4 67.23 45.80 47.19 48.96 47.92 13 Jezzine 3 39.07 39.26 21.83 53.93 53.14 14 Bint Jbeil 3 37.60 40.50 41.27 42.87 43.62 15 Marjeyoun-Hasbaya 5 30.38 41.02 41.21 46.80 49.81 16 Nabatieh 3 62.88 52.47 55.46 56.45 55.51 17 Akkar 7 30.81 39.49 40.47 53.12 53.95 48.26 18 Bcharreh 2 26.59 23.62 36.80 37.01 39.94 19 Minieh-Donnieh 3 54.70 51.74 55.53 56.41 51.16 20 Tripoli 8 38.35 37.87 41.88 45.83 39.63 21 Zgharta 3 40.89 41.29 44.20 48.43 46.61 22 Koura 3 42.47 40.37 41.64 47.44 45.52 23 Batroun 2 46.81 40.35 52.02 56.32 55.73 24 Baalbeck-Hermel 10 46.09 61.85 45.93 51.54 49.30 60.28 25 Zahle 7 47.35 43.13 49.29 53.09 53.57 26 West Bekaa-Rachaya 6 46.66 39.55 44.23 53.26 47.44 TOTAL 128 30.34 46.32 43.81 46.06 50.38 49.68 Sources : Addiyar (1996); Annahar (1996, 1997); LADE et al . (2010); Ministry of Interior and Municipalities (1996, 2005, 2006, 2009, 2010, 2018); Assafir (2005a, 2005b, 2005c, 2005d); Rizk (2002); Information International (2009a, 2009b, 2009c, 2009d, 2009e). 98 3.4. Data and model specification Panel data models are estimated for the 26 administrative districts in Lebanon during the parliamentary elections of 1996, 2000, 2005, 2009 and 2018. Our empirical model is of the following form: , , , , , , , i d t d t d t d t d t d t d t TURNOUT a bECO cPOL dSOCDEM eINST fREL         (4) d =1,…, 26 t =1996, 2000, 2005, 2009 and 2018 i = Sunni, Shiite, Druze, Alawite, Maronite, Greek Catholic, Greek Orthodox, Protestant Evangelical, Armenian Catholic, Armenian Orthodox, Christian Minorities, Others. Where d represents the district, t the election year and i the religious sect. The dependent variable is the percentage of registered voters that turned out to vote in district d in the election held in year t . represents the error term, a, and e are parameters to be estimated, and b, c, d, and f are vectors of parameters to be estimated. The dependent variable was computed using data extracted from the Ministry of the Interior and Municipalities, and Information International. The right-hand side of the equation includes explanatory variables that are associated with economic, political, sociodemographic, institutional, and religious factors. ECO is a vector of economic variables that includes: a) Log real income per capita : the logarithm of real income per capita in district d during election year t, and b) Unemployment % : the unemployment rate in district d during election year t. The literature distinguishes between two opposite effects of the economy on turnout: A mobilization effect where people participate more in elections during economic hardships to manifest their discontent about the government’s policies (Schlozman & Verba, 1979). A withdrawal effect where voters suffering from economic adversity focus more on solving their problems and improving their well-being rather than caring for politics (Rosenstone, 1982). A non-linear effect was also found where the presence and magnitude of withdrawal or mobilization effects depend on economic performance (Martins & Veiga, 2013). ,dt  99 Data for economic variables are extracted from the Central Administration of Statistics (CAS), except for 2000 log real income per capita, which was obtained from the World Development Indicators. The latter gives estimates at the country level and not the district level. Therefore, we considered that the geographic distribution of income (per district) is the same as previously reported by CAS,37 and we extrapolated for the missing year using WDI figures.38 The vector of political variables, POL , includes: a) Past Turnout : is the percentage of voter turnout in district d in the election held in year t-4.39 It reflects habit formation. The link between past turnout and the current level of participation reflects political interest, partisanship, or even an attitude that reinforces self-image (Smets & Ham, 2013). It is expected to be positively associated with the current turnout percentage (Cancela & Geys, 2016). b) Effective Number of Parties : is "the number of hypothetical equal-size parties that would have the same total effect on fractionalization of the system, as have the actual parties of unequal size" (Laakso & Taagepera, 1979: p.4). It proxies political fragmentation and is calculated by: 2 2 1 1 n i i N p   (5) Where pi is the fractional share of votes of party i . An increase in the effective number of parties is expected to be associated with a higher turnout percentage (Geys, 2006a). However, recent studies show that this relation is not very robust (Cancela & Geys, 2016). c) Winning margin : is the difference in the percentage of votes between the most voted list and the second most voted list in district d in election year t . A higher winning margin is expected to be associated with a low turnout rate (Cancela & Geys, 2016). 37 According to Ministry of Social Affairs & United Nations Development Programme (2007) and CAS (2020), the distribution of income levels per governorate did not change significantly over the years. From 1995 till 2020, the highest incomes are perceived in Beirut and Mount Lebanon, followed by the North, Bekaa and the South of Lebanon. 38 For the year 2009, we used the figures of 2007, since no study was conducted in 2009. 39 Except for 2005 that took place 5 after the previous election, and 2018 that occurred 9 years after the previous one. 100 Data regarding the share of votes of each party and voter turnout is extracted from the Ministry of Interior and Municipalities and Information International. SOCDEM is a vector of sociodemographic variables that includes: a) Population : is proxied by the logarithm of the number of registered voters in district d during election year t . It is expected that a larger population is associated with a lower turnout rate as the probability that a single vote affects the results is smaller (Geys, 2006a). Squared log number of voters will also be used to test the existence of a non-linear relationship with the turnout (Juel et al., 2014). b) School Enrollment % : is the school enrollment rate in district d during election year t and is expected to be positively correlated with the turnout. According to Blais (2000), higher education is linked to more participation in elections. Martins & Veiga (2013) used the percentage of the population employed in the tertiary sector as a measure of education and found that turnout in municipal elections falls concurrently with a more significant percentage of workers employed in the tertiary sector. Given these contradictory results, we do not have prior expectations regarding the effect of school enrollment on the turnout rate. Data for the number of registered voters is extracted from the Ministry of Interior and Municipalities and the Lebanese Center for Policy Studies. The school enrollment rate is extracted from the CAS. As for the vector of Religious variables: it includes the share of voters belonging to the religious sect i in district d during election year t . We will use these variables as a proxy for religious homogeneity. Data is extracted from the Ministry of Interior and Municipalities and from Information International. One institutional variable has been used in our model, which is the proportional system . This variable takes the value of one for the proportional electoral system and zero for the plurality electoral system. The parliamentary elections of 1996, 2000, 2005, and 2009 were conducted under a plurality system (First-PastThe-Vote) and the 2018 elections under a proportional system. Descriptive statistics for the variables used in the empirical model are reported in Table 30. Turnout was, on average, 47.5%, with the lowest rate of 21.3% and the highest of 67.6%. Regarding economic variables, the average unemployment rate is 8.97 %. It reached a maximum of 17.8% in a specific district/year, but did not go below 4.3%. Real per capita income (2010 constant USD) varies between 1,728 USD and 13,341 USD. 101 As for sociodemographic variables, population size varied between 9,798 and 190,268 persons/district, and the school enrollment rate was 88.8% on average with a maximum of 99.3% and a minimum of 61%. Regarding political variables, the winning margin was, on average, 35.3%. The winning margin reached a maximum of 96.3% and a minimum of 0.017%. The effective number of parties was, on average, 3.06. In some districts, it reached one while in other districts, it reached a maximum of 8.205. On average, past turnout is 43.4%, with a minimum of 6.52% and a maximum of 67.6%. When religious affiliation is taken into consideration, we notice the following distribution of voters: 23.8% Shiite sect; 23.7% Sunni sect; 28% Christian Maronite; 8.55% Greek Orthodox; 4.99% Druze; 4.98% Greek Catholic; 2% Armenian Orthodox; 1.48% Christian minorities; 0.591% Armenian Catholic; 0.519% belong to other religious sects; 0.487% Muslim Alawite; and 0.363% Protestant Evangelical. 102 Table 30 Descriptive statistics (1) (2) (3) (4) (5) (6) VARIABLES N mean Sd min Max Sources Turnout in district 130 0.475 0.0958 0.213 0.676 Addiyar (1996); Annahar (1996, 1997); Inf.Int. (2009a; 2009b; 2009c; 2009d; 2009e); MOIM (1996; 2005; 2006; 2009; 2010; 2018); Rizk (2002); LEDA et al. (2010) ; Lebanese Association for Democratic Elections (2014) Economic variables Unemployment % in district 130 0.0897 0.0271 0.0430 0.178 (Central Administration of Statistics, 1996, 1997c, 1997a, 2000); (Central Administration of Statistics et al ., 2008); (Central Administration of Statistics et al., 2020 ) ; CAS and UNICEF (2010); MOSA and UNDP (2007); UNDP (2008); World Bank (2019) Real income per capita (constant 2010 USD) 130 5,721 3,133 1,728 13,341 Sociodemographic variables School Enrollment % 130 0.888 0.110 0.610 0.993 (Central Administration of Statistics, 1996, 1997b, 1997a, 2000, 2005) (Central Administration of Statistics et al ., 2008); (Central Administration of Statistics et al., 2020 ) ; MOIM (1996; 2005; 2006; 2009; 2010; 2018) Population 130 55,702 28,789 9,798 190,268 Political variables Winning Margin 128 0.353 0.280 0.00170 0.963 Addiyar (1996); Annahar (1996, 1997); Inf.Int. (2009a, 2009b, 2009c, 2009d, 2009e); MOIM (1996, 2005, 2006, 2009, 2010, 2018); Rizk (2002); LEDA et al . (2010) ; Lebanese Association for Democratic Elections (2014) Effective Number of Parties 128 3.060 1.481 1 8.205 Actual Number of Parties 130 5.238 2.676 1 12 Past Turnout 130 0.434 0.121 0.0652 0.676 109 Table 33 Main model using Beta Regression and Fractional Probit with district dummies (1) (2) (3) (4) VARIABLES Fractional Probit Marginal effects FP Beta regression Marginal effects BR Past turnout 1.389*** 0.553*** 2.238*** 0.558*** (6.664) (6.674) (8.933) (8.934) Effective Number of Parties -0.0234* -0.00934* -0.0370** -0.00923** (-1.727) (-1.726) (-1.992) (-1.992) Winning Margin -0.195*** -0.0777*** -0.310*** -0.0773*** (-3.483) (-3.483) (-3.432) (-3.432) Log Population 0.112*** 0.0448*** 0.181*** 0.0451*** (2.994) (2.994) (3.394) (3.394) School Enrollment -0.226 -0.0901 -0.311 -0.0776 (-0.594) (-0.594) (-0.531) (-0.531) Unemployment -2.279*** -0.908*** -3.655*** -0.912*** (-3.089) (-3.088) (-2.894) (-2.894) 2000 election -0.289*** -0.114*** -0.457*** -0.113*** (-4.999) (-5.011) (-5.437) (-5.505) 2005 election -0.135 -0.0539 -0.227* -0.0568* (-1.499) (-1.501) (-1.784) (-1.788) 2009 election -0.102 -0.0405 -0.174 -0.0434 (-1.022) (-1.023) (-1.240) (-1.241) 2018 election -0.110 -0.0437 -0.186 -0.0464 (-0.984) (-0.985) (-1.137) (-1.138) Observations 128 128 Note: The dependent variable is the percentage of registered voters who turned out to vote. The estimation method is Fractional Probit in column 1 and Beta Regression in column 3. The marginal effects are reported in columns 2 and 4, respectively. All models include and a constant term. Robust z-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1 Second, we exclude 2018 from the analysis. This election is particular due to many factors, including its timing nine years after the previous one, and the adoption of a new electoral system (a proportional one). Similar results to those of the system-GMM baseline model were obtained in terms of statistically significant covariates. 3.6. Religious fragmentation The religious factors have proven to be strong predictors of parties’ vote share as seen in chapter 2. However, the low variability of the percentage of voters belonging to each religious sect did not allow us to estimate efficiently their effect on turnout. To overcome this weakness, we referred to two alternatives proxies of the 110 religious factor. First, we calculated the religious fragmentation by referring to equation 5 and replacing the fractional share of votes of party i with the fractional share of the number of voters belonging to confession i (ri). 2 2 1 1 n i i N r   (6) This variable was added to the main model (column 4 Table 31) and the results are reported in column 1 of Table 34. The new variable is statistically significant and negatively signed. It shows that a one percentage point increase in religious fragmentation lowers the probability of turnout by 0.035 percentage points. The result is statistically significant at a 1% level of significance and is in line with the findings of Garrote Sanchez (2021). The latter showed that municipalities with a higher level of sectarian fragmentation had lower turnout rates than those who are more confessionally homogenous during the 2018 parliamentary elections. However, when we add this variable, the lagged dependent variable loses its statistical significance. Second, we added the share of Christian voters as a percentage of total voters to the main model. The results are reported in column 2. However, no statistically result was found for this variable. A similar analysis was conducted with the share of Muslim voters and similar results were obtained. Table 34 Religious fragmentation, Christian voters and Muslim voters (1) (2) (3) VARIABLES Religious Fragmentation Share of Christians Share of Muslims Past turnout 0.181 0.306** 0.307** (0.142) (0.123) (0.125) Winning Margin -0.201*** -0.101 -0.102 (0.058) (0.078) (0.079) Effective Number of Parties -0.003 -0.010 -0.009 (0.006) (0.008) (0.008) Log Population 0.111*** 0.082** 0.081** (0.024) (0.030) (0.030) School Enrollment -0.209 -0.291 -0.290 (0.165) (0.189) (0.190) Unemployment -0.644* -0.797** -0.802** (0.354) (0.348) (0.350) Religious Fragmentation -0.035*** 111 Table 34 (Continued) (0.011) Christians 0.045 (0.061) Muslims -0.042 (0.062) 1996 election 0.052* 0.084*** 0.084*** (0.030) (0.028) (0.028) 2005 election 0.070 0.095** 0.095** (0.044) (0.045) (0.045) 2009 election 0.071 0.110* 0.111* (0.053) (0.055) (0.056) 2018 election 0.058 0.105** 0.105** (0.051) (0.050) (0.050) Observations 128 128 128 Number of instruments 19 19 19 AR(1) p-value 0.0198 0.0184 0.0179 AR(2) p-value 0.0585 0.175 0.169 Hansen p-value 0.430 0.473 0.463 Note: The dependent variable is voter turnout as a percentage of registered voters. The estimation method is system-GMM. The model includes a constant term. Religious fragmentation was added to the model in column 1. The share of Christians was added to the model in column 2. The share of Muslim voters was added to the model in column 3. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. 3.7. Conclusion This chapter records the first attempt to test the effects of different economic, sociodemographic, institutional, and political factors on the turnout rates in 26 administrative districts during the 1996, 2000, 2005, 2009, and 2018 Lebanese parliamentary elections. Despite some data limitations, where some variables are only available at the national level rather than the district one, this chapter sheds light on interesting empirical findings. Different estimation methods were applied (OLS, FE, Beta regression, and Fractional Probit). However, the system-GMM is the primary estimation method on which the remainder of the paper was based. The systemGMM main model included: one economic variable (unemployment %); political variables (past turnout, the effective number of parties, and the winning margin); and sociodemographic variables (school enrollment %, and log population). 112 Our results suggest that Lebanese voters consider political factors when participating in elections in the same way as voters from other countries do. They tend to exhibit habit formation and they participate more when the margin of votes between competing parties is small. As for the effects of the economy on the probability of turnout, our results suggest that the worsening of the economic conditions discourages Lebanese voters from participating in elections. These results are in line with the withdrawal effect where voters put more attention on their well-being rather than politics. However, opposite results to the literature were found regarding the effects of sociodemographic factors. While previous studies found that voters turn less to the polls when the size of the population increases, the results of this chapter suggest that in more populated districts, more polling stations exist, and thus it is easier to vote. The religious factor remains an important variable to be considered when talking about elections in Lebanon mainly when referring to the religious fragmentation variable. 113 CONCLUSION This thesis has been organized into three independent papers on the subject of political economy. In the first paper, we studied the determinants of successful revolutions in a large dataset of countries, before restricting our analysis, in the second and third papers, to voter behavior and turnout in a single country, Lebanon. The first paper, Chapter 1, empirically investigated the impact of the economy, governance, information and communications technology (ICT), and media censorship, among other factors, on the probability of successful revolution in more than 150 countries over the 1996-2015 period. We considered successful revolutions as processes leading to a leader's exit from the office, based on large-scale popular participation. Forty-seven successful revolutions broke out during the period under study. Chapter 1 reevaluated some factors previously mentioned in the literature and investigated new ones. There is evidence that economic performance, mainly income per capita and income growth, reduces the likelihood of successful revolutions. Some variables that have surfaced with the recent Arab uprisings reported statistically significant effects. Such factors are mainly related to media censorship, governance, and ICT. An increase in media censorship increases the probability of successful revolutions; concerning governance and ICT indicators, their increase lowers the probability of successful revolutions. On the other hand, democracy reported statistically significant results only when interacted with income level and income growth. We found that income per capita reduces the probability of successful revolutions only in democratic countries. The effect of the 5-years average rate of real GDP growth on the likelihood of successful revolutions is of concern only in non-democratic countries. There was also evidence that oil income can lower the probability of successful revolutions, particularly in non-democratic countries. In general, this chapter's findings lead to relevant recommendations, particularly concerning economic policies that stimulate medium-term economic growth, and hence higher income levels, especially in nondemocratic countries. Such policies can lower the probability that governments will face successful revolutions, making them less prone to different types of political violence, such as violence against the government, demonstrations and riots, revolutionary wars, and successful coups d’état. 114 This paper did not show a direct effect of the regime type; this does not necessarily call into question the promotion of democracy. Instead, more democratization might be needed to promote good governance, dismantle cronyism and clientelism, control corruption, and improve regulatory quality and the rule of law – – all of which are peace-promoting. Governments are called to complement such policies by lowering media censorship. Less regime control over the press, coupled with more developed information and communications technology, enables citizens to better communicate their opinions and discontent about the government, which might avoid the need to resort to the street. Like any other paper on this topic, the small number of successful revolutions analyzed is a limitation. The use of alternative datasets and different definitions of revolutions might mitigate this problem, in addition to considering a more extended period. The second paper, Chapter 2, narrowed the analysis to a particular Arab country, Lebanon. It examined the effects of religious, socio-economic, and political variables on party choice in Lebanon. The research was applied to six political parties and independent candidates in 1996, 2000, 2005, and 2009 parliamentary elections. As one would expect, in a multi-confessional and diverse society such as Lebanon, religious factors are strong predictors of parties' vote shares. The percentage of voters belonging to specific religious sects is particularly relevant for explaining the Amal Movement's vote share, the Hezbollah, the Progressive Socialist Party, and the Independents of the March 14 coalition. Despite the importance of religious factors, socio-economic variables also explain Lebanese parties' vote shares. There is evidence that the economy, namely income per capita, positively influences the vote share of the Hezbollah and the Future Movement and negatively that of Independents. The economy also plays a role in the Syrian Social Nationalist Party's vote shares, where higher unemployment is associated with lower vote shares. Finally, more developed districts, with better access to water, vote more for the Progressive Socialist Party. A richer model for our analysis is that of the Independents, who are penalized in districts that witness more bombings and assassinations. There is also evidence that Independents benefit when the number of ministers is high. 115 The third paper, Chapter 3, is the first to investigate economic, sociodemographic, institutional, and political factors on the turnout rates in 26 administrative districts during 1996, 2000, 2005, 2009, and 2018 Lebanese parliamentary elections. Despite some data limitations (some variables are only available at the national level rather than at the district level), this chapter sheds light on interesting empirical findings. The results revealed that the economy affects the probability of voters turning out to the polls. Lebanese voters participate less in elections when the economic conditions are worsening. Political and sociodemographic factors also matter to Lebanese voters. There is evidence of habit formation and more interest in voting when the competition among leading political parties is high. Furthermore, sociodemographic factors seem to affect participation in elections. Namely, more populated districts with more polling stations witness higher participation in elections. Results of both Chapters 2 and 3 suggested that the economy matters to the Lebanese voters when deciding whether to vote or not, but it matters less when deciding on the party to which they want to cast their vote. The effects of the economy on the decision to participate in elections corroborate with the withdrawal effect where the economically-stressed Lebanese voters do not vote because they are too preoccupied to pay attention to politics. They also withdraw from politics believing that their participation is worthless as it has never succeeded in impacting the government’s economic policies, especially during hardships. However, when deciding on the party to which they want to cast their vote, Lebanese voters underestimate the economy. Such behavior is not surprising in such a diverse society, where sectarian affiliation affects Lebanese voter behavior. The Lebanese political system, in place since the Taif agreement, was and is still characterized by a consociational democracy (Horn, 2008) where sectarian elites control the political scene. The electoral laws and electoral districting adopted since the end of the civil war contributed to reinforcing this political sectarianism. The phasing out of such political sectarianism requires a more decentralized electoral districting, a federal one, where voters participate in elections to express their satisfaction/dissatisfaction with the economic performance rather than withdrawing. They choose their representatives based on economic considerations rather than their religious affiliation. Smaller and more religiously homogenous constituencies contribute to 116 shifting voter's priorities from preserving and defending their sects to improving their well-being through more economic development in their districts. Optimally, the government is called to implement a secular system. However, this is a long process that starts with a rethinking of the educational system. Future generations should be raised to become more tolerant of their societies' differences and prioritize their nation's interest over their religious affiliation. In October 2019, a planned increase in gasoline, tobacco, and WhatsApp taxes triggered nationwide protests in Lebanon against the ruling elite for their failure to provide for the population's basic needs. Today, more than one year after the so-called October revolution, the same ruling elite that governed the country since the end of the civil war is still in power. The economy is suffocating. Lebanon is thrust into its worst economic crisis in decades, with its currency collapsing, businesses shutting down, and prices skyrocketing with a threedigit inflation rate. For a long time, the Lebanese people were passive and sedentary and did not believe that a change was possible. Will this status quo change? Recent events have proved that the economy has not been able to trigger a successful revolution. Will the change come through other methods, such as participating more in the next parliamentary elections and changing the voting behavior? Future research would be essential to investigate further the socio-economic and cultural differences at a more disaggregated constituency level, like municipalities. This would contribute to a better understanding of the voting behavior and turnout of Lebanese voters. 117 REFERENCES 14 March Organization. (n.d.). The Objectives of 14 March . http://www.14march.org/pages.php?cat=MTMwODkx Abramowitz, A., & Saunders, K. (2006). Exploring the Bases of Partisanship in the American Electorate : Social Identity vs . Ideology. Political Research Quarterly , 59 (2), 175. Acemoglu, D., Hassan, T., & Tahoun, A. (2014). The Power of the Street: Evidence from Egypt’s Arab Spring. In Working paper (Issue October). Acemoglu, D., & Robinson, J. A. (2001). A Theory of Political Transitions. The American Economic Review , 91 (4), 938–963. Addiyar. (1996, September). Electors and Voters. Addiyar Newspaper . Aguiar-Conraria, L., & Magalhães, P. C. (2010). Referendum design , quorum rules and turnout. Public Choice , 144 , 63–81. Al Shark. (2005, December 5). Assassinations and assaults under Taef (Arabic). 16748 . Al Shark. (2006, February 28). 126 assassinations and 94 assassination attempts since Independence... and the perpetrator is still unkown (Arabic). 28 . Alcalde, J., & Romero-Medina, A. (2006). Coalition formation and stability. Social Choice and Welfare , 27 (2), 365–375. Alesina, A., Devleeschauwer, A., Easterly, W., & Kurlat, S. (2003). Fractionalization. Journal of Economic Growth , 8 , 155–194. Alesina, A., & La Ferrara, E. (2000). Participation in Heterogeneous Communities. Quarterly Journal of Economics , 115 (3), 847–904. Alhaber, R. (2007). Opinion polls and their indications on the parliamentary elections results. In The 2005 118 Lebanese Parliamentary Elections in the mid of Local and Regional Transformations (Arabic) (First, pp. 361–406). Lebanese Centre for Policy Studies. Alhayat. (1996, September). Confession, Seats, Candidates Numbers and Electors. Issue N.12256 . Amemiya, T. (1981). Qualitative Response Models : A Survey. Journal of Economic Literature , 19 (4), 1483– 1536. Andersen, & Heath. (2003). Social Identities and Political Cleavages : The Role of Political Context. The Journal of the Royal Statistical Society , 166 (3), 301–327. Anderson, C. J. (1995). The Dynamics of Public Support For Coalition Governments. Comparative Political Studies , 28 (3), 350–383. Annahar. (1996, September 4). Confession, seats and candidates (Arabic). 19534 . Annahar. (1997, September). Electors, Voters and Canceled Votes for each District (Arabic). Issue N.19531 . Annahar. (2008, May 20). Investigation in 7 May events (Arabic). 23354 . Ansani, A., & Daniele, V. (2012). About a Revolution: the Economic Motivations of the Arab Spring. International Journal of Development and Conflict , 3 (3). Arnesen, S. (2012). Forecasting Norwegian elections: Out of work and out of office. International Journal of Forecasting , 28 (4), 789–796. Atallah, G. (2000, September 1). Official Results 2000 Parliamentary Elections (Arabic). Assafir . Atrissi, T. (2007). The Parliamentary Elections in the District of South Lebanon. In The 2005 Lebanese Parliamentary Elections in the mid of Local and Regional Transformations (Arabic) (First, pp. 153– 180). Lebanese Centre for Policy Studies. Bahgat, K., Buhaug, H., & Urdal, H. (2017). Urban Social Disorder Codebook, version 2.0 . https://www.prio.org/Data/Armed-Conflict/Urban-Social-Disorder/ 125 Elsayyad, M., & Hanafy, S. (2014). Voting Islamist or voting secular? An empirical analysis of voting outcomes in Egypt’s “Arab Spring.” Public Choice , 160 , 109–130. Enterline, A. (1998). Regime changes and interstate conflict 1816–1992. Political Research Quarterly , 51 (2), 385–409. EU. (2005). Parliamentary Elections, Lebanon 2005 . European Union. (2009). Lebanon Final Report on the 7 June Parliamentary Elections . http://aceproject.org/ero-en/regions/mideast/LB/lebanon-final-report-parliamentary-elections Fauvelle-Aymar, C., & François, A. (2005). Campaigns, political preferences and turnout. An empirical study of the 1997 French legislatives elections. French Politics , 3 (1), 49–72. Fearon, J. D., & Laitin, D. (2003). War Ethnicity , Insurgency , and Civil War. The American Political Science Review , 97 (1), 75–90. Feddersen, T. (2004). Rational Choice Theory and the Paradox of Not Voting. Journal of Economic Perspectives , 18 (1), 99–112. Feddersen, T., & Pesendorfer, W. (1999). Abstention in Elections with Asymmetric Information and Diverse Preferences. The American Political Science Review , 93 (2), 381–398. Filer, J., & Kenny, L. (1980). Voter turnout and the benefits of voting. Public Choice , 35 , 575–585. Filer, J., Kenny, L., & Morton, R. (1993). Redistribution, Income, and Voting. American Journal of Political Science , 37 (1), 63–87. Fiorina. (1981). Retrospective Voting in American National Elections. Yale University Press. Fiorina, M. P. (1976). The Voting Decision : Instrumental and Expressive Aspects. The Journal of Politics , 38 (2), 390–413. Fornos, C., Power, T., & Garand, J. (2004). Explaining voter turnout in Latin America, 1980 to 2000. Comparative Political Studies , 37 (8), 909–940. 126 Franklin, M. N. (2004). Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945 . Cambridge University Press. Freedom House. (2017). Press Freedom Survey from 1980-2016 . https://freedomhouse.org/reporttypes/freedom-press Frey, & Schneider. (1978a). A politico-economic model of the United Kingdom. The Economic Journal , 88 (350), 243–253. Frey, & Schneider. (1978b). An Empirical Study of Politico-Economic Interaction in the United States. The Review of Economics and Statistics , 60 (2), 174–183. Friedman, M. (1977). Nobel Lecture: Inflation and Unemployment. Journal of Political Economy , 85 (3), 451–472. Gallego, A., Rico, G., & Anduiza, E. (2012). Disproportionality and voter turnout in new and old democracies. Electoral Studies , 31 (1), 159–169. Garmann, S. (2014). A note on electoral competition and turnout in run-off electoral systems : Taking into account both endogeneity and attenuation bias. Electoral Studies , 34 , 261–265. Garrote Sanchez, D. (2021). Understanding Turnout in the Lebanese Elections. In The Lebanese Center for Policy Studies . Gassebner, M., Gaston, N., & Lamla, M. (2011). The Inverse Domino Effect: Are Economic Reforms Contagious? International Economic Review , 52 (1), 183–200. Gebara, K. (2007). The Parliamentary Elections in Beirut. In The 2000 Lebanese Parliamentary Elections: between Reproduction and Change (Arabic) (First, pp. 139–151). Lebanese Centre for Policy Studies. Geys, B. (2006a). Explaining voter turnout: A review of aggregate-level research. Electoral Studies , 25 (4), 637–663. Geys, B. (2006b). “Rational” theories of voter turnout: A review. Political Studies Review , 4 (1), 16–35. 127 Gharaibeh, M. K. (2007). The Voting Behavior of Kuwaiti Citizens in the 2006 Parliamentary Elections: an Analytical Field Study. Digest of Middle East Studies , 16 (2), 1–21. Gibbs. (2005). Religiosity and Voting Behavior (No. 6). https://www.mckendree.edu/academics/scholars/issue6/gibbs.htm Global Security. (2013). Lebanon Political Parties . https://www.globalsecurity.org/military/world/lebanon/political-parties.htm Goemans, H., Gleditsch, K. S., & Chiozza, G. (2016a). Archigos: A Data Set on Leaders 1875–2015 version 4.1. In Archigos . http://www.rochester.edu/college/faculty/hgoemans/Archigos_4.1.pdf Goemans, H., Gleditsch, K. S., & Chiozza, G. (2016b). ARCHIGOS A Data Set on Leaders 1875 – 2015 (Version 4.). Goldstone, J. (1980). Theories of Revolution : The Third Generation. World Politics , 32 (3), 425–453. Goodhart, C. A. E., & Bhansali, R. J. (1970). Political Economy. Political Studies , 18 (1), 43–106. Goodin, R., & Roberts, K. (1975). The Ethical Voter. The American Political Science Review , 69 (3), 926– 928. Goodman, P., & Kramer, G. H. (1975). Comment on Arcelus and Meltzer. The American Political Science Review , 1255–1265. Green, D., Palmquist, B., & Schikler, E. (2002). Partisan Hearts and Minds: Political Parties and the Social Identities of Voters . Yale University Press. Greene, W. H. (2003). Models for Discrete Choice. In Econometric Analysis (5th ed., pp. 663–756). Pearson Education. Greene, W. H. (2010). Discrete Choice. In Econometric analysis (7th ed., pp. 681–759). Pearson Education. Gurr, T. R. (1968). Psychological Factors in Civil Violence. World Politics , 20 , 245–278. 128 Gurr, T. R. (1970). Why Men Rebel . Princeton University Press. Gygli, S., Haelg, F., Potrafke, N., & Sturm, J. (2019). The KOF Globalisation Index – revisited. The Review of International Organizations . Hajnal, Z., & Lewis, P. (2003). Municipal institutions and voter turnout in local elections. Urban Affairs Review , 38 (5), 645–668. Harik, I. (1980). Voting Participation and Political Integration in Lebanon 1943-1974. Middle Eastern Studies , 16 (1), 27–48. Hegre, H., Gissinger, R., & Gleditsch, N. P. (2003). Globalization and Internal Conflict Globalization and Armed Conflict. In G. Schneider, K. Barbieri, & N. P. Gleditsch (Eds.), Globalization and Armed Conflict . Lanham, MD: Rowman & Littlefield. Hibbs, D. A. (1982). The Dynamics of Political Support for American Presidents Among Occupational and Partisan Groups. American Journal of Political Science , 26 , 312–332. Highton, B. (2010). The Contextual Causes of Issue and Party Voting in American Presidential Elections. Political Behavior , 32 , 453–471. Hlasny, V., & Verme, P. (2013). Top incomes and the measurement of inequality in Egypt. World Bank Policy Research Working Paper No. 6557 , August . http://wwwwds.worldbank.org/servlet/WDSContentServer/WDSP/IB/2013/08/07/000158349_20130807171 651/Rendered/PDF/WPS6557.pdf Hodgson, R., & Maloney, J. (2012). Economic voting in Britain, 1857-1914. Electoral Studies , 31 (4), 668– 678. Horn. (2008). Lebanese Consociation: Assessing Accountability and Representativeness . Hourani, G. G., & Sensenig-Dabbous, E. (2012). Naturalized Citizens: Political Participation, Voting Behavior, and Impact on Elections in Lebanon (1996-2007). Journal of International Migration and Integration , 13 (2), 187–202. 129 Hudson, M. (1999). Lebanon after Taif: another reform opportunity lost? Arab Studies Quarterly , 21 (1), 27– 40. Huntington. (1968). Political order in changing societies . Yale University Press. Ianchovichina, E., Mottaghi, L., & Devarajan, S. (2015). Inequality, Uprisings, and Conflict in the Arab World. In MENA Economic Monitor (Issue October). IFES. (2009a). Lebanon’s 2009 Parliamentary Elections: The Lebanese Electoral System (Issue March). https://www.ifes.org/sites/default/files/ifes_lebanon_esb_paper030209_0.pdf IFES. (2009b). The political affiliation of Lebanese Parliamentarians and the composition of the different parliamentary blocs . IFES. (2011). Elections in Lebanon: Overview of the current 26 electoral districts (Issue September). Information International. (2005a). Bekaa. In Assafir & Koutoub (Eds.), The 2009 Lebanese Parliamentary Elections According to Constituencies, Incumbents and Religions (Arabic) (First, pp. 17, 109, 299). Information International. (2005b). South and Nabatieh Governorates. In Assafir & Koutoub (Eds.), The 2009 Lebanese Parliamentary Elections According to Constituencies, Incumbents and Religions (Arabic) (First, pp. 17, 30, 35, 62, 67, 100, 105, 130, 135, 174, 179,). Information International. (2009a). Beirut Governorate. In Assafir and Koutoub (Ed.), The 2009 Lebanese Parliamentary Elections according to constituencies, incumbents and religions (Arabic) (First, pp. 55, 91, 215). Information International. (2009b). Mount Lebanon Governorate. In Assafir & Koutoub (Eds.), The 2009 Lebanese Parliamentary Elections according to constituencies, incumbents and religions (Arabic) (First, pp. 123, 175, 181, 207, 213, 300, 305, 352, 357). Information International. (2009c). North Governorate. In Assafir & Koutoub (Eds.), The 2009 Lebanese Parliamentary Elections according to constituencies, incumbents and religions (Arabic) (First, pp. 17, 111, 117, 146,151, 176, 181, 212, 217, 226, 23). 130 Inoguchi, T. (1980). Economic Conditions and Mass Support in Japan. In P. Whitely (Ed.), Models of Political Economy (pp. 121–154). Sage. Institute for Economics and Peace. (2018). Global Peace Index 2018: Measuring Peace in a Complex World . http://static.visionofhumanity.org/sites/default/files/2015 Global Terrorism Index Report_0_0.pdf Ireland, T. (1967). The Rationale of Revolt. Papers on Non-Market Decision Making , 3 (1), 49–66. Jackman, R. (1987). Political Institutions and Voter Turnout in Industrial Democracies. American Political Science Review , 81 (2), 405–423. Jackman, R., & Miller, R. (1995). Voter turnout in the industrial democracies during the 1980s. Comparative Political Studies , 27 (4), 467–492. Jonung, L., & Wadensjoe, E. (1979). The Effect of Unemployment, Inflation and Real Income Growth on Government Popularity in Sweden. Scandinavian Journal of Economics , 81 , 343–353. Joya, A. (2011). The Egyptian Revolution: Crisis of Neoliberalism and the Potential for Democratic Politics. Review of African Political Economy , 38 (129), 367*386. Juel, J., Fiva, J. H., & Natvik, G. (2014). Voting when the stakes are high. Journal of Public Economics , 110 , 157–166. Kane, J. C., Craig, S. C., & Kenneth, D. W. (2004). Religion and Presidential Politics in Florida: A List Experiment. Social Science Quarterly , 85 (2), 281–293. Kelley, S. J., Ayres, R. E., & Bowen, W. G. (1967). Registration and voting: putting first things first. American Political Science Review , 61 , 359–379. Khondker, H. H. (2011). Role of the New Media in the Arab Spring. Globalizations , 8 (5), 675–679. Kiewiet, R. D., & Udell, M. (1998). Twenty five years after Kramer: an assessment of economic retrospective voting based upon improved estimates of income and unemployment. Economics and Politics , 10 , 219–249. 131 Knusten, C. H. (2014). Income Growth and Revolutions. In Social Science Quarterly (Vol. 95, Issue 4). Kosseifi, H. (2007). Mount Lebanon Parliamentary Elections. In 2005 Lebanese Parliamentary Elections in the mid of Local and Regional Transformations (Arabic) (First, pp. 183–224). Lebanese Centre for Policy Studies. Kostadinova, T. (2003). Voter turnout dynamics in post-Communist Europe. European Journal of Political Research , 42 , 741–759. Kramer, G. H. (1971). Short-Term Fluctuations in U.S. Voting Behavior, 1896-1964. American Political Science Association , 65 (1), 131–143. L’Orient Le Jour. (2000, September 1). Official Results of 2000 Parliamentary Elections . L’Orient Le Jour. (2007, September 24). Chronologie: les principaux attentats depuis octobre 2004. 12076 . Laakso, M., & Taagepera, R. (1979). “Effective” Number of Parties: A Measure with Application to West Europe. Comparative Political Studies , 12 (1), 3–27. Labaki, B. (1988). Confessional Communities , Social Stratification and Wars in Lebanon. Social Compass , XXXV (4), 533–561. LADE, Labs, Lamba, DI, & Seed. (2010). The Lebanese Elections Data Analysis (LEDA) . Lebanese Elections Data v2.0. Lafay, J.-D. (1985). Political Change and Stability of the Popularity Function: The French General Election of 1981. In Agathon Press (Ed.), Economic Conditions and Electoral Outcomes: The United States and Western Europe (pp. 78–98). Algora Publishing. Lambert, D. (1992). Zero-inflated poisson regression, with an application to defects in manufacturing. Technometrics , 34 (1), 1–14. Lebanese Association for Democratic Elections. (2014). Opinion Poll on the Electoral Issues in Lebanon (Issue August). 132 Lebanese Centre for Studies and Development. (2017). Distribution of students in public and private schools according to administrative districts . Parliament Members Election Law, 1 (2017). Lebanonwire. (n.d.). Members of the 2000 parliament . 2000. http://www.lebanonwire.com/parliament/members_2000.asp Lewis-Beck, M. S., Martini, N. F., & Kiewiet, D. R. (2013). The nature of economic perceptions in mass publics. Electoral Studies , 32 (3), 524–528. Lewis-Beck, Michael, & Stegmaier, M. (2019). Economic voting. In C. R., G. B., & V. S. (Eds.), Oxford Handbook of Public Choice (Issue February, pp. 247–265). Oxford University Press. Lewis-Beck, Micheal. (1988). Economics and Elections: The Major European Democracies . University of Michigan Press. Lewis-Beck, & Nadeau. (2000). French electoral institutions and the economic vote. Electoral Studies , 19 (2–3), 171–182. Lewis-Beck, & Rice. (1992). Forecasting Elections. Congressional Quarterly Press . Lewis-Beck, & Stegmaier. (2000). Economic Determinants of Electoral Outcomes. Annual Review of Political Science , 3 , 183–219. Lipset, S. M. (1959). Some Social Requisites of Democracy : Economic Development and Political Legitimacy. American Political Science Review , 53 (1), 69–105. Lockerbie, B. (2013). Race and religion: Voting behavior and political attitudes. Social Science Quarterly , 94 (4), 1145–1158. Lust-Okar, E. (2006). Elections under authoritarianism: Preliminary lessons from Jordan. Democratization , 13 (3), 456–471. Madsen, H. J. (1980). Electoral Outcomes and Macro-Economic Policies: The Scandinavian Cases. In 133 Models of Political Economy (Paul White, pp. 15–46). Sage. Magalhães, & Aguiar-Conraria. (2009). Growth, Centrism and Semi-Presidentialism: Forecasting the Portuguese General Elections. Electoral Studies , 28 (2), 314–321. Magalhães, Aguiar-Conraria, & Lewis-Beck. (2012). Forecasting Spanish elections. International Journal of Forecasting , 28 (4), 769–776. Malik, A., & Awadallah, B. (2011). The economics of the Arab spring. In Centre for the Study of African Studies (Vol. 44, Issue 0). Mansfield, E., & Snyder, J. (2005). Electing to Fight: Why Emerging Democracies Go to War. MIT Press. Maoz, Z. (1989). Joining the club of nations: Political development and international conflict, 1816–1976. International Studies Quarterly , 32 (2), 199–231. Maoz, Z. (1996). Domestic Sources of Global Change . ersity of Michigan Press. Marshall, M. (2017a). Major Episodes of Political Violence 1946-2016 . http://www.systemicpeace.org/warlist/warlist.htm Marshall, M. (2017b). Major Episodes of Political Violence and Conflict Regions, 1946-2016 . www.systemicpeace.org Marshall, M. G., Gurr, T. R., & Harff, B. (2017). PITFState Failure Problem Set: Internal Wars and Failures of Governance, 1955-2016. In Political Instability Task Force – Problem Set Codebook version 2016 . http://www.systemicpeace.org/inscr/PITFProbSetCodebook2016.pd Marshall, M., Gurr, T. R., & Harff, B. (2017). Monty G. Marshall, Ted Robert Gurr, and Barbara Harff . Political Instability Task Force. http://www.systemicpeace.org/inscr/PITFProbSetCodebook2016.pdf Martins, R., & Veiga, F. J. (2013). Economic performance and turnout at national and local elections. Public Choice , 157 (3–4), 429–448. Matsusaka, J. G. (1995). Explaining voter turnout patterns : An information theory. Public Choice , 84 , 91– 134 117. McGarty, C., Thomas, E., Lala, G., Smith, L., & Bliuc, A.-M. (2013). (2013). New technologies, new identities and the growth of mass opposition in the “Arab Spring. Political Psychology , 35 (6), 725– 740. Mcguire, R. A. (1981). Economic Causes of Late-Nineteenth Century Agrarian Unrest : New Evidence. Cambridge University Press , 41 (4), 835–852. Miller, A. H. (1978). Partisanship Reinstated? A Comparison of the 1972 and 1976 U.S. Presidential Elections. British Journal of Political Science , 8 (2), 129–152. Miller, A. H., E.Miller, W., Raine, A. S., & Thad A., B. (1976). A Majority Party in Disarray: Policy Polarization in the 1972 Election. American Political Science Review , 70 (3), 753–778. Miller, A. H., Wlezien, C., & Hildreth, A. (1991). A Reference Group Theory of Partisan Coalitions of Partisan Coalitions. The Journal of Politics , 53 (4), 1134–1149. Ministry of Interior and Municipalities. (1996, September 16). Official Results 1996 Parliamentary Elections (Arabic). Annahar . Ministry of Interior and Municipalities. (2005a, May 31). Official Results Beirut 2005 Parliamentary Elections (Arabic). Assafir . Ministry of Interior and Municipalities. (2005b, June 7). Official Results South and Nabatieh 2005 Parliamentary Elections (Arabic). Assafir . Ministry of Interior and Municipalities. (2005c, June 14). Official Results Bekaa 2005 Parliamentary Elections (Arabic). Assafir . Ministry of Interior and Municipalities. (2005d, June 14). Official Results Mount Lebanon 2005 Parliamentary Elections (Arabic). Assafir . Ministry of Interior and Municipalities. (2005e, June 21). Official Results North 2005 Parliamentary Elections (Arabic). Assafir .