DEPARTMENT OF POLITICAL SCIENCE AGAINST AUTOCRATIZATION: THE IMPACT OF NON-VIOLENT RESISTANCE CAMPAIGNS ON SUCCESSFUL DEMOCRATIC RESILIENCE A Comparative Quantitative Study from 1900 to 2020 Alexander Heinrich Master’s Thesis: 30 credits Programme: Master’s Programme in Political Science Date: 21/05/2024 Supervisor: Fabio Angiolillo Words: 17,554 Abstract How can democracies stop episodes of autocratization and return to previous levels? So far, research has underestimated the role of non-violent resistance in preserving democratic structures and freedoms. This thesis presents mechanisms how civil society organizations can use means of non-violent resistance to successfully achieve democratic resilience. Using a novel panel dataset spanning from 1900 to 2020, I employ a logistic regression to analyse the relationship between non-violent resistance and democratic resilience over time. The results show that one episode of non-violent resistance increases the probability of successful democratic resilience by 14%, rising to 29.7% with two episodes. These probabilities are considerably higher during the third wave of autocratization. Moreover, I find that experiences of non-violent resistance prior to the onset of autocratization contribute significantly to democratic resilience, with a remarkable 50.8% probability when it occurs one year before. The study highlights the effectiveness of non-violent resistance in bringing about regime change and subsequent democratization, offering hope in the face of the current global crisis of democracy. Keywords: non-violent resistance; civil society organizations; democratic resilience; autocratization Table of Contents 1. Introduction .................................................................................................................1 2. Literature Review ........................................................................................................4 2.1 Autocratization and Democratic Resilience .............................................................4 2.2 Civil Society Organizations and Regime Stability ..................................................5 2.3 Origins of Political Mobilization .............................................................................9 2.4 Strategies of Non-Violent Resistance .................................................................... 11 2.5 Mechanisms of Political Change through Non-Violent Means .............................12 2.6 Challenges Facing Social Movements ...................................................................14 3. Theoretical Model and Hypotheses ..........................................................................17 4. Research Design .........................................................................................................23 4.1 Dependent Variable ...............................................................................................23 4.2 Independent Variable .............................................................................................25 4.3 Estimation Strategy ................................................................................................27 5. Main Results ...............................................................................................................31 5.1 Logit Model 1900-2020 .........................................................................................31 5.2 Logit Model 1994-2020 .........................................................................................35 5.3 Logit Model 1900-1993 .........................................................................................38 5.4 Regression Diagnostics ..........................................................................................42 5.5 Robustness Checks and Alternative Model Specifications ....................................42 5.6 Discussion and Limitations ....................................................................................44 6. Conclusion ..................................................................................................................47 7. List of References.......................................................................................................49 8. Appendix ....................................................................................................................56 1. Introduction In the early 2000s, Ukraine's young democracy struggled with authoritarian tendencies and a concentration of power in the hands of the executive. In 2004, allegations of widespread electoral fraud and voter intimidation during the presidential elections led civil society organizations (CSOs) and opposition parties to join forces and launch a non- violent resistance campaign against the winner, Viktor Yanukovych. Protests erupted across Ukraine demanding fair elections and democratic reforms. The campaign gained momentum through mass rallies, civil disobedience and ultimately the peaceful occupation of Maidan Square in Kyiv. Supported by international actors such as the European Union and the United States, the campaign finally achieved its goals when the Ukrainian Supreme Court annulled the election results and called for new elections, leading to a change of government and subsequent democratic reforms (Karatnycky, 2005). Despite such achievements, democracies are in danger worldwide (Angiolillo et al., 2024; Mounk, 2018; Runciman, 2018). Yet, autocratization is not irreversible (Gamboa, 2022; Tomini et al., 2023; van Lit et al., 2023). Scholars point out that around 70% of all episodes of autocratization in the last 30 years have been stopped and turned into re- democratization (Nord et al., 2024b, pp. 3-4). While there is a fruitful and enriching debate about how democracies emerge and die (Coppedge et al., 2022; Huntington, 1991; Levitsky, 2018), the literature overlooks possible mechanisms on how democracies can survive episodes of autocratization. Non-violent resistance has proven to be an effective means of democratization (Bayer et al., 2016; Bethke & Pinckney, 2021; Chenoweth & Stephan, 2011) and sometimes also of halting autocratizing incumbents (Bernhard et al., 2020; Laebens & Lührmann, 2021; Rakner, 2021). However, previous literature underestimates the role of non-violent resistance as a tool for successful democratic resilience. This calls for a deeper investigation and leads to the following research question: What is the role of non-violent resistance campaigns for episodes of successful democratic resilience? For this thesis, I aim to investigate the role of one of the cornerstones of democracy: civil society. Civil society, a societal sphere between the state and market, is commonly considered central for regime stability by enhancing political participation, creating trust 1 in societies, and educating democratic values and norms (Diamond, 1994; Putnam, 2000; Tocqueville, 1969 (1835)). I argue that CSOs are one of the main actors for democratic resilience as they have the organizational capacity to mobilize citizens and initiate non- violent resistance against autocratizing incumbents. To operationalize episodes of successful democratic resilience, I use a new variable capturing a two-directional regime transformation that “entails an episode of autocratization closely followed by an episode of democratization, and that the two are parts of one process” (Nord et al., 2024b, p. 2). This operationalization allows to expand the common understanding of democratic resilience which often only focuses on the halt of democratic breakdowns, but not on restoring democracy. Furthermore, I use the NAVCO dataset to identify episodes of non-violent resistance initiated by CSOs (Chenoweth & Shay, 2020a). I merge the ERT (Maerz et al., 2023), V-Dem (Coppedge et al., 2024b), and NAVCO (Chenoweth & Shay, 2020a) datasets to create a novel panel dataset spanning between 1900 and 2020 and featuring 20,272 observations across 127 countries. I implement logistic regression and I find that one episode of non-violent resistance increases the probability of successful democratic resilience by 14%, while two such episodes increase it to 29.7%. During the third wave of autocratization, the probabilities are even higher: a single episode increases the probability to 23.8%, and two episodes increases it to an impressive 53.3%. I also find that non-violent resistance prior to the onset of autocratization significantly contribute to successful democratic resilience. When non- violent resistance occurs one year before the start of an autocratization episode, the probability of achieving democratic resilience is a remarkable 50.8%. The influence gradually diminishes over time, but the “knowledge” of previous episodes of non-violent resistance among CSOs persists for decades and never disappears completely. These insights have four scholarly and societal implications. First, non-violent resistance is effective and regularly leads to regime change and democratization. Second, by operationalizing democratic resilience as a two-stage process which includes both, the halt of autocratization and the restoration of democracy, the thesis extends the common understanding of this concept within academia. Third, the study offers hope in the face of current global crises of democracy, showing that episodes of autocratization can be 2 reversed when CSOs work together and coordinate campaigns of non-violent resistance. Fourth, I show that previous engagement in non-violent resistance helps to foster a resilience network which is ready to react once autocratization starts. Hence, these findings can be helpful for democracy aid to promote civic education, grassroots mobilization, and the cultivation of democratic norms and practices at all levels of society. The rest of the thesis is divided into five sections. In the second section, I present a literature review on autocratization, democratic resilience and the role of civil society organizations in regime stability and change. Next, I outline the origins of political mobilization and strategies of non-violent resistance. I then present mechanisms that can lead to political change through non-violent means. The literature review concludes with a discussion of the challenges faced by social movements. On this basis, I develop my theoretical model and propose hypotheses in the third section. This is followed by section four in which I outline the research design. Section five presents the empirical results, which are divided into three parts: analysis of the entire period 1900 to 2020, analysis of the period 1994 to 2020, and analysis of the period 1900 to 1993. This allows me to gain a deeper insight into the relationship between non-violent resistance and democratic resilience during the third wave of autocratization, and to test whether the assumptions hold true in the more distant past. The results are discussed and validated through several robustness checks and alternative model specifications. In the final section six, I summarize the main findings and conclude with suggestions for further research. 3 2.0 Literature Review 2.1 Autocratization and Democratic Resilience The world is facing a third wave of autocratization, which refers to “any move away from [full] democracy” (Lührmann & Lindberg, 2019, p. 1099), leading to a global level of democracy equivalent to that of 1985 (Nord et al., 2024a). In contrast to previous autocratization waves that were marked by sudden regime changes such as military coups, the current one is characterized by a gradual and slow decline in key democratic qualities (Papada et al., 2023). The core actors are executive incumbents who successively weaken accountability mechanisms of civil society, courts, elections, media, and parliament (Sato et al., 2022). Episodes of autocratization happen in various regime types, ranging from liberal democracies to electoral autocracies. While a gradual and slow process is often harder to detect for observers, it also provides more time and opportunities for domestic resistance (Laebens & Lührmann, 2021). As a result, research has shifted in the last years to democratic resilience, which means the ability to withstand autocratization, or if it has already taken place, to stop it, and eventually recover from these challenges (Merkel & Lührmann, 2021). Previous research has already tried to answer what makes democracies resilient by focusing on structural factors, for example economic aspects (Boix, 2003) or the composition of the population (Beissinger, 2008). Other scholars have looked at the institutional design of democracies, i.e. whether parliamentary or presidential forms of governments are more resilient (Linz, 1990). Although some structural factors and institutional designs might support democratic resilience, the current wave shows that autocratization occurs in various regime types and is not restricted to any particular setting or structure. Therefore, recent scholars have adopted a more actor-centered approach that focuses on the actions of democratic actors against autocratizing incumbents. Laebens and Lührmann (2021) started the discussion by exploring the role of different accountability actors in halting democratic breakdowns, which refers to the collapse of democracy (Linz, 1978). They see civil society as a driving factor that can push other accountability actors, such as political parties or courts, to prevent or stop any move by the incumbent to undermine democracy. Gamboa (2022) emphasizes the importance of the right strategy choice for elites that oppose autocratization in the political arena. She 4 warns against extra-institutional methods, for instance strikes, boycotts or coups, as they are a high-risk gamble and if they fail, they can diminish the legitimacy of opposition. Tomini et al. (2023) further enrich the discourse by presenting a framework for democratic resilience that distinguishes between institutional, political, or social actors. They outline that the context of autocratization matters, and democratic resilience should be tailored to it. While institutional actors can resist autocratization in liberal democracies, political opposition is a key actor for resilience in electoral democracies. When electoral autocracies deepen their grip on power, societal actors become crucial for resilience through public contestation. van Lit et al. (2023) argue that elites can first and foremost stop autocratization through acts of non-cooperation, for example within the parliament, the courts, or bureaucratic institutions. But when elites fail to do so, due to repression, co-optation, or self-interest, social actors come into play as they can mobilize and exert pressure on the regime. In sum, scholars have so far predominantly focused on the prevention of autocratization or the halt of democratic breakdowns. In this thesis, I aim to expand the understanding of democratic resilience by taking the restoration of democratic institutions and freedoms into account after episodes of autocratization. This allows to understand how regimes can not only survive autocratization, but also how they can recover from these challenges. Hence, I define successful democratic resilience as the stop of autocratization and the restoration of democracy to previous levels. Democratic resilience can indeed be successful. According to a quantitative study by Nord et al. (2024b) almost half of all autocratization episodes between 1900 and 2022 have turned into re-democratization, frequently leading to the restoration or enhancement of democracy levels. Beside this notable exception, most studies mentioned above are based on theory-building case studies. In contrast to the studies of autocratization, the understanding of (successful) democratic resilience is still underdeveloped and needs more attention. This thesis aims to address this gap by investigating the role and effect of one of the resilience actors: civil society. 2.2 Civil Society Organizations (CSOs) and Regime Stability CSOs are a central actor for democratic resilience. Civil society, as Edwards (2011, p. 3) puts it, is “a confusing and contested concept because so many different definitions and 5 understandings exist.” In general, the concept describes a wide range of voluntary associations, ranging from small neighborhood unions to large-scale social movements, that incorporate two main aspects: voluntariness and autonomy from the state and the market (Della Porta, 2020). While CSOs are not always peaceful, the focus of this thesis is exclusively on non-violent civil society. Many scholars have attributed an important role to civil society in the breakdown of authoritarian regimes in the last 50 years, from the Carnation Revolution in Portugal in 1974 to the Arab Spring in 2011 (Beissinger, 2022; Della Porta, 2016; Linz & Stepan, 1996). By initiating protest from below, CSOs were able to mobilize citizens to take collective action against autocratic regimes. Their ongoing efforts put sustained pressure on political elites, challenging the legitimacy of their claims to power and withdrawing the support of key groups. The main contribution of CSOs are their organizational capacities that attempt to resolve collective action problems (Bernhard, 2020). CSOs are also important in maintaining high levels of mobilization, which forced many autocratic leaders to step down, often leading to a regime breakdown or collapse and subsequent democratization (Pinckney et al., 2022). One prominent example is the breakdown of the Polish communist state in the late 1980s, which was driven by the Solidarity Movement. Started as an independent trade union, the organization soon mobilized millions of citizens, organizing strikes, protests, and other forms of civil resistance to express their frustration about domestic grievances. With the support of the Catholic Church, intellectuals and Western democracies, the movement increasingly challenged the legitimacy of the communist regime, thereby significantly contributing to the democratic transition of Poland in 1989 (Ekiert, 1999). Other scholars have pointed at the ability of civil society to strengthen democratic culture and institutions (Diamond, 1994). Tocqueville famously described CSOs as “free schools of democracy” (Tocqueville, 1969 (1835), p. 522) where democratic norms are taught, defended, and contested. This has been further developed by Putnam (2000) who emphasizes the role of social capital, which is just another term for associational activity in this context. The underlying argument here is that participation in civic associations creates mutual trust between citizens and therefore helps to foster consensus and integration in societies (Siisiäinen, 2003). In Poland, for instance, the Solidarity Movement did not only coordinate protest, but also offered democratic education by 6 providing space for dissent outside of the existing institutional channels. Among others, they published underground newspapers, promoted grassroots initiatives, or held workshops and seminars to inform people about political topics and their rights (Ash, 1983). While civil society promotes democratic education, it also strengthens democracy by overseeing state power through alternative forms of accountability, in particular between elections (Bernhard et al., 2020). These alternative forms include, inter alia, public pressures like demonstrations or civil disobedience (McAdam & Tarrow, 2010), monitoring and overseeing projects (Smulovitz & Peruzzotti, 2000), attracting international attention and support (Keck & Sikkink, 1998), and initiating petitions and referenda (Altman, 2019). Not all scholars, however, agree on the democratic power of civil society. On the contrary, Berman (1997) and Riley (2010) show that civil society contributed to the rise of fascist authoritarian regimes in Italy, Romania, Spain, and the Weimar Republic in Germany. They argue that highly mobilized societies experiencing enormous grievances are more prone to support anti-democratic actors. More recently, Lorch (2021) argues that in weakly institutionalized democracies, such as Bangladesh, the Philippines, and Thailand, CSOs are likely to be captured by political elites and may therefore contribute to democratic erosion. Thus, civil society can also become a dangerous space for democracy. As we delve deeper into understanding the mechanisms influencing regime stability, it becomes clear that civil society can play an important role during episodes of autocratization. What follows is a critical discussion of recent studies that have started to analyze the role of civil society for democratic resilience. Bernhard et al. (2020) demonstrate the central role of an active civil society in strengthening democratic resilience. They highlight how mobilized citizens effectively hold political elites accountable through the threat of social sanctions, whether by informal means such as protest or formal channels such as elections. This ability to impose sanctions puts pressure on accountability actors to uphold democratic principles and raises the political costs of autocratization for incumbents. The study is an important contribution to the puzzle of resilience but does not cover most recent developments that happened after 2010. Another critic point is that the study uses a binary classification for 7 democratic breakdowns which may not capture the complex dynamics of democratic resilience. Autocratization does not always end in a regime transition, therefore the findings may be restricted due to the operationalization. Rakner (2021) uses a comparative study on Malawi and Zambia to argue that previous, non-violent experiences of pro-democracy movements in the 1990s help contemporary civil society to mobilize and act when democracy is under attack. The author’s argument is based on social movement theory, which suggests that protest often relies on institutionalized CSOs with earlier experiences of contention, such as churches or labor unions (Chalcraft, 2016; Mueller, 2018). Periodic elections then provide a decisive window of opportunity for civil society to act against an autocratizing incumbent. By taking civil resistance to the streets during elections campaigns, CSOs can mobilize voters to express their dissatisfaction at the ballot box (Rakner, 2021). The author further emphasizes the importance of the autonomy of civil society from political parties. While civil society has remained autonomous in Malawi and stopped autocratic attempts in 2003, 2011, and 2019, Zambia’s key civil society actors were co-opted into government, which opened the way for substantial autocratization under the Patriotic Front between 2011 and 2021. While these insights are further helping to understand democratic resilience, the study is constrained due to its small sample size. Other recent contributions also provide valuable insights but are not directly connected to the study of democratic resilience. Bernhard et al. (2017) stress that civil society’s strength during critical junctures, such as revolutions, matters more for the outcome of the regime than its strength before. In contrast, Grahn and Lührmann (2021) exemplify the importance of pro-democracy movements prior to independence for the democracy level of countries with colonial past. By analyzing the effects of different kinds of CSOs on democratization, Pinckney et al. (2022) show that the nature of CSOs are crucial. Their study on resistance movements in Africa from 1990 to 2015 illustrates that so-called quotidian CSOs (QCSOs), which are voluntary associations meeting on a regular basis and that do not compete for political power, are the key driver for regime change. In particular, they highlight the role of three QCSOs - trade unions, professional- and religious organizations - which are expected to have the most democratic and durable networks to achieve change. Their democratic power is even stronger when the organizations are older and not connected to political parties. Notably, these results are 8 restricted to the African context. Finally, Hellmeier and Bernhard (2023) analyze the effect of pro-democracy and pro-autocracy mobilization on democracy levels, and they do not find statistical evidence that mass mobilization of civil society can halt democratic breakdowns. In sum, there is a large body of research that supports the idea of a positive relationship between civil society and democracy. There is also contested and ongoing discussion whether the nature or scale of civil society matter more for democracy. What is, however, not yet systematically investigated, is the interplay of civil society and democratic resilience (Lorch, 2021; Rakner, 2021). This thesis focuses on the ability of CSOs to mobilize citizens, forming a campaign that aims to stop autocratization and reinstates democracy through a diverse set of non-violent, contentious actions. The following parts explain 1) how mobilization originates and when it evolves to a movement, 2) which strategies the movements can apply, 3) how movements can achieve political change, and 4) what challenges movements face in doing so. 2.3 Origins of Political Mobilization Mobilization can be understood as the culmination of efforts leading to collective political action (Opp, 2019). It can either emerge from citizens, known as bottom-up mobilization, or be initiated by political elites, known as top-down mobilization. Drawing from social movement theory, we know that there are different explanations about how mobilization processes originate: the political opportunity approach, the resource mobilization perspective, and the theory of collective action. The key argument about the political opportunity approach is straightforward: open political structures lead to more protest, while restricted political structures lead to less protest (Martin, 2015). Hence, political environments determine when (and if) mobilization occurs (Eisinger, 1973). The resource mobilization perspective argues, as the name suggests, that resources are the main determinant for mobilization: the more resources a movement can gather, the more likely it becomes to achieve its goals (Edwards & Gillham, 2013). Resources can be anything what matters for mobilization, 9 for example time, number of supporters, or money. Both approaches have been criticized for their narrow focus on macro factors (Opp, 2019). In contrast, the theory of collective action takes micro factors into account as it highlights the role of individual behavior as the central element to explain mobilization. It seeks to answer why people contribute to a public good when their individual role has no significant effect on the outcome (Ostrom, 2009). According to Olson Jr (1971), the contribution of a single individual to a common good is explained by selective incentives, which refer to benefits that individuals receive only if they contribute, along with costs that arise if they do not contribute. These can be for example the integration in protest- encouraging social networks or a sense of duty to participate (Opp, 2019). Some related scholars note that a relatively small, but very engaged critical mass is necessary to start mobilization processes before less active citizens can join (Marwell & Oliver, 1993). Another line of research combines features of contention, politics, and collective action. It mainly seeks to explain the mechanisms behind episodes of contention, such as revolutions, strikes, or social movements (Tarrow, 2015). The concept does not focus on a particular form of contention as it believes that all of them are created in a similar way. A core feature of the concept is the political opportunity approach which is used to describe the onset of political mobilization. Once in motion, contention works through three main mechanisms: brokerage, diffusion, and coordinated action. Brokerage refers to a process that connects people with other people that have not met previously. Diffusion means that one particular form of contention, for example sit-ins, spread across multiple actors and locations. And coordinated action occurs when two or more people work together by communicating and making similar demands for the same thing at the same time (Tilly & Tarrow, 2015). Whether mobilization is successful or not, depends on the so-called scale shift. A scale shift happens when contentious actions increase in their size, impact, and reach (Soule, 2013). Not all episodes of contention manage to do that because “it takes strong organization, determined leadership, or the onset of new opportunities and threats for ordinary people to overcome their collective action problem” (Tilly & Tarrow, 2015, p. 121). 10 CSOs provide great structural pre-conditions for political mobilization. According to the civic voluntarism model, people become politically active when other people ask them to do so (Verba et al., 1995). This is more likely to happen in civic organizations because there information are exchanged and discussed, democratic skills developed, and citizens simply meet other people, thereby expanding their social network and increasing the chances of brokerage (Dalton, 2017). A number of scholars emphasize the importance of social networks in triggering and maintaining collective actions (Diani, 2015; Diani & McAdam, 2003; Opp & Gern, 1993). Political mobilization can sometimes develop into a social movement if it succeeds. A social movement is a consistent effort to make political claims by repeatedly presenting them through different actions, supported by various organizations, networks, and alliances (Tilly & Tarrow, 2015, p. 145). Tilly and Tarrow (2015) further differentiate between the bases and campaigns of social movements. The former refers to everyone that is part of the movement or supports them, whereas the latter refers to the movement’s actions. Social movements are expected to have more resources, a wider social network, and thus more opportunities to achieve its goals. Eventually, every mobilization process ends with demobilization. This happens when the goal behind the onset of mobilization has been achieved, for instance the passing or the prevention of a law. Mobilization can also end when main actors are co-opted into political parties. In the worst case, mobilization ends with repression. 2.4 Strategies of Non-Violent Resistance Movements can choose between a wide range of contentious actions, something that Tilly and Tarrow (2015) are referring to as repertoires. Repertoires can be violent or non- violent and follow a relational dynamic which means that they are shaped by the interactions and responses of both the movement and the authorities or opponents they confront (Alimi, 2015). Some movements aim to achieve political change through the use of a diverse set of violence or the threat of doing so (Della Porta, 1995). Movements use violence because they either see it as the most effective mean to achieve change or as a consequence of continued escalation between the movement and its adversaries (McAdam, 1983). 11 Movement campaigns can be successful without using violence. Repertoires are non- violent when the people involved in the movement are unarmed and refrain from the use or threat of violence (Chenoweth & Lewis, 2013, p. 418). In his classic work, Sharp (1973) identifies 198 different kinds of non-violent repertoires, such as boycotts, hunger strikes, public speeches, sit-downs, and social disobedience. Some of these forms have a more moderate character, other aims to explicitly disrupt the political process. Chenoweth and Stephan (2011) show in a large-N study that non-violent resistance campaigns are twice as effective as their violent counterpart in achieving their movement’s goals. They argue that the main reason for this is that more people participate in non-violent movements, as the moral and physical barriers of participation are lower than for violent movements. However, quantity is not all that matters for a movement’s success. The composition of the movement is also important because more diversity among the participants means that the movement represents common demands that are supported by a large part of the population, thus increasing their legitimacy (Chenoweth & Stephan, 2011). It then becomes more difficult for opponents such as the regime to stop the movement. Another important aspect is that the movement can adapt and change their repertories in response to their opponent’s reaction (Tilly & Tarrow, 2015). This enables movements to overcome challenges and sustain mobilization. Finally, the cohesion of the movement is crucial to effectively coordinate and organize collective action. This is easier to achieve when sharing a collective identity (Polletta & Jasper, 2001). 2.5 Mechanisms of Political Change through Non-Violent Means Chenoweth and Stephan (2011) outline different ways how non-violent movements achieve political change. Firstly, movements can withdraw support of important groups for the regime through their continued actions against them – a process that scholars refer to as leveraging (Schock, 2005, p. 142). Sustained resistance poses political, economic, and military costs for the regime which sooner or later results in serious problems for the country. This, in turn, can gradually undermine public support for the regime (DeNardo, 2014). Ongoing resistance also draws international attention to domestic problems, potentially leading to sanctions or the stop of diplomatic support by trade partners and 12 international organizations, thereby further intensifying leverage processes (Marinov, 2005). Secondly, diminished support may also lead to loyalty shifts which means that political elites in important institutions may no longer back the regime (Chenoweth & Stephan, 2011, p. 46). One result could be for example that bureaucratic superiors refuse to follow orders from the government. Sooner or later, the regime will lose its ability to function. Thirdly, the defection of security forces can lead to the downfall of the regime because it significantly weakens the regime’s ability to maintain control over the monopoly on the use of force. Defections happen when security forces join the movement or resists to act against them. Thus, the role of security forces, such as the army or the police, are crucial for a movement’s success as well (Anisin, 2020). Serbia serves as a great example to show how civil resistance achieves political change. In 1998, the student group Otpor started to oppose the regime after they curtailed the autonomy of the University in Belgrade. Two years later, they had become a major movement that had spread across the country united by the common goal of removing the autocrat Slobodan Milošević from office. During this time, they used various forms of civil resistance to gather support, such as creative street theatre, grassroots initiatives, and workshops. After Milošević did not concede his election defeat in 2000, Otpor was ready to initiate a series of mass civil resistance in alliance with the political opposition, including a general strike and the peaceful occupation of the capital city. They received much support, for example from the orthodox church or Western democracies. The pressure on the regime increased significantly, leading to the defection of security forces. This made it clear that large sections of the population were against a continuation of autocratic rule. Ultimately, the costs to stay in power were higher than conceding for Milošević, leading to his resignation which opened the way for democratization (Kurtz, 2010). Another line of research emphasizes that elections are a decisive window of opportunity in which the demands of the movement can be realised (McAdam & Tarrow, 2010). While movements can put pressure on those in power to hold free and fair elections and mobilize citizens to go voting, they can also compete in elections either by joining an election coalition or by standing as a candidate (Heaney, 2022). 13 From a public choice perspective, the regime will only give in to the demands of the movement if the costs of alternative courses of action are greater (Mueller, 2003). Since the autocrat’s primary goal is to remain in power, they will explore all possibilities to stop or weaken the movement. For meaningful political change to occur, the costs to the incumbent must be as high as possible, including all the mechanisms described above. In sum, increasing pressure on the regime from inside and outside could lead to minimalist goals like concessions to the movement or maximalist goals like the resignation of the incumbent. Research has shown that concessions in authoritarian regimes often fuel further protest because citizens realize that change is possible and if they continue to mobilize even more goals may be achieved (Leuschner & Hellmeier, 2024). In other words: concessions often create new political opportunities which stimulates even more mobilization and resistance. However, the movement’s success depends on a variety of factors and is not easy to achieve as the next part shows by presenting challenges for movements. 2.6 Challenges facing Social Movements Depending on the regime context, the movement’s ability to push for political change can be restricted by certain factors. The most fundamental factor is repression, which is defined as the threat or use of violence against a person or group that opposes the regime (Davenport, 2007). Repression is one of the stabilizing pillars for the survival of autocracies and is commonly used to suppress dissent (Gerschewski, 2013). To understand repression in its depth, two distinctions are necessary. First, scholars differentiate between overt and covert repression acts. The former are visible acts of repression, for instance when the police forcefully stop a demonstration (Della Porta & Reiter, 1998), whereas the latter are more subtle forms that aim to prevent contentious actions in the first place, for example the surveillance of dissidents (Cunningham, 2004). Second, scholars further distinguish between coercion and channeling (Earl, 2022). While coercion involves a variety of repressive acts, ranging from intimidation to murder (Davenport, 2007), channeling intends to gain control over resistance by allowing certain forms of contention but sanctioning others, for example by withdrawing state funds (McCarthy et al., 1991). 14 The use of repression as an act to restrict dissent by the state or related actors lead to different outcomes. Some scholars point to the weakening effect on political participation as repression raises the personal costs for individuals to take part in contentious activities (Honari, 2018). Similarly, repression raises the costs for the allocation of resources which weakens the movement’s ability to mobilize (McCarthy & Zald, 1977). Other scholars emphasize that the use of repression leads to a tactical shift because movements change their repertoires in response to the state’s coercive actions (Lichbach, 1987). However, repression can also cause feelings of public outrage when citizens perceive the state’s actions as unjust. If public anger becomes widely publicized, repression can trigger a backfiring process that strengthens support for the movement and increases its resilience, ultimately leading to broader participation and solidarity (Hess & Martin, 2006). Autocratic regimes also use co-optation, which refers to the inclusion of important actors of the movement in the political elite, to use the power of these actors for their own favor (Bertocchi & Spagat, 2001). Co-optation is another of the stabilizing pillars for the survival of autocracies and is used in this context to limit the power of movements by offering benefits in exchange for cooperation and loyalty (Gandhi & Przeworski, 2006; Gerschewski, 2013). Co-optation weakens movements as it undermines the credibility of their leaders and opens a channel for the regime to influence the movement’s behavior and goals from within. Another obstacle are countermovements that mobilize simultaneously but share contrary demands (Tarrow, 2015). Countermovements usually arise under three conditions: first, when one movement achieves some success, second, when the interests of some social groups become threatened by the movements goals, and third, when political elites support resistance to the movement (Meyer & Staggenborg, 1996, p. 1635). The stronger the countermovement, the more difficult it becomes for movements to achieve political change, because they have to fight on two fronts: against the regime and against the countermovement. Internal disputes can lead to the radicalization of one part of the movement that demands other tactics than the moderate part. Scholars refer to this phenomenon as the radical flank which be seen as a further obstacle (Haines, 2013). The effect of radical flanks on movements are contested among scholars. Some believe that radicals weaken the 15 legitimacy of the movement and its ability to mobilize (Hoffman & Bertels, 2012). Others note a positive effect as radical acts can attract greater public attention and enhance the reputation of the moderate part of the movement (Ellefsen, 2018). Finally, recent literature has outlined that the third wave of autocratization is accompanied by a rise in misinformation and polarization (Boese et al., 2022; McCoy & Somer, 2019). Misinformation can be used by the autocratic regime to spread fake news and myths about a movement to diminish its potential to mobilize. Polarization on the other hand can either help the movement to mobilize or fueling countermovements as people are divided into two camps that oppose each other. 16 3. Theoretical Model and Hypotheses As outlined in the literature review, episodes of autocratization are associated with multiple attacks on accountability mechanisms. These attacks include, among others, restrictions on the freedom of speech and assembly, limitations on the power of the judiciary to exert control over the government, censorships on independent media outlets, and crackdown of civil society. Over time, incumbents progressively undermine accountability mechanisms, thus concentrating power on one person. However, CSOs have repeatedly demonstrated their ability to counter these attacks, even under very restrictive conditions. I argue that attacks on the accountability mechanisms by incumbents can trigger a response by the CSOs. While minor attacks might not find much public attention, greater attacks, such as constitutional changes, the extension of term limits, or a rigged election, may increase discontent among the population. This can result in political opportunities for CSOs to mobilize. Sometimes the same process is triggered by scandals that uncover violations of the democratic order. To realize democratic resilience, the mobilization of civil society must spread across several actors, united by the common goal of defending democracy against an autocratizing incumbent. For that purpose, CSOs can rely on their personal networks. CSOs are important actors to collect resources and help to overcome collective action problems by providing incentives for citizens to participate. Beside their coordination and organization efforts, CSOs also have the potential to empower citizens by educating them through leaflets, seminars, workshops, and more. One effective tool to oppose the autocrat is non-violent resistance which can take on various forms of collective action. Non-violent resistance has at least two advantages over violent resistance. First, a commitment to non-violent means raises the profile of the campaign, thereby attracting more participants and supporters. The advantage in the size of participation strengthens the legitimacy of the campaign and increases its political impact. Second, the government faces a substantial challenge in repressing non-violent campaigns because it risks a backfiring process as people might perceive the state’s action as unjust and more people join the resistance in reaction. Non-violent resistance thus emerges not only as morally superior, but also as strategically advantageous which is capable of challenging authoritarian rule (Stephan & Chenoweth, 2008). 17 It is important that the resistance campaign continues and involves as many people as possible. Persistent and large-scale resistance imposes political and economic costs on the regime, thereby gradually eroding the support of central groups for the incumbent. It also attracts international attention which can lead to sanctions or the stop of diplomatic support by countries or organizations that provide democratic assistance. The less domestic and international support an incumbent has, the more problematic it becomes for them to legitimate the claim on power. This can lead to loyalty shifts of important political elites or the defection of security forces, factors that significantly intensify the pressure on the incumbent. As long as it costs less than conceding, the incumbent will try to fight the campaign and those involved in it. Common tactics are repression and co-optation. Campaigns might also face other obstacles such as countermovements, the radical flank effect, polarization, or misinformation. It is crucial for the campaign to withstand these setbacks and continue to mobilize. The more and diverse the people involved in the campaign, the higher the likelihood that the campaign will survive the attacks of the incumbent. In answering how non-violent resistance campaigns can achieve democratic resilience, I present three possible scenarios. First, the domestic and international pressure increases to such a high degree that the incumbent loses its legitimacy and capability to govern. This creates further political opportunities as citizens realize that change might be possible. As a result, citizens incessantly participate in non-violent resistance until maximalist goals are achieved, such as new elections or a revolution that is followed by a regime change. New elected political elites can then implement democratic reforms. For instance, civil society in North Macedonia launched mass mobilization against the autocratic incumbent Nikola Gruevski after a massive wiretapping scandal erupted in 2015 (Marusic, 2015). Although the space for civil society was restricted by the government, CSOs were able to form a movement in alliance with the main opposition party. The movement favoured non-violent marches that always ended at the main government building and citizens could hold speeches. Other non-violent actions were initiated over time, such as releasing colourful balloons or painting monuments, to draw attention to the broad and diverse participation of the population (Stojadinovic, 2019). A successful boycott of the national elections and international pressure eventually led to a democratic transfer of power which is also known as the Colourful Revolution. The new 18 government then implemented several democratic reforms and reinstated accountability mechanisms (Papada et al., 2023, p. 29). In a second scenario, the public pressure and decreasing legitimacy for the incumbent stimulates accountability actors to stop the incumbent from their autocratic aspirations. This can be achieved by the use of legal mechanisms within the democratic institutions, such as impeachment proceedings initiated by the political opposition or rulings by the constitutional court. However, in this scenario, accountability mechanisms must still be in place to accomplish democratic resilience. South Korea’s democratic turnaround in 2017 illustrates how accountability actors reinstate democracy. After President Park Geun-hye attacked some of country’s liberal democratic principles and was alleged to be engaged in corrupt practices, civil society was ready to respond. Important labour groups and other CSOs initiated mass mobilization and took millions of citizens to protest on the streets which endured for six months (Cho & Hwang, 2021). Every Saturday night, citizens went to the streets with candles in their hands, a well-known Korean symbol from previous contentious episodes during authoritarian rule, to express their displeasure with the government. A few months later, the approval rate of Park dropped to less than five percent (Jung, 2023, p. 778). While opposition parties were hesitant to join the movement at first, public pressure pushed them to start impeachment proceedings. The parliament finally removed Park from office and the Constitutional Court unanimously upheld the judgement (Shin & Moon, 2017). Third, non-violent resistance campaigns often take place around election times. In this case, supporters of the campaign can vote the incumbent out of office if movements align with the political opposition or standing as a candidate by themselves. Opposition parties’ elites can mobilize citizens to go voting by drawing attention to grievances and the possibility of political change. Slovenia serves as a great example to illustrate the third scenario. After Prime Minister Janez Janša took the country on a path towards autocratization by curtailing press freedoms and disregarding the judiciary in 2020, more than 100 different CSOs initiated mobilizations for democracy (Narsee, 2022). Every Friday, people went on the streets to protest against Janša and his policies. Because assembly rights were restricted due to the COVID-Pandemic, people started bicycle rallies to comply with the rules (Monitor, 2022, p. 7). The rallies persisted until the national parliamentary elections in 2022, which the newly founded Freedom Movement 19 won and thereby defeated Janša and his party. In the run-up to the elections, CSOs organized joint activities to inform people about the current political situation and to mobilize them to turn out and vote. This resulted in a large voter turnout of around 70%, a 18% increase compared to the previous elections in 2018 (Andrinek, 2022). In summary, I argue that CSOs have the organizational power and capacity to resist autocratization by initiating and coordinating non-violent resistance campaigns. While not every campaign is successful in stopping and reversing episodes of autocratization, the three outlined scenarios have shown how CSOs can be effective in achieving successful democratic resilience. This leads to the first hypothesis: Hypothesis One: “Non-violent resistance campaigns significantly contribute to successful democratic resilience during episodes of autocratization.” Furthermore, I argue that experiences of non-violent resistance in the past helps civil society to act against an incumbent if they attack accountability mechanisms. On the one hand, past experiences build a shared knowledge of resistance strategies from which CSOs can learn. While past strategic mistakes can be avoided, successful strategies can be reused or adapted. Moreover, CSOs with past experiences have a better understanding of how to overcome challenges such as repression, co-optation or misinformation and are therefore less likely to be defeated by the government. On the other hand, previous engagement in non-violent activities builds trust and creates a culture of democratic resilience. As people remember that they aligned together before in alliance with CSOs, the barrier to participate in non-violent resistance becomes lower for individuals. Finally, CSOs with earlier experiences are better connected among each other domestically, but also internationally. This provides them with more resources to mobilize when democracy is under attack. At this point, we can refer back to Rakner (2021) who points at the role of CSOs in Malawi to prevent several autocratic take-overs. In the 1990s, bishops of the catholic church demanded political and economic reforms for the country, which was supported by student marches and worker’s strikes. After the successful democratic transition in 1994, the new government tried to change the constitution to allow it to stay in power for a third term. Meanwhile, CSOs had set up a joint network and played a key role in stopping the parliament to change the constitution by mobilizing citizens and motivating 20 them to participate in non-violent resistance. As Dionne (2024, p. 130) emphasizes, “among civil society actors, Malawi’s faith leaders stand out as trusted voices and leaders” that can encourage the population to take to the streets and peacefully fight for their democratic rights. In 2011, President Bingu wa Mutharika tried to hand over power to his brother after two consecutive terms, which led again to the mobilization of CSOs that coordinated nation-wide non-violent resistance. While being confronted with severe repression by the government, the resistance campaign was eventually successful as the highest court ruled the elected vice-president as president (VonDoepp, 2020). However, in the 2014 elections, Mutharika’s brother came into office and restricted the space for CSOs. Nevertheless, CSOs once again successfully mobilized in 2019 after allegations of fraud elections. Non-violent resistance led to the annulation of the election results by the Highest Court. The Malawi case demonstrates how past experiences of contention empower civil society to mobilize when democracy is under attack by providing CSOs with important knowledge, establishing a resilience network, and lowering the barriers for citizens to participate. This leads to the second hypothesis: Hypothesis Two: “Non-violent resistance campaigns that happen before the onset of autocratization significantly contribute to successful democratic resilience.” Figure 1 summarizes the mechanisms that can link non-violent resistance campaigns and democratic resilience. As explained, attacks on accountability mechanisms or scandals that uncover violations of the democratic order can create political opportunities for CSOs to mobilize. If several CSOs join forces, they can initiate a non-violent resistance campaign. Previous experiences of non-violent resistance help CSOs to coordinate and sustain mobilization of society. The increasing pressure on the incumbent may lead to democratic resilience as outlined in the three scenarios before. To realize democratic resilience, the support of the judiciary and the legislative is helpful because they can use institutional mechanisms, such as impeachment proceedings or rulings by the constitutional court, to translate the resistance into politically effective measures. However, the institutional power may be restricted due to the accountability attacks of the autocratizing incumbent. 21 Figure 1: Mechanisms linking non-violent resistance campaigns and successful democratic resilience 22 4.0 Research Design In order to answer the research question “What is the role of non-violent resistance campaigns for episodes of successful democratic resilience?” and to test the proposed hypotheses on a global scale, I choose a quantitative research design. My data comes from two different and highly regarded sources: the Varieties of Democracy (V-Dem) project (Coppedge et al., 2024a) and the Non-Violent and Violent Campaigns and Outcomes (NAVCO) data project (Chenoweth & Shay, 2020a). 4.1 Dependent Variable (DV) Measuring the quality of democracy is a complex process, as there are many different definitions, and it cannot be observed directly. Over the past decade, the V-Dem project has become one of the leading indices, outperforming other commonly used datasets on democracy such as PolityIV and Freedom House in terms of methodology rigor and aggregation procedures (Boese, 2019). V-Dem uses a continuous classification of regimes which is based on Dahl’s institutional prerequisites for democracy (Dahl, 1971). To generate the data, at least five independent experts answer questionnaires for each country indicator. In total, more than 4,000 experts are part of the data generation process. With the help of an established measurement model, these answers are than aggregated into point estimates for each indicator, accounting for potential biases (Pemstein et al., 2023). V-Dem updates their dataset annually, with the latest version encompassing “over 31 million data points for 202 countries from 1789 to 2023” (Nord et al., 2024a, p. 2). To operationalize the concept of successful democratic resilience, I make use of Nord et al.’s (2024b) Democratic turnarounds, which refers to a new type of regime transformation that “entails an episode of autocratization closely followed by an episode of democratization, and that the two are parts of one process” (Nord et al., 2024b, p. 2). To identify episodes of democratization and autocratization, the authors use the Episodes of Regime Transformation (ERT) dataset (Maerz et al., 2023). The operationalization of the ERT is based on changes on the level of V-Dem’s Electoral Democracy Index (EDI), measuring the quality of electoral processes within a country on a range from 0 to 1 (Coppedge et al., 2024b). Autocratization represents a decline of at least 0.01 on the EDI, whereas democratization represents an increase of at least 0.01 on the EDI. The threshold to be considered as regime transformation is a total change of at least ±0.1 within the 23 episode’s duration. The regime transformation episode is ongoing if the EDI changes once every five year and does not experience an annual reverse trend of 0.03 or greater. The last year of an episode is reached when one of the mentioned criteria is not fulfilled anymore (Maerz et al., 2023, p. 6). Nord et al. (2024b) distinguish different subtypes of turnarounds. This thesis makes use of U-Turns which are regime transformations that restore the level of democracy to that prior to the onset of autocratization (Nord et al., 2024b, p. 10). U-Turns are coded by calculating the difference of the EDI value between the final year of autocratization and the first year of democratization. If the difference is less than 0.1, it is considered a U- Turn. Figure 2 illustrates how an average U-Turn looks like. The EDI level is represented on the Y-axis, while the X-axis represents the time in years. Further on, it shows the average duration of a U-Turn, which is approximately 8 years (Nord et al., 2024b, p. 25). Figure 2: Illustration of an average U-Turn between 1900 and 2022. Source: Nord et al. (2024b, p. 24) 24 The variable “uturn” is dichotomous, indicating whether a given country has experienced a U-Turn in a given year. The whole process, from the start to the end of the turnaround, which lasts several years, is coded as a U-Turn. The data spans from 1900 to 2023, thus allowing to investigate the research question and hypotheses over time. The unit of analysis of the data set is country-year. The U-Turn dataset is yet to be published by Nord et al. (2024b), and they shared it with me on request. I merge it with the 2024 V-Dem dataset (Coppedge et al., 2024a). Figure 3 shows the frequency of U-Turns over decades. The number of U-Turns has increased over time and experienced a first massive rise in the 1940s after the end of the Second World War. In the 21st century, however, U-Turns have become much more common on average than in previous decades. This shows that democratic resilience is often successful and underlines the importance of studying its causes. Figure 3: Frequency of U-Turns over time, displayed by decades. 4.2 Independent Variable (IV) The independent variable derives from the research project NAVCO that tracks violent and non-violent mass movements with the goals of regime change, anti-occupation and secession (Chenoweth & Lewis, 2013). NAVCO uses campaigns as the main unit of analysis. To qualify as a campaign, contentious events with more than 1,000 participants must take place at least twice and be coordinated by CSOs with the same objectives. The data for non-violent resistance relies on consensus data based on various sources, such as 25 encyclopaedias, case studies and other scientific bibliography. Data for violent resistance on the other hand stems from already established datasets on armed conflict and counterinsurgency operations. The coding of campaigns is subject to a review process by a number of experts (Chenoweth & Shay, 2020b). The most recent dataset is the version 1.3 which includes all resistance campaigns between 1900 and 2019 (Chenoweth & Shay, 2020a). For every campaign, it presents information about its location, target, start and end years, and whether it has been primarily violent or non-violent. Moreover, it includes information about the success and the purpose of the campaign. In order to use both, V-Dem and NAVCO, I changed the structure of NAVCO towards country-year-campaign, so that every country has at least one observation for a single year between 1900 and 2019. This allows me to merge the two datasets. Furthermore, I created two new variables based on the existing variables which will be used as the main IVs. To test Hypothesis One, I create the variable “NVR” which counts how many episodes of non-violent resistance campaigns have occurred per year and country.1 To test Hypothesis Two, I build a stock variable that represents that accumulated knowledge of NVRs. I argue that older non-violent resistance episodes become less relevant to current mobilizations. To account for the diminishing effect, I use an annual depreciation value of 0.1 within an interval of 0 to 1. 1 means that CSOs have experienced non-violent resistance one year before and have full “knowledge” of it, whereas 0 means that the last non-violent resistance episode happened (more than) 10 years ago and CSOs do not have any meaningful “knowledge” of it anymore. The threshold of 10 years as well as the depreciation value is debatable and may be adjusted to longer time periods. The variable name is “NVR stock knowledge”. Figure 4 shows the frequency of episodes of non-violent resistance over time. Non-violent resistance was not very widespread from 1900 to 1970. In the 1980s and 1990s, non- violent resistance experienced its first upsurge. From 1980 to 2009, non-violent resistance episodes were more than twice as common than it had been in previous decades. More importantly to this study, there has been a significant rise in the last decade, indicating 1 The same is done for violent protest episodes which will be used as a control variable. The variable name is “Violent Resistance”. 26 that non-violent resistance is a highly relevant phenomenon that is clearly evident in the current wave of autocratization. This underscores the relevance of my study. Figure 4: Episodes of non-violent resistance over time, displayed by decades. 4.3 Estimation strategy As the outcome variable is binary, I can choose between a non-linear probability (probit) model or a non-linear logistic (logit) model to estimate the influence of the IVs on the DV. Both models can be used to estimate the influence of certain variables on the probability of an event occuring, in this case a U-Turn. The difference between these two models is the calculation of the probability. While probit models use the cumulative standard normal distribution function, logit models are based on logistic functions (Aldrich, 1984). Since there is no reason to believe that one model fits better than the other regarding the data of my study, I decided to choose the logit model and include the probit model as a robustness check. The logit model use the following formula to calculate the probability of observing a U- turn: 1 1 𝑃𝑟(𝑌 = 1 | 𝑋) = = 1 + 𝑒−𝑙𝑜𝑔𝑖𝑡 1 + 𝑒−(β0+ β1𝑋1+ β2𝑋2+⋯ +β𝑘𝑋𝑘) 27 In this formula, 𝑃𝑟(𝑌 = 1 | 𝑋) stands for the conditional probability of a U-Turn occuring given the values of the independent variables X1, X2, …, Xk. The logistic function 1 calculates the probability of observing a U-Turn, where β0, β 1+𝑒−(β0+ β 1, 1𝑋1+ β2𝑋2+⋯ +β𝑘𝑋𝑘) …, βk are the coefficients estimated by the model. ′𝑒′ is the base for the natural logarithm (Mehmetoglu, 2022, pp. 179-182). I include several control variables in the logit model, which are expected to influence both, the DV and the IVs. By accounting for these factors, one can more precisely estimate the influence of the IVs on the propability of observing the DV. First of all, the variable “Violent Resistance” is added to the model to check whether the kind of protest differs on the outcome. In contrast to non-violent resistance, violent resistance is not expected to lead to democratic resilience as it increases the risk of severe repression and delegitimates the movement’s goals (Chenoweth & Stephan, 2011). In return, not as many people will take part in violent resistance which decreases the power of the movement. Second, the level of democracy, measured by V-Dem’s EDI, is included in the model. The variable is called “v2x_polyarchy” and scaled from low to high (0-1) (Coppedge et al., 2024b, p. 47; Teorell et al., 2019). As described earlier, U-Turns are dependent on the EDI. However, the level of democracy also affects non-violent resistance because in well- developed democracies citizens have simply more legal options to oppose the incumbent. I expect that countries with higher democracy levels are less prone to autocratization in the first place and thus less likely to experience a U-Turn. Third, GDP per capita is included to account for economic factors which may trigger non-violent resistance when countries experience economic downturns. Economic imbalances also increase the likelihood of regime change and thus affects U-Turns (Haggard & Kaufman, 1995). I use the GDP per capita variable “gdppc” from the Maddison Project Database (Bolt & Van Zanden, 2020a). Fourth, I use two indices from the V-Dem dataset to account for the power of the judiciary and the legislative to hold the executive accountable. The former index asks “To what extent the executive respect the constitution and comply with court rulings, and to what extent is the judiciary able to act in an independent fashion?” (Coppedge et al., 2024b, p. 51), whereas the latter asks “To what extent are the legislature and government agencies […] capable of questioning, investigating, and exercising oversight over the executive?” (Coppedge et al., 2024b, p. 51). Both indices are scaled 28 from low to high (0-1). Next, the level of political polarization is added to the model as research has shown that it increases the likelihood of autocratization (Boese et al., 2022; McCoy & Somer, 2019). I also expect that higher polarization levels lead to more civil resistance as people become increasingly motivated to challenge opposing views. The variable derives from V-Dem and is called “v2cacamps”, ranging from low (0) to high (4) (Coppedge et al., 2024b, p. 232). In addition, the average years of education are added to the model. Education fosters democracy and higher education levels may help societies to withstand autocratization as they might be more informed about political developments and are more aware of their rights and tools to counter attacks on democractic rights (Barro, 1999; Glaeser et al., 2007; Lipset, 1959). The variable “e_peaveduc” is part of V- Dem’s dataset and shows the average years of education for citizens that are older than 15 years (Coppedge et al., 2024b, p. 381; Fariss et al., 2021). Lastly, the population size is added as the eight control variable. Accounting for the size of population is important to ensure that the influence of the IVs on the DV are not driven by differences in the population size across observations. The variable stems from V-Dem and is called “e_pop” (Coppedge et al., 2024b, p. 394; Fariss et al., 2021). Before running the models, I transformed some of the variables to strengthen the assumptions underlying logistic regression. To improve the linearity of the relationship between the independent and dependent variable, I use the logarithmic (log) version for the GDP per capita and the Population variable. For the Electoral Democracy Index (EDI), I use the moving average of five years to smooth fluctuations, reduce noise in the data, and to address long-term trends more precisely. Furthermore, some countries experience up to five U-Turns, whereas other countries only experience one or none. This variability in the number of U-Turns across countries can lead to within-country correlation among the observations, which may bias the standard errors. I address this issue by clustering the standard errors by countries. Additionally, I include time fixed effects to control for trends that may arise over time. In a robustness check, I also include country fixed effects to control for unobserved heterogeneity across countries. However, because U-Turns are still rare events (N=68), incorporating these fixed effects significantly reduces the sample size. This reduction occurs because STATA automatically excludes all observations from countries where no U-Turns were recorded, 29 resulting in the loss of over half of the original sample. This substantial decrease in sample size is the reason why I do not include country fixed effects in the initial models. Table 1 provides a summary of all included variables. All V-Dem variables are coded from 1900 to 2023, whereas the NAVCO variables are coded from 1900 to 2019. The GDP variable is coded from 1900 to 2018. Some variables have been renamed for clarity. Table 1: Descriptive statistics of all included variables Concept Source Observat- Mean Std. dev. Min Max ions U-Turn V-Dem 19,499 .047 .021 0 1 Non-violent NAVCO 18,629 .041 .205 0 2 resistance (NVR) NVR stock NAVCO 19,620 .009 .082 0 1 knowledge Violent resistance NAVCO 18,629 .131 .473 0 7 Electoral V-Dem 19,236 .305 .275 .006 .921 Democracy Index (moving average) GDP per capita MADDISON 13,415 8.392 1.110 5.922 11.960 (log) Judicial constraints V-Dem 19,259 .516 .296 .002 .992 Legislative V-Dem 16,681 .470 .308 .012 .99 constraints Political V-Dem 18,976 -.145 1.372 -3.806 3.934 Polarization Education V-Dem 13,059 4.993 3.560 .01 13.3 Population (log) V-Dem 15,532 6.434 1.708 1.082 11.907 30 5.0 Main results This section is divided into six parts. In the first part, I test the hypotheses for the entire period from 1900 to 2020. In the second part, I subset the time span to the period after 1993. This makes it possible to look more closely at the relationship between non-violent resistance and democratic resilience during the third wave of autocratization. In a third step, I test the hypotheses for the years 1900 to 1993, in order to analyze whether the assumptions hold true in the more distant past. I then conduct regression diagnostics and various robustness checks, as well as alternative model specifications, to test the validity and sensitivity of the results. Finally, the main findings are discussed and interpreted. 5.1 Logit Model 1900 to 2020 Table 2 shows the results for the Logit Models for the time span from 1900 to 2020. The interpretation of the coefficients from the table is not straightforward, since the logit regressions show “how much the natural logarithm of the odds for Y = 1 changes for each one-unit change in X” (Mehmetoglu, 2022, p. 178). However, the direction of the coefficients can be interpreted as well as their significance. Model 1 represents the bivariate relationship between non-violent resistance campaigns (NVR) and U-Turns. The coefficient is statistically significant at p < 0.001 and positive, indicating a positive association between NVRs and U-Turns. In Model 2, the variable “NVR stock knowledge” is added which indicates the accumulated knowledge of CSOs regarding past non-violent resistance campaigns. The relationship between this variable and the U-Turn is positive and statistically significant at p < 0.001 as well, which suggests that higher levels of NVR knowledge increase the likelihood of a U-Turn. The coefficient of the “NVR” variable remains positive and statistically significant (p < 0.001). In Model 3, I add the third NAVCO variable. We can observe that the two non-violent NAVCO variables remain statistically significant at p < 0.001 and positive associated with U-Turns. In contrast, the Violent Resistance variable is not statistically significant (p > 0.05), meaning that violent resistance episodes do not influence U-Turns by enough certainty. In Model 4, two more control variables are added: GDP per capita (log) and the Electoral Democracy Index (moving average) variable. These variables are expected to have the 31 most explanatory power among the control variables which is why we look at them separately from the other controls. The two non-violent NAVCO variables remain statistically significant (p < 0.001) and positively associated with U-Turns, and the coefficient of the Violent Resistance indicator remains statistically insignificant (p > 0.05) when controlling for other factors that could influence the relationship between the main IV and DV. Looking at the added control variables, we can see that a one-unit increase in the logged GDP per capita is associated with a decrease in the log odds of observing a U- Turn. The relationship is statistically significant at p < 0.01. This tells us that more developed economies are less likely to experience U-Turn. The EDI variable is not statistically significant (p > 0.05). In the final model (Model 5), all other control variables are included. The coefficients of the non-violent NAVCO variables remain statistically significant (p < 0.001) and have a positive direction as in the models before. Violent resistance continues to be insignificant (p > 0.05). Two of the control variables are statistically significant: the GDP per capita (log) at p < 0.05 and the Polarization variable at p < 0.001. The GDP per capita coefficient exhibits the same direction as in Model 4. As theoretically expected, a one-unit increase in the Polarization variable increases the log odds of observing a U-Turn, indicating a positive relationship. All other control variables are statistically insignificant (p > 0.05). 32 Table 2: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 2020 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.010*** 1.123*** 1.096*** 1.066*** 0.904*** (0.209) (0.217) (0.214) (0.216) (0.200) L.NVR stock knowledge 3.156*** 3.152*** 3.020*** 2.916*** (0.534) (0.537) (0.534) (0.552) L.Violent Resistance 0.199 0.104 0.0280 (0.125) (0.121) (0.113) ** * L.GDP per capita (log) -0.491 -0.658 (0.177) (0.279) L.EDI (moving average) 0.557 -0.0203 (0.611) (1.034) L.Judicial constraints 1.076 (1.213) L.Legislative constraints 0.401 (0.729) L.Polarization 0.378*** (0.105) L.Education 0.0797 (0.113) L.Population (log) 0.00357 (0.104) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.539*** 0.141 1.377 (1.019) (1.019) (1.019) (1.685) (2.180) Observations 9266 9266 9266 9266 9266 AIC 4312 4188 4189 4110 4020 BIC 5054 4937 4974 4937 4883 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 33 To investigate and illustrate the coefficients in depth, I use marginal effect plots to display the marginal effects of the main independent variables. Marginal effects indicate “how changes in a focal independent variable affect the predicted value of the outcome” (Mize et al., 2019, p. 155) Figure 5 shows the marginal effect plot for the influence of NVRs on the probability of observing a U-Turn when looking at the bivariate relationship. From Figure 5, we can observe a positive relationship between NVRs and U-Turns. The probability of observing a U-Turn is 14% when a country experiences NVR. When experiencing two NVRs, the likelihood of observing a U-Turn is 29.7%. What is also observable from Figure 4 is that without experiencing NVR, the baseline likelihood of experiencing a U-Turn is 5.8%. We can see that the width of the confidence intervals becomes greater the more NVRs occur, indicating greater uncertainty for these estimates. This might be driven by the distribution of the NVR variable because there are considerably less observations for NVR = 2 than for NVR = 1.2 Figure 5: Marginal effect plot for influence of NVRs on U-Turns with 95% confidence intervals for the time span 1900-2020 The next figure (Figure 7) shows the marginal effects for the “NVR stock knowledge” variable. As the plot before, the marginal effects are calculated only for the bivariate relationship. Figure 7 illustrates a positive relationship between the stock variable and U- Turns: the more NVR “knowledge”, the higher the probability of experiencing a U-Turn. 2 A histogram of the NVR variable can be found in the Appendix at Figure 6 34 Looking on the graph from the opposite direction (from the right to left), we can see the expected diminishing effect over time: a one-unit decrease (-0.1) of the NVR knowledge stock variable is associated with a decrease in the probability of observing a U-Turn. Regarding the confidence intervals, we see more uncertainty for the values greater than 0.5. The baseline likelihood for a U-Turn without any previous NVR knowledge is 5.8%. In comparison, when having maximum “knowledge”, the likelihood of experiencing a U- Turn is 50.8%. After 5 years, the likelihood diminishes to 20.6% which is less than half. Figure 7: Marginal effect plot for influence of NVR stock knowledge on U-Turns with 95% confidence intervals for the time span 1900-2020 5.2 Logit Model 1994 to 2020 As a next step, I run the same models but only for the period 1994 to 2020. The results can be seen in Table 3. The relationships between the two non-violent NAVCO variables and U-Turns remain statistically significant at p < 0.001 and positive across all models. Violent Resistance remains statistically insignificant (p > 0.05). 35 Table 3: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1994 to 2020 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.316*** 1.460*** 1.466*** 1.478*** 1.302*** (0.260) (0.266) (0.261) (0.249) (0.229) L.NVR stock knowledge 2.969*** 2.971*** 2.798*** 2.541*** (0.598) (0.599) (0.596) (0.614) L.Violent Resistance -0.0337 -0.133 -0.278 (0.205) (0.200) (0.196) L.GDP per capita (log) -0.618** -1.294** (0.204) (0.406) L.EDI (moving average) 0.609 0.671 (0.804) (1.491) L.Judicial constraints 1.876 (1.616) L.Legislative constraints -1.048 (1.032) L.Polarization 0.423* (0.179) L.Education 0.288 (0.165) L.Population (log) 0.239 (0.162) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -2.959*** -3.080*** -3.071*** 1.792 3.310 (0.404) (0.437) (0.434) (1.330) (2.508) Observations 3136 3136 3136 3136 3136 AIC 1900 1814 1816 1723 1631 BIC 2057 1977 1985 1905 1844 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 36 Looking closer at the main variables of interest, we can see in Figure 8 that an increase in the NVR variable from 0 to 1 lead to a probability of observing a U-Turn by 23.8% for the bivariate relationship. In comparison to Figure 5, this plot shows an increase by 9.8%. When countries experience two NVRs, the probability of observing a U-Turn is 53.3%, an increase of 23.6% in comparison to Figure 5. The baseline likelihood of observing a U-Turn without the occurrence of any NVR is 7.8%, an increase of only 2% in comparison to Figure 5. This tells us that during the third wave of autocratization, the influence of NVRs on observing U-Turns is much higher than for the full-time span. Figure 8: Marginal effect plot for influence of NVRs on U-Turns with 95% confidence intervals for the time span 1994-2020 Figure 9 shows the marginal effects of the variable “NVR stock knowledge”. The marginal effects are very similar to those in Figure 7, although the slope of this plot is slightly steeper. This indicates that while the general interpretation remains the same, the influence of “NVR stock knowledge” is slightly bigger for the period 1994-2020 than for the entire period 1900-2020. For example, having “full knowledge” (NVR stock knowledge = 1) leads to a probability of observing a U-Turn of 56.1%, an increase of 5.3% compared to Figure 7. After 5 years, the likelihood of observing a U-Turn diminishes to 25.4%, an increase of 4.8% compared to Figure 7. 37 Figure 9: Marginal effect plot for influence of “NVR stock knowledge” on U-Turns with 95% confidence intervals for the time span 1994-2020 In sum, the results for the time span 1994 to 2020 indicate that NVRs during episodes of autocratization have a much greater influence on U-Turns compared to the full-time span. The same, but less noticeable, can be said about the “NVR stock knowledge” variable. One of the reasons for these results could be that NVRs are increasing in reaction to autocratization between 1994-2020, which also leads to more “NVR knowledge”. 5.3 Logit Model 1900 to 1993 In a final step, I run the same models for the time span 1900 to 1993. From Table 4 we can see that the main relationship between NVRs and U-Turns is statistically insignificant (p > 0.05) across all models, indicating that non-violent resistance does not influence the log odds of observing a U-Turn by enough certainty for this time span. By contrast, violent resistance is statistically significant at p < 0.05 with a positive association with U- Turns in Model 3-5. 38 Table 4: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 1993 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 0.392 0.434 0.347 0.297 0.218 (0.379) (0.383) (0.418) (0.430) (0.445) L.NVR stock knowledge 4.221*** 4.240*** 4.228*** 3.978*** (1.092) (1.092) (1.054) (1.077) L.Violent Resistance 0.512*** 0.460** 0.366* (0.143) (0.142) (0.159) L.GDP per capita (log) -0.199 -0.111 (0.284) (0.385) L.EDI (moving average) 0.0490 -0.671 (0.759) (1.268) L.Judicial constraints 1.518 (1.279) L.Legislative constraints 1.100 (1.032) L.Polarization 0.369** (0.117) L.Education -0.109 (0.117) L.Population (log) -0.0786 (0.136) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.563*** -2.015 -2.040 (1.019) (1.019) (1.021) (2.496) (3.091) Observations 6008 6008 6008 6008 6008 AIC 2347 2300 2290 2290 2234 BIC 2876 2836 2873 2893 2870 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 39 This finding underscore how violent, rather than non-violent resistance, had been a driving force for U-Turns in the more distant. This can be illustrated by the case example of Argentina. After a military coup in 1955, the country experienced a significant decline on their democracy levels. The coup was supported by a coalition of military factions, conservative political groups, and members of middle and upper classes who opposed the regime (Romero, 2015). The overthrow of the former regime included violent means and is coded in NAVCO as a violent resistance campaign (Chenoweth & Shay, 2020a). A few years later, a new constitution was adopted and lead to national elections in 1958, which was the starting point for a period of democratization and is now considered as a U-Turn (Jones et al., 2005; Nord et al., 2024b, p. 14). Figure 10 illustrates the much smaller (and insignificant) marginal effects for NVRs on U-Turns. When countries experience NVR, the probability of observing a U-Turn is only 7%. When experiencing two NVRs, the probability is 9%. The baseline likelihood is below 5%. Figure 10: Marginal effect plot for influence of NVR pre-autocratization on U-Turns with 95% confidence intervals for the time span 1900-1993 However, the variable “NVR stock knowledge” is statistically significant in all models with much greater coefficients than in the previous models. This might sound counterintuitive in the first place since “NVR” is insignificant and the marginal effects of “NVR” much smaller than in the previous models. Again, a case example helps to 40 understand the mechanisms behind the result. South Korea experienced a major student- led revolution with hundreds of thousands of people participating in non-violent resistance after a rigged election in 1960 (Sonn, 2013). The non-violent resistance campaign was eventually successful and led to the downfall of the Rhee regime. However, these democratic achievements were short-lived as General Park Chung-Hee staged a military coup and established a military dictatorship in 1961. A few years later, Park called for national elections to legitimate his power claim (Kim & Vogel, 2011). Even though the elections were manipulated, they lead to an increase in the level of democracy again, which is now considered a U-Turn (Nord et al., 2024b). This historical example illustrates that while non-violent resistance initially led to democratic gains, subsequent events, such as military coups and manipulated elections, complicated the relationship between NVR and U-Turns and explains why the stock knowledge exhibits a significant and large influence on U-Turns. Figure 11 shows the marginal effect plot for “NVR stock knowledge” on U-Turns. We can see that the marginal effects are greater for the values 0.6 to 1 than in the plots before. For instance, the probability of observing a U-Turn when “NVR stock knowledge” = 1 is 71%, an increase of 14.9% compared to Figure 9 and an increase of 20.2% compared to Figure 7. Figure 11: Marginal effect plot for influence of NVR stock knowledge on U-Turns with 95% confidence intervals for the time span 1900-1993 41 All in all, these results suggests that the main relationship between non-violent resistance and U-Turns might not be as strong as in recent years when looking at the more distant past. By contrast, violent resistance seems to have a relatively large influence on U-Turns for the time period 1900 to 1993, as well as “NVR stock knowledge”. 5.4 Regression Diagnostics Two commonly used measurements were included to compare the goodness of fit across these five models. The first one is the Akaike information criterion (AIC), and the second one is the Bayesian information criterion (BIC). Both can be used to assess “which of the models best approximates the data” (Fabozzi et al., 2014, p. 399). The lower the values of the AIC and BIC, the better the goodness of fit. Looking at the AIC and BIC values across the models, we can observe that in Table 2, 3, and 4, the most refined Model 5 has the lowest value, indicating that the goodness of fit improves as the models becomes more refined. Hence, it can be said that Model 5 is the best model to approximate the data according to both, the AIC and the BIC. In addition, I conducted the variance inflation test (VIF) to check whether some of the explanatory variables are highly correlated with each other. The output can be found in the Appendix at Table 5. The test shows that none of the independent variables have tolerance values (1/VIF) below 0.2, indicating that there are no strong signs for multicollinearity in the model. Moreover, I checked for influential observations that could have manipulated the results by using Pregibon’s (1981) ∆?̂? influence statistics. Figure 12 in the Appendix shows the units with the highest influence on the model. I run an additional model and exclude the most influential outliers (db > 0.05). Results can be found in Table 6 in the Appendix and demonstrate that these outliers have not manipulated any result. No coefficients substantially change which increases our confidence in the previous results. 5.5 Robustness checks and alternative model specifications First, I include country fixed effects to account for country-specific differences. Results in Table 7 in the Appendix show that the main results hold when adding country fixed effects to the models, even though the sample size drops from 9266 in Table 2 to 4030 in 42 Table 7. One notable difference can be seen, which is that violent resistance is statistically significant in Model 3 (p < 0.5), but then turns insignificant in Model 4 and 5 (p > 0.05). In general, this robustness test tells us that the results are not driven by any country specific differences and increases our confidence in the previous results. Since I detected that violent resistance seems to be a driver for U-Turns in the more distant past, I run an additional model in which I change the non-violent NAVCO variables with the violent ones to gain a deeper insight into this relationship. This also means that the “NVR stock knowledge” is replaced by a new variable “Violent Resistance stock knowledge”. Results from Table 8 in the Appendix cover the period from 1900 to 1993. The Violent Resistance coefficient is statistically significant at p < 0.05 and has a positive association with U-Turns across all models. By contrast, “Violent Resistance stock knowledge” as well as the “NVR” variable are not statistically significant (p > 0.05). This result support the observation that violent resistance has a large influence on U-Turns for this time span. To check the assumption that the knowledge of non-violent resistance diminishes after 10 years, I run three more models. One accounts for non-violent resistance of the last 20 years, the second for the last 30 years, and the third one without a threshold. The variable “NVR stock knowledge” was adjusted accordingly. Table 9 (threshold of 20 years) and Table 10 (threshold of 30 years) in the Appendix show slightly bigger coefficients of the log odds for the variable “NVR stock knowledge” than in the original model, whereas Table 11 (no threshold) show slightly smaller coefficients. In general, one can say that the results do not substantially change. This indicates that even when accounting for longer historical time frames, the NVR stock knowledge remains a significant factor in explaining U-Turns. When looking at the distribution of the variable that measures how many years have passed since the last NVR and autocratization3, we can see that there are simply much more NVRs that happened in the last ten years which could explain this result. Next, I only use data for non-violent- and violent resistance which primary aims for regime change. NAVCO also codes resistance campaigns with the aim of expelling 3 A histogram of this variable can be found in the Appendix at Figure 13 43 foreign occupation, self-determination or separatism, or other types of social change, such as anti-apartheid campaigns (Chenoweth & Shay, 2020b). These campaigns were included in all regression so far, and this test aims to isolate the influence of resistance campaigns that primarily aim for regime change since this is the core interest of this study. The results in Table 12 in the Appendix shows consistency in comparison to the main model from 1900 to 2020. This test indicates that the influence of the models before were not driven by data on resistance episodes that accounts for more factors than regime change. Finally, I re-run the first three initial models by using the Probit estimation to check if the results substantially change when choosing a different probability calculation. Results in Table 13, 14, and 15 in the Appendix do not substantially differ from previous results, although the coefficients are smaller for the main independent variables than in Table 2,3, and 4. This result is expectable since the probit model uses a different link function to calculate the scale of the coefficients. Thus, these results further increase the confidence in the robustness of my results. 5.6 Discussion and Limitations This part summarizes the main insights of the study and outlines its limitations. The study provides robust support for Hypothesis One: “Non-violent resistance campaigns during episodes of autocratization significantly contribute to successful democratic resilience.” Specifically, the analysis reveals that when countries experience non-violent resistance during episodes of autocratization, there is a notable 14% probability of achieving a U- Turn. Furthermore, if two NVRs occur, this probability increases substantially to 29.7%. During the third wave of autocratization, the impact of non-violent resistance is even greater. A single episode of non-violent resistance increases the probability of a U-turn to 23.8%, while two such episodes raise the probability to an impressive 53.3%. In the more distant past, violent resistance appears to be more influential in increasing the likelihood of U-Turns. The study also provides robust support for Hypothesis Two: “Non-violent resistance campaigns that happen before the onset of autocratization significantly contribute to successful democratic resilience.” The analysis shows that when non-violent resistance occurs one year before the onset of autocratization, the probability of achieving a U-Turn 44 is a remarkable 50.8%. During the third wave of autocratization, this probability increases to 56.1%. As theoretically expected, the influence of the variable “NVR stock knowledge” gradually diminishes over time. After 5 years, the probability falls to less than half. However, the additional models which extended the threshold of years shows that the “knowledge” of non-violent resistance among CSOs persists for decades and never completely disappears. In the more distant past, the impact of the “NVR stock knowledge” is considerably higher as explained by the case example of South Korea in the 1960s. A notable limitation of this study is the inability to establish causality. In the absence of a counterfactual, i.e. a scenario in which certain conditions are changed, I can not infer causality from the study’s findings. For example, it remains uncertain whether countries that have experienced non-violent resistance would still make a U-Turn in the absence of civil resistance. It is not possible to answer such questions because I do not have control units for each country, as it would be the case in a controlled experimental setting. Another limitation could be a potential underreporting bias of the NAVCO dataset. The threshold of being included as a campaign is relatively high (1,000 participants) which could imply that civil resistance that failed in its very early stages are not recorded. This could lead to a bias towards successful campaigns (Chenoweth & Shay, 2020b, pp. 7-9). In addition, the use of U-Turns as an operationalization for successful democratic resilience can be questioned as well. While U-Turns reflect effective recovery from autocratization which restores the old level of democracy, it does not reflect cases that stopped autocratization but remained at a lower democracy level or gain much higher democracy levels than previous to the onset of autocratization. Furthermore, some codings of U-Turns in the 20th century represent the victory of the Allies over Nazi Germany and are not to be seen in the classical sense of democratic resilience. The same can be said about U-Turns that partially happened because of international intervention, for example in Mali between 2007 and 2014. After a military coup in 2012, Mali faced a violent uprising in its Northern region caused by Islamist rebels. French army forces helped the government to defeat the rebellion which lead to new elections in 2013 (Pezard & Shurkin, 2015). This U-Turn is different in its mechanisms and outcomes from the NVR approach. However, I believe these limitations 45 do not significantly limit the interpretation of the study, but are rather a call for further research on U-Turns and its causes. Finally, the main focus of this study has been on non- violent resistance. As we have seen, violent resistance had been a driving factor for U- Turns in the more distant past that may merit further examination. 46 6. Conclusion The aim of this thesis is to answer the research question “What is the role of non-violent resistance campaigns for episodes of successful democratic resilience?”. I argue that non- violent resistance initiated by CSOs can not only stop episodes of autocratization, but also often lead to re-democratization. Using a logistic regression model between 1900 and 2020, I demonstrate that non-violent resistance contributes significantly to successful democratic resilience. This result is particularly evident for the last three decades, i.e. in the period of the third wave of autocratization. I also show that earlier experiences of non- violent experience empower civil society to mobilize when democracy is under attack by providing CSOs with important knowledge, establishing a resilience network, and lowering the barriers for citizens to participate in non-violent resistance. These findings provide important insights for research on democratic resilience which has so far underestimated the role of civil society in preserving democratic structures and freedoms. The overall take-away of this thesis is that CSOs can stop autocratizing incumbents when they join forces and use means of non-violent resistance, and that earlier experiences of non-violent resistance helps to foster a resilience network which is ready to react once autocratization starts. Future studies should investigate if and how specific features of CSOs, for example the autonomy from government, help to mobilize citizens in times of autocratization. Understanding the mechanisms through which CSOs effectively mobilize citizens can inform strategies to strengthen civil society’s role in democratic resilience. Future research should also delve deeper into the interplay of CSOs with other resilience actors, such as parliament, judiciary, and political parties. Exploring how these different actors help or hinder each other can provide a more comprehensive understanding of the dynamics at play. Finally, future research can adapt an expanded definition of democratic resilience - as this thesis did by using the U-Turn operationalization - to not only explaining the halt of democratic breakdowns, but also the restoration of democracy. Democracy is entering a crucial phase, with a record number of countries holding national elections in 2024, many of them taking place in regimes with declining levels of democracy (Nord et al., 2024a). As this study shows, CSOs are an important cornerstone for preserving democratic structures and freedoms. The organization of non-violent 47 resistance by CSOs to mobilize people to vote could help democracies to halt autocratization (and eventually restore it) in one of its most difficult times. 48 7. List of References Aldrich, J. H. (1984). Linear probability, logit and probit models. Sage. Alimi, E. Y. (2015). Repertoires of Contention. In D. della Porta & M. Diani (Eds.), The Oxford Handbook of Social Movements. Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199678402.013.42 Altman, D. (2019). Citizenship and contemporary direct democracy. Cambridge University Press. Andrinek, G. (2022). Why Slovenia booted a populist incumbent for a green rookie. Deutsche Welle. https://www.dw.com/en/what-political-newcomer-robert-golobs-election-win- means-for-slovenia/a-61594261 Angiolillo, F., Lundstedt, M., Nord, M., & Lindberg, S. I. (2024). State of the world 2023: democracy winning and losing at the ballot. Democratization, 1-25. https://doi.org/10.1080/13510347.2024.2341435 Anisin, A. (2020). Unravelling the complex nature of security force defection. Global Change, Peace & Security, 32(2), 135-155. https://doi.org/10.1080/14781158.2020.1767046 Ash, T. G. (1983). The Polish revolution : Solidarity 1980-82. Cape. Barro, R. J. (1999). Determinants of Democracy. The Journal of political economy, 107(S6), 158- 183. https://doi.org/10.1086/250107 Bayer, M., Bethke, F. S., & Lambach, D. (2016). The democratic dividend of nonviolent resistance. Journal of Peace Research, 53(6), 758-771. https://doi.org/https://doi.org/10.1177/0022343316658090 Beissinger, M. R. (2008). A New Look at Ethnicity and Democratization. Journal of democracy, 19(3), 85-97. https://doi.org/10.1353/jod.0.0017 Beissinger, Mark R. (2022). The Revolutionary City: Urbanization and the Global Transformation of Rebellion. Princeton University Press. https://doi.org/10.2307/j.ctv2175r9q Berman, S. (1997). Civil Society and the Collapse of the Weimar Republic. World Politics, 49(3), 401-429. http://www.jstor.org/stable/25054008 Bernhard, M. (2020). What do we know about civil society and regime change thirty years after 1989? East European Politics, 36(3), 341-362. https://doi.org/10.1080/21599165.2020.1787160 Bernhard, M., Fernandes, T., & Branco, R. (2017). Introduction: Civil Society and Democracy in an Era of Inequality. Comparative Politics, 49(3), 297-309. http://www.jstor.org/stable/26330959 Bernhard, M., Hicken, A., Reenock, C., & Lindberg, S. I. (2020). Parties, Civil Society, and the Deterrence of Democratic Defection. Studies in Comparative International Development, 55(1), 1-26. https://doi.org/10.1007/s12116-019-09295-0 Bertocchi, G., & Spagat, M. (2001). The politics of co-optation. Journal of Comparative Economics, 29(4), 591-607. Bethke, F. S., & Pinckney, J. (2021). Non-violent resistance and the quality of democracy. Conflict Management and Peace Science, 38(5), 503-523. https://doi.org/10.1177/0738894219855918 Boese, V. A. (2019). How (not) to measure democracy. International Area Studies Review, 22(2), 95-127. https://doi.org/10.1177/2233865918815571 Boese, V. A., Lundstedt, M., Morrison, K., Sato, Y., & Lindberg, S. I. (2022). State of the world 2021: autocratization changing its nature? Democratization, 29(6), 983-1013. https://doi.org/10.1080/13510347.2022.2069751 Boix, C. (2003). Democracy and redistribution. Cambridge University Press. Bolt, J., & Van Zanden, J. L. (2020a). Maddison Project Database, version 2020. https://www.rug.nl/ggdc/historicaldevelopment/maddison/releases/maddison- project-database-2020?lang=en 49 Chalcraft, J. (2016). Popular politics in the making of the modern Middle East. Cambridge University Press. Chenoweth, E., & Lewis, O. A. (2013). Unpacking nonviolent campaigns:Introducing the NAVCO 2.0 dataset. Journal of Peace Research, 50(3), 415-423. https://doi.org/10.1177/0022343312471551 Chenoweth, E., & Shay, C. W. (2020a). List of Campaigns in NAVCO 1.3. https://doi.org/10.7910/DVN/ON9XND/PTMCCV&version=1.0 Chenoweth, E., & Shay, C. W. (2020b). NAVCO 1.3 Codebook List.pdf. V1. https://doi.org/https://doi.org/10.7910/DVN/ON9XND/PTMCCV Chenoweth, E., & Stephan, M. J. (2011). Why Civil Resistance Works The Strategic Logic of Nonviolent Conflict. Columbia University Press. http://www.jstor.org/stable/10.7312/chen15682 Cho, Y. H., & Hwang, I. (2021). Who defends democracy and why? Explaining the participation in the 2016-2017 candlelight protest in South Korea. Democratization, 28(3), 625-644. https://doi.org/10.1080/13510347.2020.1845149 Coppedge, M., Edgell, A. B., Knutsen, C. H., & Lindberg, S. I. (2022). Why democracies develop and decline. Cambridge University Press. Coppedge, M., Gerring, J., Knutsen, C. H., Lindberg, S. I., Teorell, J., Altman, D., Angiolillo, F., Bernhard, M., Borella, C., Cornell, A., Fish, M. S., Fox, L., Gastaldi, L., Gjerløw, H., Glynn, A., Good God, A., Grahn, S., Hicken, A., Kinzelbach, K., . . . Ziblatt, D. (2024a). V-Dem [Country-Year/Country-Date] Dataset v14” Varieties of Democracy (V-Dem) Project. https://doi.org/10.23696/mcwt-fr58 Coppedge, M., Gerring, J., Knutsen, C. H., Lindberg, S. I., Teorell, J., Altman, D., Angiolillo, F., Bernhard, M., Borella, C., Cornell, A., Fish, S. M., Fox, L., Gastaldi, L., Gjerløw, H., Glynn, A., God, A. G., Grahn, S., Hicken, A., Kinzelbach, K., . . . Ziblatt, D. (2024b). V-Dem Codebook v14" Varieties of Democracy (V-Dem) Project. Cunningham, D. (2004). There’s something happening here: The new left, the Klan, and FBI counterintelligence. University of California Press. Dahl, R. A. (1971). Polyarchy: Participation and opposition. Yale university press. Dalton, R. J. (2017). Civil Society Mobilizing Action. In R. J. Dalton (Ed.), The Participation Gap: Social Status and Political Inequality (pp. 63-82). Oxford University Press. https://doi.org/10.1093/oso/9780198733607.003.0004 Davenport, C. (2007). State Repression and Political Order. Annual Review of Political Science, 10(1), 1-23. https://doi.org/10.1146/annurev.polisci.10.101405.143216 Della Porta, D. (1995). Social movements, political violence, and the state: A comparative analysis of Italy and Germany. Cambridge University Press. Della Porta, D. (2016). Where Did the Revolution Go?: Contentious Politics and the Quality of Democracy. Cambridge University Press. https://doi.org/10.1017/9781316783467 Della Porta, D. (2020). Building Bridges: Social Movements and Civil Society in Times of Crisis. VOLUNTAS: International Journal of Voluntary and Nonprofit Organizations, 31(5), 938- 948. https://doi.org/10.1007/s11266-020-00199-5 Della Porta, D., & Reiter, H. (1998). Policing protest: The control of mass demonstrations in Western democracies (Vol. 6). University of Minnesota Press. DeNardo, J. (2014). Power in numbers : the political strategy of protest and rebellion. Princeton University Press. Diamond, L. (1994). Rethinking Civil Society: Toward Democratic Consolidation. Journal of democracy, 5(3), 4-17. https://doi.org/10.1353/jod.1994.0041 Diani, M. (2015). The Cement of Civil Society: Studying Networks in Localities. Cambridge University Press. https://doi.org/10.1017/CBO9781316163733 50 Diani, M., & McAdam, D. (2003). Social Movements and Networks: Relational Approaches to Collective Action. Oxford University Press. https://doi.org/10.1093/0199251789.001.0001 Dionne, K. Y. (2024). Why Malawi's Democracy Endures. Journal of democracy, 35(2), 122-135. https://doi.org/10.1353/jod.2024.a922838 Earl, J. (2022). Repression and Social Movements. In D. A. Snow, D. Della Porta, B. Klandermans, & D. McAdam (Eds.), The Wiley-Blackwell Encyclopedia of Social and Political Movements: John Wiley & Sons Ltd. Edwards, B., & Gillham, P. F. (2013). Resource Mobilization Theory. In D. A. Snow, D. Della Porta, B. Klandermans, & D. McAdam (Eds.), The Wiley-Blackwell Encyclopedia of Social and Political Movements: Blackwell Publishing Ltd. Edwards, M. (2011). Introduction: Civil Society and the Geometry of Human Relations. In M. Edwards (Ed.), The Oxford Handbook of Civil Society. Oxford University Press. https://doi.org/10.1093/oxfordhb/9780195398571.013.0001 Eisinger, P. K. (1973). The conditions of protest behavior in American cities. American political science review, 67(1), 11-28. Ekiert, G. (1999). Rebellious civil society : popular protest and democratic consolidation in Poland, 1989-1993. University of Michigan Press. https://doi.org/10.3998/mpub.16117 Ellefsen, R. (2018). Deepening the Explanation of Radical Flank Effects: Tracing Contingent Outcomes of Destructive Capacity. Qualitative Sociology, 41(1), 111-133. https://doi.org/10.1007/s11133-018-9373-3 Fabozzi, F. J., Focardi, S. M., Rachev, S. T., Arshanapalli, B. G., & Hoechstoetter, M. (2014). The Basics of Financial Econometrics: Tools, Concepts, and Asset Management Applications (1 ed.). Wiley. https://doi.org/10.1002/9781118856406 Fariss, C., Anders, T., Markowitz, J., & Barnum, M. (2021). Replication Data for: New Estimates of Over 500 Years of Historic GDP and Population Data Version V4) Harvard Dataverse. https://doi.org/doi:10.7910/DVN/DC0ING Gamboa, L. (2022). Resisting backsliding : opposition strategies against the erosion of democracy. Cambridge University Press. Gandhi, J., & Przeworski, A. (2006). Cooperation, cooptation, and rebellion under dictatorships. Economics & politics, 18(1), 1-26. https://doi.org/https://doi.org/10.1111/j.1468- 0343.2006.00160.x Gerschewski, J. (2013). The three pillars of stability: legitimation, repression, and co-optation in autocratic regimes. Democratization, 20(1), 13-38. https://doi.org/10.1080/13510347.2013.738860 Glaeser, E. L., Ponzetto, G. A. M., & Shleifer, A. (2007). Why does democracy need education? Journal of economic growth (Boston, Mass.), 12(2), 77-99. https://doi.org/10.1007/s10887-007-9015-1 Grahn, S., & Lührmann, A. (2021). Good seed makes a good crop? The relationship between civil society and post-independence democracy levels. Journal of Civil Society, 17(3-4), 297-322. https://doi.org/10.1080/17448689.2021.2003139 Haggard, S., & Kaufman, R. R. (1995). The political economy of democratic transitions. Princeton University Press. Haines, H. H. (2013). Radical Flank Effects. In D. A. Snow, D. Della Porta, B. Klandermans, & D. McAdam (Eds.), The Wiley-Blackwell Encyclopedia of Social and Political Movements: John Wiley & Sons, Ltd. Heaney, M. T. (2022). Elections and Social Movements. In D. A. Snow, D. Della Porta, B. Klandermans, & D. McAdam (Eds.), The Wiley-Blackwell Encyclopedia of Social and Political Movements: Blackwell Publishing Ltd. 51 Hellmeier, S., & Bernhard, M. (2023). Regime Transformation From Below: Mobilization for Democracy and Autocracy From 1900 to 2021. Comparative Political Studies, 56(12), 1858-1890. https://doi.org/10.1177/00104140231152793 Hess, D., & Martin, B. (2006). Backfire, Repression, and the Theory of Transformative Events. Mobilization, 11, 249-267. https://doi.org/10.17813/maiq.11.2.3204855020732v63 Hoffman, A. J., & Bertels, S. (2012). Who is part of the environmental movement? In Good Cop/Bad Cop (pp. 48-69). Routledge. Honari, A. (2018). From 'the effect of repression' toward 'the response to repression'. Current Sociology, 66(6), 950-973. https://doi.org/10.1177/0011392118787585 Huntington, S. P. (1991). The third wave. University of Oklahoma Press. Jones, M., Lauga, M., & León-Roesch, M. (2005). Argentina. In D. Nohlen (Ed.), Elections in the Americas. A data handbook (Vol. 2, pp. 59-108). https://doi.org/https://doi.org/10.1093/oso/9780199283583.003.0002. Jung, J. K. (2023). The candlelight protests in South Korea: a dynamics of contention approach. Social Movement Studies, 22(5-6), 767-785. https://doi.org/10.1080/14742837.2022.2053515 Karatnycky, A. (2005). Ukraine's orange revolution. Foreign Aff., 84, 35-53. https://heinonline.org/HOL/P?h=hein.journals/fora84&i=249 Keck, M. E., & Sikkink, K. (1998). Activists beyond borders: Advocacy networks in international politics. Cornell University Press. Kim, B.-K., & Vogel, E. F. (2011). The Park Chung Hee Era: The Transformation of South Korea. Harvard University Press. Kurtz, L. R. (2010). Otpor and the Struggle for Democracy in Serbia. International Center on Nonviolent Conflict. https://www.nonviolent-conflict.org/wp- content/uploads/2016/02/Kurtz-Otpor-Serbia-5.pdf Laebens, M. G., & Lührmann, A. (2021). What halts democratic erosion? The changing role of accountability. Democratization, 28(5), 908-928. https://doi.org/10.1080/13510347.2021.1897109 Leuschner, E., & Hellmeier, S. (2024). State Concessions and Protest Mobilization in Authoritarian Regimes. Comparative Political Studies, 57(1), 3-31. https://doi.org/10.1177/00104140231169022 Levitsky, S. (2018). How democracies die. Penguin Books. Lichbach, M. I. (1987). Deterrence or escalation? The puzzle of aggregate studies of repression and dissent. Journal of Conflict Resolution, 31(2), 266-297. Linz, J. J. (1978). The breakdown of democratic regimes. Linz, J. J. (1990). The perils of presidentialism. Journal of democracy, 1(1), 51-69. Linz, J. J., & Stepan, A. (1996). Problems of democratic transition and consolidation: Southern Europe, South America, and post-communist Europe. Johns Hopkins University Press. Lipset, S. M. (1959). Some Social Requisites of Democracy: Economic Development and Political Legitimacy. The American political science review, 53(1), 69-105. https://doi.org/10.2307/1951731 Lorch, J. (2021). Elite capture, civil society and democratic backsliding in Bangladesh, Thailand and the Philippines. Democratization, 28(1), 81-102. https://doi.org/10.1080/13510347.2020.1842360 Lührmann, A., & Lindberg, S. I. (2019). A third wave of autocratization is here: what is new about it? Democratization, 26(7), 1095-1113. https://doi.org/10.1080/13510347.2019.1582029 Maerz, S. F., Edgell, A. B., Wilson, M. C., Hellmeier, S., & Lindberg, S. I. (2023). Episodes of regime transformation. Journal of Peace Research. https://doi.org/10.1177/00223433231168192 52 Marinov, N. (2005). Do Economic Sanctions Destabilize Country Leaders? American Journal of Political Science, 49(3), 564-576. https://doi.org/10.2307/3647732 Martin, G. (2015). Understanding social movements (1 Edition. ed.). Routledge. Marusic, S. J. (2015). Macedonians Stage Mass Protest for PM’s Resignation. BalkanInsight. https://balkaninsight.com/2015/05/17/macedonia-braces-for-big-anti-government- protest/ Marwell, G., & Oliver, P. (1993). The critical mass in collective action. Cambridge University Press. McAdam, D. (1983). Tactical Innovation and the Pace of Insurgency. American Sociological Review, 48(6), 735-754. https://doi.org/10.2307/2095322 McAdam, D., & Tarrow, S. (2010). Ballots and Barricades: On the Reciprocal Relationship between Elections and Social Movements. Perspectives on Politics, 8(2), 529-542. https://doi.org/10.1017/S1537592710001234 McCarthy, J. D., Britt, D. W., & Wolfson, M. (1991). The institutional channeling of social movements by the state in the United States. Research in Social Movements, Conflicts and Change, 13(2), 45-76. McCarthy, J. D., & Zald, M. N. (1977). Resource mobilization and social movements: A partial theory. American journal of sociology, 82(6), 1212-1241. McCoy, J., & Somer, M. (2019). Toward a Theory of Pernicious Polarization and How It Harms Democracies: Comparative Evidence and Possible Remedies. The ANNALS of the American Academy of Political and Social Science, 681(1), 234-271. https://doi.org/10.1177/0002716218818782 Mehmetoglu, M. (2022). Applied statistics using Stata : a guide for the social sciences (Second edition ed.). SAGE. Merkel, W., & Lührmann, A. (2021). Resilience of democracies: responses to illiberal and authoritarian challenges. Democratization, 28(5), 869-884. https://doi.org/10.1080/13510347.2021.1928081 Meyer, D. S., & Staggenborg, S. (1996). Movements, Countermovements, and the Structure of Political Opportunity. American journal of sociology, 101(6), 1628-1660. http://www.jstor.org/stable/2782114 Mize, T. D., Doan, L., & Long, J. S. (2019). A General Framework for Comparing Predictions and Marginal Effects across Models. Sociological Methodology, 49(1), 152-189. https://doi.org/10.1177/0081175019852763 Monitor, C. (2022). Country Update: Slovenia. Overview of Recent Restrictions to Civic Freedoms. https://civicus.org/documents/SloveniaCountryBrief.6April2022.pdf Mounk, Y. (2018). The people vs. democracy : why our freedom is in danger and how to save it. Harvard University Press. Mueller, D. C. (2003). Public Choice III (3 ed.). Cambridge University Press. https://doi.org/10.1017/CBO9780511813771 Mueller, L. (2018). Political protest in contemporary Africa. Cambridge University Press. Narsee, A. (2022). Euroviews. EU leaders must pay attention to threatened civic freedoms in Slovenia. Euronews. https://www.euronews.com/2022/01/17/eu-leaders-must-pay- attention-to-threatened-civic-freedoms-in-slovenia-view Nord, M., Angiolillo, F., Lundstedt, M., Wiebrecht, F., & Lindberg, S. I. (2024b). When Autocratization is Reversed: Episodes of Democratic Turnarounds since 1900. V-Dem Working Paper, 147. https://v-dem.net/media/publications/wp_147_yvOYnKU.pdf Nord, M., Lundstedt, M., Altman, D., Angiolillo, F., Borella, C., Fernandes, T., Gastaldi, L., Good God, A., Natsika, N., & Lindberg, S. I. (2024a). Democracy Report 2024: Democracy Winning and Losing at the Ballot. https://v-dem.net/documents/44/v- dem_dr2024_highres.pdf 53 Olson Jr, M. (1971). The Logic of Collective Action: Public Goods and the Theory of Groups, with a new preface and appendix (Vol. 124). Harvard University Press. Opp, K.-D. (2019). Political Mobilization Approaches. In W. Merkel, R. Kollmorgen, & H.-J. Wagener (Eds.), The Handbook of Political, Social, and Economic Transformation. Oxford University Press. https://doi.org/10.1093/oso/9780198829911.003.0014 Opp, K.-D., & Gern, C. (1993). Dissident Groups, Personal Networks, and Spontaneous Cooperation: The East German Revolution of 1989. American Sociological Review, 58(5), 659-680. https://doi.org/10.2307/2096280 Ostrom, E. (2009). Collective Action Theory. In C. Boix & S. C. Stokes (Eds.), The Oxford Handbook of Comparative Politics. Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199566020.003.0008 Papada, E., Altman, D., Angiolillo, F., Gastaldi, L., Köhler, T., Lundstedt, M., Natsika, N., Nord, M., Sato, Y., & Wiebrecht, F. (2023). Defiance in the face of autocratization. Democracy report 2023. Democracy Report. https://v-dem.net/documents/29/V- dem_democracyreport2023_lowres.pdf Pemstein, D., Marquardt, K. L., Tzelgov, E., Wang, Y.-t., Krusell, J., & Miri, F. (2023). The V-Dem measurement model: latent variable analysis for cross-national and cross-temporal expert-coded data. V-Dem Working Paper, 21. https://v- dem.net/media/publications/Working_Paper_21_z5BldB1.pdf Pezard, S., & Shurkin, M. (2015). Achieving Peace in Northern Mali: Past Agreements, Local Conflicts, and the Prospects for a Durable Settlement (1 ed.). Santa Monica: RAND Corporation. https://doi.org/10.7249/j.ctt15zc57q Pinckney, J., Butcher, C., & Braithwaite, J. M. (2022). Organizations, Resistance, and Democracy: How Civil Society Organizations Impact Democratization. International Studies Quarterly, 66(1). https://doi.org/10.1093/isq/sqab094 Polletta, F., & Jasper, J. M. (2001). Collective Identity and Social Movements. Annual Review of Sociology, 27, 283-305. http://www.jstor.org/stable/2678623 Pregibon, D. (1981). Logistic Regression Diagnostics. The Annals of statistics, 9(4), 705-724. https://doi.org/10.1214/aos/1176345513 Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. Touchstone Books/Simon & Schuster. Rakner, L. (2021). Don’t Touch My Constitution! Civil Society Resistance to Democratic Backsliding in Africa´s Pluralist Regimes. Global Policy, 12(S5), 95-105. https://doi.org/doi/10.1111/1758-5899.12991 Riley, D. (2010). The Civic Foundations of Fascism in Europe: Italy, Spain, and Romania, 1870– 1945. Johns Hopkins University Press. Romero, L. A. (2015). A History of Argentina in the Twentieth Century: Updated and Revised Edition. Penn State Press. Runciman, D. (2018). How democracy ends. Profile Books. Sato, Y., Lundstedt, M., Morrison, K., Boese, V. A., & Lindberg, S. I. (2022). Institutional order in episodes of autocratization. V-Dem Working Paper, 133. https://v- dem.net/media/publications/WP_133.pdf Schock, K. (2005). Unarmed insurrections people power movements in nondemocracies. University of Minnesota Press. Sharp, G. (1973). The Politics of Nonviolent Action: Power and struggle (Vol. 1-3). Porter Sargent Shin, G.-W., & Moon, R. J. (2017). South Korea After Impeachment. Journal of democracy, 28(4), 117-131. https://doi.org/https://doi.org/10.1353/jod.2017.0072 Siisiäinen, M. (2003). Two concepts of social capital: Bourdieu vs. Putnam. International Journal of Contemporary Sociology, 40, 183-204. 54 Smulovitz, C., & Peruzzotti, E. (2000). Societal Accountability in Latin America. Journal of democracy, 11, 147. Sonn, J. (2013). April Revolution (Korea). In D. A. Snow, D. Della Porta, B. Klandermans, & D. McAdam (Eds.), The Wiley-Blackwell Encyclopedia of Social and Political Movements: Blackwell Publishing Ltd. Soule, S. (2013). Diffusion and Scale Shift. In D. A. Snow, D. Della Porta, B. Klandermans, & D. McAdam (Eds.), The Wiley-Blackwell Encyclopedia of Social and Political Movements: John Wiley & Sons, Ltd. Stephan, M. J., & Chenoweth, E. (2008). Why Civil Resistance Works: The Strategic Logic of Nonviolent Conflict. International security, 33(1), 7-44. https://doi.org/10.1162/isec.2008.33.1.7 Stojadinovic, S. (2019). North Macedonia's Colorful Revolution is Over. What's Next? . https://www.nonviolent-conflict.org/blog_post/north-macedonias-colorful-revolution- whats-next/ Tarrow, S. (2015). Contentious Politics. In D. della Porta & M. Diani (Eds.), The Oxford Handbook of Social Movements. Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199678402.013.8 Teorell, J., Coppedge, M., Lindberg, S., & Skaaning, S.-E. (2019). Measuring Polyarchy Across the Globe, 1900–2017. Studies in Comparative International Development, 54(1), 71-95. https://doi.org/10.1007/s12116-018-9268-z Tilly, C., & Tarrow, S. (2015). Contentious Politics. Oxford University Press, Incorporated. http://ebookcentral.proquest.com/lib/ub-wuerzburg/detail.action?docID=2121281 Tocqueville, A. (1969 (1835)). Democracy in America (G. Lawrence, Trans.; J. P. Mayer, Ed.). Doubleday, Anchor Books. Tomini, L., Gibril, S., & Bochev, V. (2023). Standing up against autocratization across political regimes: a comparative analysis of resistance actors and strategies. Democratization, 30(1), 119-138. https://doi.org/10.1080/13510347.2022.2115480 van Lit, J., van Ham, C., & Meijers, M. J. (2023). Countering autocratization: a roadmap for democratic defence. Democratization, 1-23. https://doi.org/https://doi.org/10.1080/13510347.2023.2279677 Verba, S., Schlozman, K. L., & Brady, H. E. (1995). Voice and equality: Civic voluntarism in American politics. Harvard University Press. VonDoepp, P. (2020). Resisting Democratic Backsliding: Malawi’s Experience in Comparative Perspective. African Studies Review, 63(4), 858-882. https://doi.org/10.1017/asr.2019.62 55 8. Appendix Figure 6: Histogram for frequency of variable “NVR” (0 is excluded) Table 5: VIF-Test Variable VIF 1/VIF L.EDI (moving average) 4.80 0.208 L.Education 4.08 0.245 L.Legislative constraints 3.95 0.253 L.GDP per capita (log) 3.70 0.270 L.Judicial constraints 3.53 0.283 L.Polarization 1.47 0.678 L.Population (log) 1.16 0.860 L.Violent Resistance 1.14 0.880 L.NVR 1.16 0.946 L.NVR stock knowledge 1.03 0.975 Mean VIF 2.63 56 Figure 12: Pregibon’s measure of influential observations 57 Table 6: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 2020, excluding most influential observations Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.025*** 1.138*** 1.126*** 1.089*** 0.930*** (0.208) (0.214) (0.209) (0.210) (0.196) L.NVR stock knowledge 3.242*** 3.246*** 3.111*** 3.011*** (0.566) (0.569) (0.565) (0.592) L.Violent Resistance 0.218 0.115 0.0405 (0.135) (0.133) (0.124) L.GDP per capita (log) -0.498** -0.670* (0.180) (0.284) L.EDI (moving average) 0.520 -0.0692 (0.616) (1.058) L.Judicial constraints 1.191 (1.234) L.Legislative constraints 0.295 (0.743) L.Polarization 0.376*** (0.109) L.Education 0.0828 (0.115) L.Population 0.00586 (0.106) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.540*** 0.202 1.604 (1.019) (1.019) (1.019) (1.709) (2.193) Observations 8995 8995 8995 8995 8995 AIC 4081 3960 3961 3885 3802 BIC 4792 4685 4721 4695 4647 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 58 Table 7: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 2020, including country fixed effects Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 0.647* 0.837** 0.857** 0.901*** 0.584* (0.270) (0.302) (0.285) (0.273) (0.265) L.NVR stock knowledge 3.285*** 3.252*** 3.200*** 2.872*** (0.852) (0.877) (0.890) (0.805) L.Violent Resistance 0.403* 0.356 0.113 (0.194) (0.185) (0.175) L.GDP per capita (log) -0.501 -0.201 (0.473) (0.569) L.EDI (moving average) 0.666 -0.242 (1.440) (1.349) L.Judicial constraints 1.947 (1.748) L.Legislative constraints 0.626 (1.046) L.Polarization 0.813*** (0.227) L.Education -0.154 (0.294) L.Population 0.526 (0.738) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Country Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -1.757 -1.958 -1.892 1.221 -8.667 (1.202) (1.196) (1.199) (3.135) (8.543) Observations 4030 4030 4030 4030 4030 AIC 2670 2622 2607 2595 2487 BIC 3034 2950 2934 2922 2815 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 59 Table 8: Logit Model of violent resistance campaigns on U-Turns from 1900 to 1993 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.Violent Resistance 0.517*** 0.518*** 0.507*** 0.450** 0.335* (0.143) (0.143) (0.143) (0.145) (0.160) L.Violent resistance stock 0.640 0.648 0.544 0.466 knowledge (0.857) (0.857) (0.921) (0.917) L.NVR 0.311 0.261 0.150 (0.412) (0.424) (0.437) L.GDP per capita (log) -0.213 -0.154 (0.279) (0.379) L.EDI (moving average) 0.125 -0.576 (0.754) (1.239) L.Judicial constraints 1.539 (1.289) L.Legislative constraints 1.101 (1.020) L.Polarization 0.397*** (0.113) L.Education -0.0943 (0.116) L.Population (log) -0.0592 (0.133) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.564*** -3.564*** -3.563*** -1.926 -1.713 (1.021) (1.021) (1.021) (2.461) (3.044) Observations 6008 6008 6008 6008 6008 AIC 2331 2333 2335 2339 2278 BIC 2880 2889 2905 2942 2914 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 60 Table 9: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 1993, changing the threshold for NVR stock knowledge to 20 years Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.010*** 1.175*** 1.148*** 1.112*** 0.970*** (0.209) (0.219) (0.217) (0.220) (0.205) L.NVR stock knowledge 3.291*** 3.290*** 3.152*** 3.113*** (0.481) (0.485) (0.466) (0.513) L.Violent Resistance 0.201 0.111 0.0511 (0.128) (0.123) (0.114) ** * L.GDP per capita (log) -0.473 -0.633 (0.179) (0.282) L.EDI (moving average) 0.493 -0.248 (0.617) (1.021) L.Judicial constraints 1.143 (1.217) L.Legislative constraints 0.485 (0.733) L.Polarization 0.367*** (0.106) L.Education 0.0775 (0.115) L.Population (log) -0.0171 (0.104) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.539*** 0.0228 1.162 (1.019) (1.019) (1.019) (1.700) (2.225) Observations 9266 9266 9266 9266 9266 AIC 4312 4104 4106 4035 3947 BIC 5054 4853 4891 4862 4811 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 61 Table 10: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 1993, changing the threshold for NVR stock knowledge to 30 years Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.010*** 1.209*** 1.186*** 1.144*** 1.014*** (0.209) (0.216) (0.213) (0.218) (0.202) L.NVR stock knowledge 3.271*** 3.261*** 3.122*** 3.116*** (0.439) (0.443) (0.423) (0.475) L.Violent Resistance 0.178 0.0936 0.0448 (0.138) (0.131) (0.120) * * L.GDP per capita (log) -0.457 -0.621 (0.179) (0.284) L.EDI (moving average) 0.414 -0.415 (0.616) (1.023) L.Judicial constraints 1.134 (1.208) L.Legislative constraints 0.558 (0.728) L.Polarization 0.359*** (0.106) L.Education 0.0799 (0.114) L.Population (log) -0.0292 (0.104) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.537*** -0.0777 1.153 (1.019) (1.019) (1.019) (1.700) (2.239) Observations 9266 9266 9266 9266 9266 AIC 4312 4063 4066 4000 3913 BIC 5054 4812 4851 4827 4776 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 62 Table 11: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 1993, no threshold for NVR stock knowledge Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.010*** 1.255*** 1.236*** 1.184*** 1.060*** (0.209) (0.219) (0.215) (0.220) (0.201) L.NVR stock knowledge 2.999*** 2.979*** 2.835*** 2.851*** (0.374) (0.377) (0.361) (0.409) L.Violent Resistance 0.138 0.0615 0.0197 (0.158) (0.146) (0.131) * * L.GDP per capita (log) -0.436 -0.615 (0.179) (0.286) L.EDI (moving average) 0.322 -0.549 (0.608) (1.009) L.Judicial constraints 1.128 (1.206) L.Legislative constraints 0.582 (0.737) L.Polarization 0.357*** (0.107) L.Education 0.0860 (0.113) L.Population (log) -0.0376 (0.104) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.535*** -0.215 0.985 (1.019) (1.019) (1.019) (1.703) (2.263) Observations 9266 9266 9266 9266 9266 AIC 4312 4048 4052 3990 3901 BIC 5054 4804 4836 4818 4764 All independent variables are run with one-year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 63 Figure 13: Frequency of variable NVR_last (displayed in Histogram) which measures the duration in years between the most recent NVR before the onset of autocratization (0 is excluded from the histogram) 64 Table 12: Logit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 1993, only accounting for resistance episodes that primarily aim for regime change. Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 1.117*** 1.230*** 1.229*** 1.235*** 1.054*** (0.224) (0.233) (0.234) (0.216) (0.203) L.NVR stock knowledge 3.078*** 3.079*** 2.887*** 2.896*** (0.519) (0.526) (0.513) (0.529) L.Violent Resistance 0.446 0.246 0.0656 (0.230) (0.233) (0.269) L.GDP per capita (log) -0.478** -0.592* (0.170) (0.273) L.EDI (moving average) 0.679 0.00338 (0.593) (0.974) L.Judicial constraints 1.200 (1.144) L.Legislative constraints 0.464 (0.669) L.Polarization 0.394*** (0.0992) L.Education 0.0501 (0.113) L.Population 0.0607 (0.0991) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -3.526*** -3.526*** -3.542*** 0.00563 0.435 (1.019) (1.019) (1.020) (1.650) (2.192) Observations 9367 9367 9367 9367 9367 AIC 4505 4399 4394 4323 4216 BIC 5248 5149 5173 5152 5080 All independent variables are run with one year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 65 Table 13: Probit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 2020 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 0.520*** 0.569*** 0.552*** 0.537*** 0.443*** (0.119) (0.123) (0.122) (0.120) (0.111) L.NVR stock knowledge 1.760*** 1.748*** 1.676*** 1.616*** (0.311) (0.312) (0.307) (0.319) L.Violent Resistance 0.118 0.0633 0.0198 (0.0735) (0.0705) (0.0663) L.GDP per capita (log) -0.241** -0.335* (0.0849) (0.136) L.EDI (moving average) 0.189 -0.0678 (0.289) (0.492) L.Judicial constraints 0.478 (0.555) L.Legislative constraints 0.230 (0.340) L.Polarization 0.194*** (0.0518) L.Education 0.0418 (0.0520) L.Population 0.00642 (0.0490) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -1.902*** -1.902*** -1.909*** -0.0872 0.557 (0.433) (0.433) (0.433) (0.777) (1.077) Observations 9266 9266 9266 9266 9266 AIC 4318 4190 4189 4107 4012 BIC 5067 4939 4974 4934 4875 All independent variables are run with one year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 66 Table 14: Probit Model of non-violent resistance campaigns (NVR) on U-Turns from 1994 to 2020 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 0.720*** 0.789*** 0.791*** 0.794*** 0.687*** (0.155) (0.156) (0.154) (0.146) (0.129) L.NVR stock knowledge 1.672*** 1.674*** 1.566*** 1.404*** (0.350) (0.351) (0.343) (0.353) L.Violent Resistance -0.0161 -0.0694 -0.144 (0.108) (0.107) (0.106) L.GDP per capita (log) -0.319** -0.683*** (0.101) (0.200) L.EDI (moving average) 0.222 0.287 (0.404) (0.734) L.Judicial constraints 0.924 (0.777) L.Legislative constraints -0.524 (0.508) L.Polarization 0.231* (0.0902) L.Education 0.151 (0.0785) L.Population 0.126 (0.0810) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -1.637*** -1.678*** -1.673*** 0.892 1.823 (0.187) (0.196) (0.196) (0.678) (1.289) Observations 3136 3136 3136 3136 3136 AIC 1902 1818 1820 1723 1628 BIC 2059 1981 1989 1904 1839 All independent variables are run with one year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 67 Table 15: Probit Model of non-violent resistance campaigns (NVR) on U-Turns from 1900 to 1993 Dependent variable: U-Turn Model 1 Model 2 Model 3 Model 4 Model 5 L.NVR 0.182 0.201 0.161 0.140 0.0833 (0.183) (0.184) (0.196) (0.199) (0.203) L.NVR stock knowledge 2.302*** 2.310*** 2.305*** 2.170** (0.678) (0.679) (0.665) (0.666) L.Violent Resistance 0.283*** 0.256*** 0.202* (0.0779) (0.0759) (0.0841) L.GDP per capita (log) -0.0854 -0.0572 (0.123) (0.167) L.EDI (moving average) -0.0615 -0.372 (0.337) (0.568) L.Judicial constraints 0.690 (0.563) L.Legislative constraints 0.520 (0.445) L.Polarization 0.186** (0.0580) L.Education -0.0467 (0.0528) L.Population -0.0316 (0.0605) Time Fixed Effects ✓ ✓ ✓ ✓ ✓ Constant -1.902*** -1.902*** -1.919*** -1.234 -1.169 (0.433) (0.433) (0.435) (1.072) (1.313) Observations 6008 6008 6008 6008 6008 AIC 2347 2299 2285 2284 2227 BIC 2877 2835 2868 2887 2863 All independent variables are run with one year lag Standard errors in parentheses, clustered by country * p < 0.05, ** p < 0.01, *** p < 0.001 68 Results_MAThesis - Printed on 20.05.2024 13:01:22 1 ***Masterthesis Alexander Heinrich*** 2 3 clear 4 5 import excel "C:\Users\alexa\OneDrive\Desktop\NAVCO 1.3 List_prepmerge_finish.xlsx", firstrow //import NAVCO 6 7 browse 8 9 * Build variable protest-episodes per year 10 11 egen "NVR" = total(NONVIOL), by (country_name year) // only non-violent 12 13 egen "VR" = total(VIOL), by (country_name year) // only violent 14 15 * Create descriptive statistics for years in the count of protest events 16 17 keep if NONVIOL == 1 18 tab NONVIOL 19 20 gen decade = floor(year / 10) * 10 21 tab decade 22 23 collapse (count) count_nv_resistance = NONVIOL, by(decade) 24 list 25 26 graph bar (sum) count_nv_resistance /// 27 ,over(decade) /// 28 title("Non-violent Resistance over Decades") /// 29 bar(1, color(blue)) /// 30 graphregion(color(white)) /// 31 bgcolor(white) /// 32 legend(off) 33 34 * Create descriptive statistics for years in the count of uturns 35 36 use "C:\Users\alexa\OneDrive\Desktop\V-Dem14+U-Turn.dta", clear 37 38 keep if uturn == 1 39 keep if year <= 2019 40 41 tab uturn 42 43 gen decade = floor(year / 10) * 10 44 tab decade 45 46 collapse (count) U_Turn = uturn, by(decade) 47 list 48 49 graph bar (sum) U_Turn /// 50 ,over(decade) /// 51 title("U-Turns over Decades") /// 52 bar(1, color(blue)) /// 53 graphregion(color(white)) /// 54 bgcolor(white) /// 55 legend(off) 56 57 ***** 58 59 merge m:1 country_name year using "C:\Users\alexa\OneDrive\Desktop\V-Dem14+U-Turn.dta" // merge NACO with V-Dem+U-Turn 60 61 ***** 62 63 * create new variable that calculates the year when last protest happened 64 65 by country_name (year), sort: gen last_nonviol = cond(NONVIOL==1, year, 0) if _n == 1 66 by country_name (year): replace last_nonviol = /// Page 1 Results_MAThesis - Printed on 20.05.2024 13:01:22 67 cond((NONVIOL==1) & (year > last_nonviol[_n-1]), year, last_nonviol[_n-1]) if _n > 1 68 69 tab last_nonviol 70 gen years_since_last_nonviol = year - last_nonviol if aut_ep == 1 71 replace years_since_last_nonviol = 0 if last_nonviol == 0 72 replace years_since_last_nonviol = 0 if missing(years_since_last_nonviol) 73 74 rename years_since_last_nonviol NVR_last 75 tab NVR_last 76 77 * create protest knowledge variable based on NVR_last 78 79 gen resistance_knowledge = . 80 replace resistance_knowledge = 0 if NVR_last == 0 81 replace resistance_knowledge = 1 if NVR_last == 1 82 replace resistance_knowledge = 0.9 if NVR_last == 2 83 replace resistance_knowledge = 0.8 if NVR_last == 3 84 replace resistance_knowledge = 0.7 if NVR_last == 4 85 replace resistance_knowledge = 0.6 if NVR_last == 5 86 replace resistance_knowledge = 0.5 if NVR_last == 6 87 replace resistance_knowledge = 0.4 if NVR_last == 7 88 replace resistance_knowledge = 0.3 if NVR_last == 8 89 replace resistance_knowledge = 0.2 if NVR_last == 9 90 replace resistance_knowledge = 0.1 if NVR_last == 10 91 replace resistance_knowledge = 0 if NVR_last > 10 92 93 * renaming variables 94 95 rename v2x_polyarchy EDI 96 rename v2cacamps Polariz 97 ***** 98 99 * Prepare control variables 100 drop _merge 101 merge m:1 country_text_id year using "C:\Users\alexa\OneDrive\Desktop\mpd2020.dta" // insert GDP per capita from Maddison Project 102 gen gdppc_log = ln(gdppc) 103 104 gen e_pop_log = ln(e_pop) 105 106 107 collapse uturn NVR resistance_knowledge VR gdppc gdppc_log EDI v2x_jucon v2xlg_legcon Polariz e_peaveduc e_pop_log, by (country_id year) // collapse to avoid repeated time values within panel 108 tsset country_id year 109 110 tssmooth ma ma_edi = EDI, window(5) 111 112 ************* 113 * Running logit model 1900-2020 * 114 115 * Preparation 116 collapse uturn NVR resistance_knowledge VR gdppc gdppc_log ma_edi v2x_jucon v2xlg_legcon Polariz e_peaveduc e_pop_log, by (country_id year) // collapse to avoid repeated time values within panel 117 tsset country_id year // set panel structure 118 119 * Start with (most refined) Model 5: all navco + all control variables 120 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 121 estimates store m5 122 gen in_model_5 = e(sample) 123 estat ic 124 125 * Bivariate Model 1 126 logit uturn l.NVR i.year if in_model_5==1, vce(cluster country_id) 127 estimates store m1 128 estat ic 129 130 * Multivariate Model 2: add second NAVCO variable Page 2 Results_MAThesis - Printed on 20.05.2024 13:01:22 131 logit uturn l.NVR l.resistance_knowledge i.year if in_model_5==1, vce(cluster country_id) 132 estimates store m2 133 estat ic 134 135 * Model 3: all navco variables 136 logit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5==1, vce(cluster country_id) 137 estimates store m3 138 estat ic 139 140 * Model 4: all nacvo variables + democracy & gdp 141 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5==1, vce( cluster country_id) 142 estimates store m4 143 estat ic 144 145 * Create Output to Word 146 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020.rtf", se r2 ar2 replace 147 148 * Marginsplot for NVR on U-Turn 149 logit uturn l.NVR i.year if in_model_5==1, vce(cluster country_id) 150 margins, at(l.NVR=(0(1)2)) // margins for NVR 151 marginsplot, name(logit_plot, replace) xtitle("l.NVR") ytitle("Pr(uturn)") 152 153 * Marginsplot for NVR stock knowledge 154 logit uturn l.resistance_knowledge i.year if in_model_5==1, vce(cluster country_id) 155 margins, at(l.resistance_knowledge=(0(0.1)1)) // margins for resistance_knowledge 156 marginsplot, name(logit_plot, replace) xtitle("l.NVR Stock Knowledge") ytitle("Pr(uturn)") 157 158 ************************ DIAGNOSTICS ********************************************************************** 159 quietly regress uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l. v2xlg_legcon l.Polariz l.e_peaveduc l.e_pop_log 160 161 estat vif // check for multicollinearity 162 163 quietly logit uturn NVR resistance_knowledge VR gdppc_log ma_edi v2x_jucon v2xlg_legcon Polariz e_peaveduc e_pop_log // check for influential outliers 164 predict p 165 predict db, dbeta 166 scatter db p 167 168 sort db 169 browse country_id db 170 171 keep if db < .05 172 173 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) // run model without influential outliers 174 estimates store m5 175 gen in_model_5_outliers = e(sample) 176 estat ic 177 178 logit uturn l.NVR i.year if in_model_5_outliers==1, vce(cluster country_id) 179 estimates store m1 180 estat ic 181 182 logit uturn l.NVR l.resistance_knowledge i.year if in_model_5_outliers==1, vce(cluster country_id) 183 estimates store m2 184 estat ic 185 186 logit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5_outliers==1, vce(cluster country_id) 187 estimates store m3 188 estat ic 189 190 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5_outliers== 1, vce(cluster country_id) Page 3 Results_MAThesis - Printed on 20.05.2024 13:01:22 191 estimates store m4 192 estat ic 193 194 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020_outliers.rtf", se r2 ar2 replace 195 196 ********* ROBUSTNESS CHECKS 197 198 *** 2-way fixed effects *** 199 200 * Start with (most refined) Model 5: all navco + all control variables 201 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.country i.year, vce(cluster country_id) 202 estimates store m5 203 gen in_model_5 = e(sample) 204 estat ic 205 206 * Bivariate Model 1 207 logit uturn l.NVR i.country i.year if in_model_5==1, vce(cluster country_id) 208 estimates store m1 209 estat ic 210 211 * Multivariate Model 2: add second NAVCO variable 212 logit uturn l.NVR l.resistance_knowledge i.country i.year if in_model_5==1, vce(cluster country_id) 213 estimates store m2 214 estat ic 215 216 * Model 3: all navco variables 217 logit uturn l.NVR l.resistance_knowledge i.country i.year l.VR if in_model_5==1, vce(cluster country_id) 218 estimates store m3 219 estat ic 220 221 * Model 4: all nacvo variables + democracy & gdp 222 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.country i.year if in_model_5 ==1, vce(cluster country_id) 223 estimates store m4 224 estat ic 225 226 * Create Output to Word 227 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020_fe.rtf", se r2 ar2 replace 228 229 *** Model with alternative specification: Violent Resistance *** 230 231 * create new variable that calculates the year when last protest happened 232 233 by country_name (year), sort: gen last_viol = cond(VIOL==1, year, 0) if _n == 1 234 by country_name (year): replace last_viol = /// 235 cond((VIOL==1) & (year > last_viol[_n-1]), year, last_viol[_n-1]) if _n > 1 236 237 tab last_viol 238 gen years_since_last_viol = year - last_viol if aut_ep == 1 239 replace years_since_last_viol = 0 if last_viol == 0 240 replace years_since_last_viol = 0 if missing(years_since_last_viol) 241 242 rename years_since_last_viol VR_last 243 tab VR_last 244 245 * create protest knowledge variable based on NVR_last 246 247 gen resistance_knowledge = . 248 replace resistance_knowledge = 0 if VR_last == 0 249 replace resistance_knowledge = 1 if VR_last == 1 250 replace resistance_knowledge = 0.9 if VR_last == 2 251 replace resistance_knowledge = 0.8 if VR_last == 3 252 replace resistance_knowledge = 0.7 if VR_last == 4 253 replace resistance_knowledge = 0.6 if VR_last == 5 254 replace resistance_knowledge = 0.5 if VR_last == 6 255 replace resistance_knowledge = 0.4 if VR_last == 7 Page 4 Results_MAThesis - Printed on 20.05.2024 13:01:22 256 replace resistance_knowledge = 0.3 if VR_last == 8 257 replace resistance_knowledge = 0.2 if VR_last == 9 258 replace resistance_knowledge = 0.1 if VR_last == 10 259 replace resistance_knowledge = 0 if VR_last > 10 260 261 * Preparation 262 collapse uturn VR resistance_knowledge NVR gdppc gdppc_log EDI v2x_jucon v2xlg_legcon Polariz e_peaveduc e_pop_log, by (country_id year) // collapse to avoid repeated time values within panel 263 tsset country_id year // set panel structure 264 265 tssmooth ma ma_edi = EDI, window(5) 266 267 268 drop if year > 1993 269 270 * Start with (most refined) Model 5: all navco + all control variables 271 logit uturn l.VR l.resistance_knowledge l.NVR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 272 estimates store m5 273 gen in_model_5_violent = e(sample) 274 estat ic 275 276 * Bivariate Model 1 277 logit uturn l.VR i.year if in_model_5_violent==1, vce(cluster country_id) 278 estimates store m1 279 estat ic 280 281 * Multivariate Model 2: add second NAVCO variable 282 logit uturn l.VR l.resistance_knowledge i.year if in_model_5_violent==1, vce(cluster country_id) 283 estimates store m2 284 estat ic 285 286 * Model 3: all navco variables 287 logit uturn l.VR l.resistance_knowledge l.NVR i.year if in_model_5_violent==1, vce(cluster country_id) 288 estimates store m3 289 estat ic 290 291 * Model 4: all nacvo variables + democracy & gdp 292 logit uturn l.VR l.resistance_knowledge l.NVR l.gdppc_log l.ma_edi i.year if in_model_5_violent== 1, vce(cluster country_id) 293 estimates store m4 294 estat ic 295 296 * Create Output to Word 297 esttab m1 m2 m3 m4 m5 using "logitresults_1900-1994_violent.rtf", se r2 ar2 replace 298 299 *** Model with alternative specification: Change threshold for NVR pre-autocratization to 20 years *** 300 301 * create new variable that calculates the year when last protest happened 302 303 by country_name (year), sort: gen last_nonviol = cond(NONVIOL==1, year, 0) if _n == 1 304 by country_name (year): replace last_nonviol = /// 305 cond((NONVIOL==1) & (year > last_nonviol[_n-1]), year, last_nonviol[_n-1]) if _n > 1 306 307 tab last_nonviol 308 gen years_since_last_nonviol = year - last_nonviol if aut_ep == 1 309 replace years_since_last_nonviol = 0 if last_nonviol == 0 310 replace years_since_last_nonviol = 0 if missing(years_since_last_nonviol) 311 312 rename years_since_last_nonviol NVR_last 313 tab NVR_last 314 315 * create protest knowledge variable based on NVR_last 316 317 gen resistance_knowledge = . 318 replace resistance_knowledge = 0 if NVR_last == 0 Page 5 Results_MAThesis - Printed on 20.05.2024 13:01:23 319 replace resistance_knowledge = 1 if NVR_last == 1 320 replace resistance_knowledge = 0.95 if NVR_last == 2 321 replace resistance_knowledge = 0.9 if NVR_last == 3 322 replace resistance_knowledge = 0.85 if NVR_last == 4 323 replace resistance_knowledge = 0.8 if NVR_last == 5 324 replace resistance_knowledge = 0.75 if NVR_last == 6 325 replace resistance_knowledge = 0.7 if NVR_last == 7 326 replace resistance_knowledge = 0.65 if NVR_last == 8 327 replace resistance_knowledge = 0.6 if NVR_last == 9 328 replace resistance_knowledge = 0.55 if NVR_last == 10 329 replace resistance_knowledge = 0.5 if NVR_last == 11 330 replace resistance_knowledge = 0.45 if NVR_last == 12 331 replace resistance_knowledge = 0.4 if NVR_last == 13 332 replace resistance_knowledge = 0.35 if NVR_last == 14 333 replace resistance_knowledge = 0.3 if NVR_last == 15 334 replace resistance_knowledge = 0.25 if NVR_last == 16 335 replace resistance_knowledge = 0.2 if NVR_last == 17 336 replace resistance_knowledge = 0.15 if NVR_last == 18 337 replace resistance_knowledge = 0.1 if NVR_last == 19 338 replace resistance_knowledge = 0.05 if NVR_last == 20 339 replace resistance_knowledge = 0 if NVR_last > 20 340 341 * Start with (most refined) Model 5: all navco + all control variables 342 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 343 estimates store m5 344 gen in_model_5 = e(sample) 345 estat ic 346 347 * Bivariate Model 1 348 logit uturn l.NVR i.year if in_model_5==1, vce(cluster country_id) 349 estimates store m1 350 estat ic 351 352 * Multivariate Model 2: add second NAVCO variable 353 logit uturn l.NVR l.resistance_knowledge i.year if in_model_5==1, vce(cluster country_id) 354 estimates store m2 355 estat ic 356 357 * Model 3: all navco variables 358 logit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5==1, vce(cluster country_id) 359 estimates store m3 360 estat ic 361 362 * Model 4: all nacvo variables + democracy & gdp 363 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5==1, vce( cluster country_id) 364 estimates store m4 365 estat ic 366 367 * Create Output to Word 368 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020.rtf", se r2 ar2 replace 369 370 *** Model with alternative specification: Change threshold for NVR pre-autocratization to 30 years *** 371 372 * create new variable that counts the protest episodes during the last 30 years if autocratization happened 373 374 rangestat (sum) NONVIOL, by(country_name) interval(year -30 -1) 375 replace NONVIOL_sum = . if aut_ep != 1 376 replace NONVIOL_sum = 0 if missing(NONVIOL_sum) 377 378 rename NONVIOL_sum NVR_history 379 tab NVR_history 380 381 * create new variable that calculates the year when last protest happened 382 Page 6 Results_MAThesis - Printed on 20.05.2024 13:01:23 383 by country_name (year), sort: gen last_nonviol = cond(NONVIOL==1, year, 0) if _n == 1 384 by country_name (year): replace last_nonviol = /// 385 cond((NONVIOL==1) & (year > last_nonviol[_n-1]), year, last_nonviol[_n-1]) if _n > 1 386 387 tab last_nonviol 388 gen years_since_last_nonviol = year - last_nonviol if aut_ep == 1 389 replace years_since_last_nonviol = 0 if last_nonviol == 0 390 replace years_since_last_nonviol = 0 if missing(years_since_last_nonviol) 391 392 rename years_since_last_nonviol NVR_last 393 tab NVR_last 394 395 * create protest knowledge variable based on NVR_last 396 397 gen resistance_knowledge = . 398 replace resistance_knowledge = 0 if NVR_last == 0 399 replace resistance_knowledge = 1 if NVR_last == 1 400 replace resistance_knowledge = 0.966 if NVR_last == 2 401 replace resistance_knowledge = 0.933 if NVR_last == 3 402 replace resistance_knowledge = 0.9 if NVR_last == 4 403 replace resistance_knowledge = 0.866 if NVR_last == 5 404 replace resistance_knowledge = 0.833 if NVR_last == 6 405 replace resistance_knowledge = 0.8 if NVR_last == 7 406 replace resistance_knowledge = 0.766 if NVR_last == 8 407 replace resistance_knowledge = 0.733 if NVR_last == 9 408 replace resistance_knowledge = 0.7 if NVR_last == 10 409 replace resistance_knowledge = 0.666 if NVR_last == 11 410 replace resistance_knowledge = 0.633 if NVR_last == 12 411 replace resistance_knowledge = 0.6 if NVR_last == 13 412 replace resistance_knowledge = 0.566 if NVR_last == 14 413 replace resistance_knowledge = 0.533 if NVR_last == 15 414 replace resistance_knowledge = 0.5 if NVR_last == 16 415 replace resistance_knowledge = 0.466 if NVR_last == 17 416 replace resistance_knowledge = 0.433 if NVR_last == 18 417 replace resistance_knowledge = 0.4 if NVR_last == 19 418 replace resistance_knowledge = 0.366 if NVR_last == 20 419 replace resistance_knowledge = 0.333 if NVR_last == 21 420 replace resistance_knowledge = 0.3 if NVR_last == 22 421 replace resistance_knowledge = 0.266 if NVR_last == 23 422 replace resistance_knowledge = 0.233 if NVR_last == 24 423 replace resistance_knowledge = 0.2 if NVR_last == 25 424 replace resistance_knowledge = 0.166 if NVR_last == 26 425 replace resistance_knowledge = 0.133 if NVR_last == 27 426 replace resistance_knowledge = 0.1 if NVR_last == 28 427 replace resistance_knowledge = 0.066 if NVR_last == 29 428 replace resistance_knowledge = 0.033 if NVR_last == 30 429 replace resistance_knowledge = 0 if NVR_last > 30 430 431 * Start with (most refined) Model 5: all navco + all control variables 432 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 433 estimates store m5 434 gen in_model_5 = e(sample) 435 estat ic 436 437 * Bivariate Model 1 438 logit uturn l.NVR i.year if in_model_5==1, vce(cluster country_id) 439 estimates store m1 440 estat ic 441 442 * Multivariate Model 2: add second NAVCO variable 443 logit uturn l.NVR l.resistance_knowledge i.year if in_model_5==1, vce(cluster country_id) 444 estimates store m2 445 estat ic 446 447 * Model 3: all navco variables 448 logit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5==1, vce(cluster country_id) 449 estimates store m3 Page 7 Results_MAThesis - Printed on 20.05.2024 13:01:23 450 estat ic 451 452 * Model 4: all nacvo variables + democracy & gdp 453 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5==1, vce( cluster country_id) 454 estimates store m4 455 estat ic 456 457 * Create Output to Word 458 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020_mod.rtf", se r2 ar2 replace 459 460 *** Model with alternative specification: Change threshold for NVR pre-autocratization to 64 years (64 is the last entry point for the NVR_last variable) *** 461 462 * create new variable that calculates the year when last protest happened 463 464 by country_name (year), sort: gen last_nonviol = cond(NONVIOL==1, year, 0) if _n == 1 465 by country_name (year): replace last_nonviol = /// 466 cond((NONVIOL==1) & (year > last_nonviol[_n-1]), year, last_nonviol[_n-1]) if _n > 1 467 468 tab last_nonviol 469 gen years_since_last_nonviol = year - last_nonviol if aut_ep == 1 470 replace years_since_last_nonviol = 0 if last_nonviol == 0 471 replace years_since_last_nonviol = 0 if missing(years_since_last_nonviol) 472 473 rename years_since_last_nonviol NVR_last 474 tab NVR_last 475 476 * create protest knowledge variable based on NVR_last 477 gen resistance_knowledge = . 478 replace resistance_knowledge = 1 if NVR_last == 1 479 replace resistance_knowledge = 0 if NVR_last == 0 480 replace resistance_knowledge = 0.9844 if NVR_last == 2 481 replace resistance_knowledge = 0.9688 if NVR_last == 3 482 replace resistance_knowledge = 0.9531 if NVR_last == 4 483 replace resistance_knowledge = 0.9375 if NVR_last == 5 484 replace resistance_knowledge = 0.9219 if NVR_last == 6 485 replace resistance_knowledge = 0.9063 if NVR_last == 7 486 replace resistance_knowledge = 0.8906 if NVR_last == 8 487 replace resistance_knowledge = 0.8750 if NVR_last == 9 488 replace resistance_knowledge = 0.8594 if NVR_last == 10 489 replace resistance_knowledge = 0.8438 if NVR_last == 11 490 replace resistance_knowledge = 0.8281 if NVR_last == 12 491 replace resistance_knowledge = 0.8125 if NVR_last == 13 492 replace resistance_knowledge = 0.7969 if NVR_last == 14 493 replace resistance_knowledge = 0.7813 if NVR_last == 15 494 replace resistance_knowledge = 0.7656 if NVR_last == 16 495 replace resistance_knowledge = 0.7500 if NVR_last == 17 496 replace resistance_knowledge = 0.7344 if NVR_last == 18 497 replace resistance_knowledge = 0.7188 if NVR_last == 19 498 replace resistance_knowledge = 0.7031 if NVR_last == 20 499 replace resistance_knowledge = 0.6875 if NVR_last == 21 500 replace resistance_knowledge = 0.6719 if NVR_last == 22 501 replace resistance_knowledge = 0.6563 if NVR_last == 23 502 replace resistance_knowledge = 0.6406 if NVR_last == 24 503 replace resistance_knowledge = 0.6250 if NVR_last == 25 504 replace resistance_knowledge = 0.6094 if NVR_last == 26 505 replace resistance_knowledge = 0.5938 if NVR_last == 27 506 replace resistance_knowledge = 0.5781 if NVR_last == 28 507 replace resistance_knowledge = 0.5625 if NVR_last == 29 508 replace resistance_knowledge = 0.5469 if NVR_last == 30 509 replace resistance_knowledge = 0.5313 if NVR_last == 31 510 replace resistance_knowledge = 0.5156 if NVR_last == 32 511 replace resistance_knowledge = 0.5000 if NVR_last == 33 512 replace resistance_knowledge = 0.4844 if NVR_last == 34 513 replace resistance_knowledge = 0.4688 if NVR_last == 35 514 replace resistance_knowledge = 0.4531 if NVR_last == 36 515 replace resistance_knowledge = 0.4375 if NVR_last == 37 Page 8 Results_MAThesis - Printed on 20.05.2024 13:01:23 516 replace resistance_knowledge = 0.4219 if NVR_last == 38 517 replace resistance_knowledge = 0.4063 if NVR_last == 39 518 replace resistance_knowledge = 0.3906 if NVR_last == 40 519 replace resistance_knowledge = 0.3750 if NVR_last == 41 520 replace resistance_knowledge = 0.3594 if NVR_last == 42 521 replace resistance_knowledge = 0.3438 if NVR_last == 43 522 replace resistance_knowledge = 0.3281 if NVR_last == 44 523 replace resistance_knowledge = 0.3125 if NVR_last == 45 524 replace resistance_knowledge = 0.2969 if NVR_last == 46 525 replace resistance_knowledge = 0.2813 if NVR_last == 47 526 replace resistance_knowledge = 0.2656 if NVR_last == 48 527 replace resistance_knowledge = 0.2500 if NVR_last == 49 528 replace resistance_knowledge = 0.2344 if NVR_last == 50 529 replace resistance_knowledge = 0.2188 if NVR_last == 51 530 replace resistance_knowledge = 0.2031 if NVR_last == 52 531 replace resistance_knowledge = 0.1875 if NVR_last == 53 532 replace resistance_knowledge = 0.1719 if NVR_last == 54 533 replace resistance_knowledge = 0.1563 if NVR_last == 55 534 replace resistance_knowledge = 0.1406 if NVR_last == 56 535 replace resistance_knowledge = 0.1250 if NVR_last == 57 536 replace resistance_knowledge = 0.1094 if NVR_last == 58 537 replace resistance_knowledge = 0.0938 if NVR_last == 59 538 replace resistance_knowledge = 0.0781 if NVR_last == 60 539 replace resistance_knowledge = 0.0625 if NVR_last == 61 540 replace resistance_knowledge = 0.0469 if NVR_last == 62 541 replace resistance_knowledge = 0.0313 if NVR_last == 63 542 replace resistance_knowledge = 0.0156 if NVR_last == 64 543 544 * Start with (most refined) Model 5: all navco + all control variables 545 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 546 estimates store m5 547 gen in_model_5 = e(sample) 548 estat ic 549 550 * Bivariate Model 1 551 logit uturn l.NVR i.year if in_model_5==1, vce(cluster country_id) 552 estimates store m1 553 estat ic 554 555 * Multivariate Model 2: add second NAVCO variable 556 logit uturn l.NVR l.resistance_knowledge i.year if in_model_5==1, vce(cluster country_id) 557 estimates store m2 558 estat ic 559 560 * Model 3: all navco variables 561 logit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5==1, vce(cluster country_id) 562 estimates store m3 563 estat ic 564 565 * Model 4: all nacvo variables + democracy & gdp 566 logit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5==1, vce( cluster country_id) 567 estimates store m4 568 estat ic 569 570 * Create Output to Word 571 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020_mod64.rtf", se r2 ar2 replace 572 573 * Show distribution of NVR_last variable 574 hist NVR_last if NVR_last > 0, freq 575 576 *** Test with data that focuses on REGIMECHANGE *** 577 578 egen NVR_regchange = total(NONVIOL * (REGCHANGE == 1)), by(country_name year) 579 egen VR_regchange = total(VIOL * (REGCHANGE == 1)), by(country_name year) 580 581 * Generate the variable NVR_history Page 9 Results_MAThesis - Printed on 20.05.2024 13:01:23 582 rangestat (sum) NONVIOL if REGCHANGE == 1, by(country_name) interval(year -10 -1) 583 replace NONVIOL_sum = . if aut_ep !=1 584 replace NONVIOL_sum = 0 if missing(NONVIOL_sum) 585 rename NONVIOL_sum NVR_history_regchange 586 tab NVR_history_regchange 587 588 * Generate the variable last_nonviol 589 by country_name (year), sort: gen last_nonviol = cond((NONVIOL==1 & REGCHANGE == 1), year, 0) if _n == 1 590 by country_name (year): replace last_nonviol = /// 591 cond((NONVIOL==1 & REGCHANGE == 1) & (year > last_nonviol[_n-1]), year, last_nonviol[_n-1]) if _n > 1 592 593 tab last_nonviol 594 gen years_since_last_nonviol = year - last_nonviol if aut_ep == 1 595 replace years_since_last_nonviol = 0 if last_nonviol == 0 596 replace years_since_last_nonviol = 0 if missing(years_since_last_nonviol) 597 598 rename years_since_last_nonviol NVR_last_regchange 599 tab NVR_last_regchange 600 601 * create protest knowledge variable based on NVR_last 602 603 gen resistance_knowledge_regchange = . 604 replace resistance_knowledge = 0 if NVR_last_regchange == 0 605 replace resistance_knowledge = 1 if NVR_last_regchange == 1 606 replace resistance_knowledge = 0.9 if NVR_last_regchange == 2 607 replace resistance_knowledge = 0.8 if NVR_last_regchange == 3 608 replace resistance_knowledge = 0.7 if NVR_last_regchange == 4 609 replace resistance_knowledge = 0.6 if NVR_last_regchange == 5 610 replace resistance_knowledge = 0.5 if NVR_last_regchange == 6 611 replace resistance_knowledge = 0.4 if NVR_last_regchange == 7 612 replace resistance_knowledge = 0.3 if NVR_last_regchange == 8 613 replace resistance_knowledge = 0.2 if NVR_last_regchange == 9 614 replace resistance_knowledge = 0.1 if NVR_last_regchange == 10 615 replace resistance_knowledge = 0 if NVR_last > 10 616 617 * renaming variables 618 619 rename v2x_polyarchy EDI 620 rename v2cacamps Polariz 621 ***** 622 623 * Prepare control variables 624 drop _merge 625 merge m:1 country_text_id year using "C:\Users\alexa\OneDrive\Desktop\mpd2020.dta" // insert GDP per capita from Maddison Project 626 gen gdppc_log = ln(gdppc) 627 628 gen e_pop_log = ln(e_pop) 629 630 631 collapse uturn NVR_regchange resistance_knowledge_regchange VR_regchange gdppc gdppc_log EDI v2x_jucon v2xlg_legcon Polariz e_peaveduc e_pop_log, by (country_id year) // collapse to avoid repeated time values within panel 632 tsset country_id year 633 634 tssmooth ma ma_edi = EDI, window(5) 635 636 637 * Run logit 638 * Preparation 639 collapse uturn NVR_regchange resistance_knowledge_regchange VR_regchange gdppc_log ma_edi v2x_jucon v2xlg_legcon Polariz e_peaveduc e_pop_log, by (country_id year) // collapse to avoid repeated time values within panel 640 tsset country_id year // set panel structure 641 642 * Start with (most refined) Model 5: all navco + all control variables Page 10 Results_MAThesis - Printed on 20.05.2024 13:01:23 643 logit uturn l.NVR_regchange l.resistance_knowledge_regchange l.VR_regchange l.gdppc_log l.ma_edi l .v2x_jucon l.v2xlg_legcon l.Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 644 estimates store m5 645 gen in_model_5_regchange = e(sample) 646 estat ic 647 648 * Bivariate Model 1 649 logit uturn l.NVR_regchange i.year if in_model_5_regchange==1, vce(cluster country_id) 650 estimates store m1 651 estat ic 652 653 * Multivariate Model 2: add second NAVCO variable 654 logit uturn l.NVR_regchange l.resistance_knowledge_regchange i.year if in_model_5_regchange==1, vce(cluster country_id) 655 estimates store m2 656 estat ic 657 658 * Model 3: all navco variables 659 logit uturn l.NVR_regchange l.resistance_knowledge_regchange l.VR_regchange i.year if in_model_5_regchange==1, vce(cluster country_id) 660 estimates store m3 661 estat ic 662 663 * Model 4: all nacvo variables + democracy & gdp 664 logit uturn l.NVR_regchange l.resistance_knowledge_regchange l.VR_regchange l.gdppc_log l.ma_edi i .year if in_model_5_regchange==1, vce(cluster country_id) 665 estimates store m4 666 estat ic 667 668 * Create Output to Word 669 esttab m1 m2 m3 m4 m5 using "logitresults_1900-2020_regchange.rtf", se r2 ar2 replace 670 671 672 673 *** Last robustness: Probit *** 674 675 * Start with (most refined) Model 5: all navco + all control variables 676 probit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 677 estimates store m5 678 gen in_model_5 = e(sample) 679 estat ic 680 681 * Bivariate Model 1 682 probit uturn l.NVR i.year if in_model_5==1, vce(cluster country_id) 683 estimates store m1 684 estat ic 685 686 * Multivariate Model 2: add second NAVCO variable 687 probit uturn l.NVR l.resistance_knowledge i.year if in_model_5==1, vce(cluster country_id) 688 estimates store m2 689 estat ic 690 691 * Model 3: all navco variables 692 probit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5==1, vce(cluster country_id) 693 estimates store m3 694 estat ic 695 696 * Model 4: all nacvo variables + democracy & gdp 697 probit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5==1, vce( cluster country_id) 698 estimates store m4 699 estat ic 700 701 * Create Output to Word 702 esttab m1 m2 m3 m4 m5 using "probitresults_1900-2020.rtf", se r2 ar2 replace 703 704 Page 11 Results_MAThesis - Printed on 20.05.2024 13:01:23 705 **************** Do the same but only for year > 1994************************** 706 drop if year < 1994 707 708 * Start with (most refined) Model 5: all navco + all control variables 709 probit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 710 estimates store m5 711 gen in_model_5_post = e(sample) 712 estat ic 713 714 * Bivariate Model 1 715 probit uturn l.NVR i.year if in_model_5_post==1, vce(cluster country_id) 716 estimates store m1 717 estat ic 718 719 * Multivariate Model 2: add second NAVCO variable 720 probit uturn l.NVR l.resistance_knowledge i.year if in_model_5_post==1, vce(cluster country_id) 721 estimates store m2 722 estat ic 723 724 * Model 3: all navco variables 725 probit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5_post==1, vce(cluster country_id) 726 estimates store m3 727 estat ic 728 729 * Model 4: all nacvo variables + democracy & gdp 730 probit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5_post==1, vce(cluster country_id) 731 estimates store m4 732 estat ic 733 734 esttab m1 m2 m3 m4 m5 using "probitresults_1994-2020.rtf", se r2 ar2 replace 735 736 **************** Do the same but only for year < 1994************************** 737 drop if year >= 1994 738 739 * Start with (most refined) Model 5: all navco + all control variables 740 probit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi l.v2x_jucon l.v2xlg_legcon l. Polariz l.e_peaveduc l.e_pop_log i.year, vce(cluster country_id) 741 estimates store m5 742 gen in_model_5_pre = e(sample) 743 estat ic 744 745 * Bivariate Model 1 746 probit uturn l.NVR i.year if in_model_5_pre==1, vce(cluster country_id) 747 estimates store m1 748 estat ic 749 750 * Multivariate Model 2: add second NAVCO variable 751 probit uturn l.NVR l.resistance_knowledge i.year if in_model_5_pre==1, vce(cluster country_id) 752 estimates store m2 753 estat ic 754 755 * Model 3: all navco variables 756 probit uturn l.NVR l.resistance_knowledge l.VR i.year if in_model_5_pre==1, vce(cluster country_id) 757 estimates store m3 758 estat ic 759 760 * Model 4: all nacvo variables + democracy & gdp 761 probit uturn l.NVR l.resistance_knowledge l.VR l.gdppc_log l.ma_edi i.year if in_model_5_pre==1, vce(cluster country_id) 762 estimates store m4 763 estat ic 764 765 esttab m1 m2 m3 m4 m5 using "probitresults_1900-1993.rtf", se r2 ar2 replace Page 12