2.3.1 The Timing of Community-Initiated Militia Formation

When do militias form? The timing of community-initiated militia formation refers to the strategic context in which militias form. The strategic context shapes the expectations of community residents and elites about future violence and thus their preferences about what actions to take.

Community residents form militias when insurgent and incumbent armies enter a military stalemate at the local level, which I refer to as a “community-empowering stalemate.” A military stalemate can be defined as “a situation where neither combatant is able to make noteworthy advances on the battlefield due to the strength of the opposing side, and neither side believes that the situation will improve in the near future” (Walter Reference Walter1997, 347). This is a common definition of a military stalemate related to the balance of military capabilities (Licklider Reference Licklider1993, 309; Schulhofer-Wohl Reference Schulhofer-Wohl2019, 7). A stalemate on the district or provincial level can coincide with a stalemate on the national level, although it does not have to. Usually, stalemates characterize the strategic situation in which governments and insurgents find themselves on the national level – a situation that prevents decisive victory of one side (Fearon Reference Fearon2004, 276).Footnote ⁵ However, given the disjunction between war dynamics on the local and the national levels (Kalyvas Reference Kalyvas2003; Lubkemann Reference Lubkemann2005), local military stalemates may exist regionally when there is no stalemate nationally, and in turn a national stalemate may not reflect the strategic situation at regional or local levels.

There are two possible types of military stalemates. A military stalemate arises either when territorial control is relatively constant, which means that the level of violence against civilians by the incumbent is relatively low, or when territorial control changes frequently and incumbent violence is high (Kalyvas Reference Kalyvas2006, 207). In the first case, incumbents – the armed actor in control – use civil war violence selectively to punish those civilians who they suspect of supporting the enemy. In this way, armed groups seek to ensure that civilians collaborate and provide them with the necessary intelligence about those who support the enemy (Kalyvas Reference Kalyvas2006). Incumbents attempt to expand their base of popular support, as occurred in El Salvador with the decrease in state violence against civilians and the provision of basic services (Wood Reference Wood2003, 132, 150). Thus, a “local political equilibrium” between the El Salvadoran government and the Farabundo Martí National Liberation Front (FMLN) emerged (Wood Reference Wood2003, 150).Footnote ⁶ In such situations war economies flourish, as the warring parties seek to reap benefits from the ongoing war, as happened in Liberia when a military stalemate arose two years after President Samuel Doe’s death in September 1990 (Ellis Reference Ellis1999, 25, 169; cf. Le Billon Reference Le Billon2001).

The second type of military stalemate is particularly conducive to the bottom-up mobilization of militias as it allows for more civilian agency. In this second type, the military stalemate results in a war of attrition, during which both sides try to hurt the other as much as possible. After the war in Chechnya, for example, evolved into a stalemate in 2002, “Chechen fighters were ‘no longer able to confront Russian troops head-on, but they remain determined to inflict as much pain as possible in the name of Chechen independence’” (Kalyvas Reference Kalyvas2006, 164n33, citing a report from The New York Times). In this context, states often adopt scorched-earth tactics as their main counterinsurgency measure to kill potential collaborators indiscriminately and destroy the insurgents’ popular support base. This was the case, for instance, in Mozambique during the anticolonial war, when the Portuguese colonial regime indiscriminately targeted potential collaborators of the liberation movement (Reis and Oliveira Reference Reis and Oliveira2012). The high cost of military stalemates has led analysts to contend that conflict parties should prefer negotiated settlements to wars of attrition (Zartman Reference Zartman1989). However, problems in credibly committing to settlements in an environment of fluctuating state capabilities lead to prolonged civil war. The state’s commitments are not credible, and the hope remains that one day one’s own side prevails over the other (Fearon Reference Fearon2004).

Though civilians must bear much of the violence in such costly military stalemates of the second type, these stalemates can be community-empowering. Community residents form militias when indiscriminate violence is high, the incumbent’s territorial control over a core area is relatively stable but constantly under threat, and peripheral areas change control often (see Kalyvas Reference Kalyvas2008a, 1059). In such a situation, incumbents find themselves virtually under siege; limited control over access roads leads to a lack of resources, which in turn leads to a failure of the incumbent to protect the population under its control.

Community-empowering military stalemates shape the preferences of communities and elites in the following ways. First, regarding community preferences, the population receives sufficient protection from neither the insurgent nor the incumbent forces. In such situations, neither cooperation with the dominant armed group nor defection to the enemy promises to improve civilians’ security. By fleeing to an area between zones of control, civilians lose access to both their farming plots and relief goods, which also endangers their food security (Justino Reference Justino2009). Civilians may think that they are better protected when they become combatants (Kalyvas and Kocher Reference Kalyvas and Kocher2007). However, joining the insurgents is unlikely as people become alienated with the armed group that inflicts violence upon them. Similarly, people may avoid joining the incumbent security forces either because of the army’s perpetration of violence against civilians or because joining does not promise protection as the fighting intensifies due to the lack of a military advantage. This difficult situation makes civilians vulnerable and war-weary, inducing a preference to form militias.

Second, from the administrative elite perspective, the population’s frequent displacement disrupts the control over people, the delivery of relief supplies, and the provision of state services. Thus the local administration’s primary goal is to protect and hold the population under its control. If the state security forces present in a community – these may include regular security forces or state-initiated militias – are too small or otherwise unable to fulfill this goal, and the administrative elites’ demands for regional or national military commando support are not met, community residents and their leaders look for alternative solutions. This provides an opportunity for alternative forms of security provision. Administrative elites are forced to tolerate or even actively support the formation of armed forces outside the regular state security apparatus and with substantial influence of community leaders.

In sum, community residents form militias when community-empowering military stalemates emerge. The population sees no better option for survival than organizing its own protection in the form of militias. In the absence of military support from regional or national commandos, administrative elites are forced to tolerate or even actively support the formation of such irregular forces.

2.3.2 The Location of Community-Initiated Militia Formation

Where do militias form? The strategic context of militia formation indicates that the experience of violence and state responses to that violence shape people’s expectations about future violence and, thus, their preferences about their own response to that violence. However, expectations are influenced not only by events that occur within one’s own community but also by events in neighboring communities. Communities’ decisions to form militias are often influenced by neighboring communities’ experience with militias. In fact, like other forms of violent collective action – riots, rebellions, revolutions – militias diffuse across geographic boundaries, be they defined by national borders or local boundaries.

I distinguish between initial diffusion – the spread of a militia – and sustained diffusion – the integration of a militia into the local security apparatus. Initial diffusion of militias depends on the relationship between the communities among which diffusion occurs. Ethnic, ideological, cultural, and historical bonds between communities facilitate initial diffusion, in particular when these bonds are reinforced by the (temporary) migration of “diffusion agents” – community residents and local elites – between these communities. Moreover, when militias reduce violence in neighboring communities, community residents learn that forming militias serves them better than staying passive. Sustained diffusion depends on the institutional context of the receiving community. When elites are relatively unified, local officials integrate militias into existing institutions. Relative unity among elites prevents community-initiated militias from becoming the private army of individual leaders. The community and its leaders need to trust and support a civilian-based armed group during war to facilitate its widespread mobilization. When elites are united in their understanding of the new armed group’s aims and strategies, community residents and local elites have more trust in the militia’s ability to curb violence, and hence support the new institution, ensuring its success.

The notion that revolutions do not occur in and of themselves but need to be nourished by the spread of ideas and actions across geographic and group boundaries is a recurring theme in many different episodes of collective action. Diffusion of ideas creates solidarity and fosters identification with others in the same situation.Footnote ⁷ However, the concept of diffusion applies to much broader areas of scholarly research. Scholars interested in the evolution of policies, institutions, and organizations use diffusion to explain how ideas and models of action – opinions, attitudes, and practices – spread or disseminate across geographic or group boundaries. This research agenda assumes that “there are enduring, cross-boundary dependencies in the evolution of policies and institutions” (Gleditsch and Ward Reference Gleditsch and Ward2006, 912). Social movement scholars have adopted this perspective to inquire into the conditions and mechanisms of how both established and innovative collective action frames and repertoires across communities, movements, and nation-states spread (Givan, Roberts, and Soule Reference Givan, Roberts and Soule2010, 4; Strang and Soule Reference Strang and Soule1998).Footnote ⁸

Concerning the study of armed conflict, the regional and transnational dimensions of civil wars in Africa and Asia in the 1990s and 2000s triggered renewed interest in the geographic spread of civil war. Scholars have long recognized the interdependence of civil wars and nationalist movements (Lake and Rothchild Reference Lake and Rothchild1998; Buhaug and Gleditsch Reference Buhaug and Gleditsch2008; Beissinger Reference Beissinger2002). The movement of insurgent groups, refugee populations, and weapons across porous borders regionalizes civil wars (Salehyan and Gleditsch Reference Salehyan2006; Salehyan Reference Serapião2007).Footnote ⁹ However, scholarly efforts to understand the process and precise mechanisms of how civil wars spread are relatively recent (Checkel Reference Checkel2013; Wood Reference Wood and Checkel2013, 233–34).

Dominated by studies of international politics, conflict scholars have analyzed the diffusion of violent collective action repertoires primarily as a transnational phenomenon rather than something that also occurs within states across community and group boundaries. By repertoires of collective action, I mean a set of forms of collective action available to a community or group (Tilly Reference Tilly1978; Gutiérrez-Sanín and Wood Reference Gutiérrez-Sanín and Wood2017; Ahram Reference Alexander2016). I focus on how and why a particular form of violent collective action (forming militias) spreads across communities during civil war. Militias can spread across community boundaries either through direct contacts traveling between communities or mediation through third actors (see Tarrow and McAdam Reference Tarrow, McAdam, Porta and Tarrow2005; Givan, Roberts, and Soule Reference Givan, Roberts and Soule2010). I call such actors who temporarily migrate and spread forms of collective action “diffusion agents.” I explain why diffusion is sometimes sustained, meaning that the community integrates the new form of violent collective action into its collective action repertoire. Communities – influenced by surrounding communities or diffusion agents – may try to adopt various forms of collective action during war; what is more compelling is whether these forms persist and become institutionalized, thereby influencing the dynamics of war across time and space.

Diffusion depends on the relation between two units and on the characteristics of the receiving unit. Research on the diffusion of civil wars suggests that ethnic or ideological bonds between units or refugee movements, including rebel combatants, to a neighboring state make the spread of violence more likely (Saideman Reference Saideman2002; Salehyan and Gleditsch Reference Salehyan2006; Salehyan Reference Serapião2007). Research on how policies and norms diffuse shows that diffusion depends on the coalition of actors in the receiving unit and their ability to translate new policies and norms into the local context (Acharya Reference Acharya2004; Gilardi Reference Gilardi2010). This means that once a bond is established or reinforced between two units, sustained diffusion occurs when the coalition of actors allows new forms to integrate into the local institutional context.

Applied to militia formation, militias initially diffuse when two communities are linked via ethnic, ideological, cultural, or historical bonds. A crucial mechanism that accounts for diffusion is the temporary or permanent migration of diffusion agents who reinforce such preexisting bonds (see Franzese and Hays Reference Franzese and Hays2008). Diffusion agents spread stories about militias’ success in decreasing violence against civilians, leading to initial diffusion to other communities. Through such success stories, community residents learn that active resistance rather than passivity helps to achieve security. Success can be understood as the perceived ability of a militia to create stability in a village so that community residents no longer have to flee to protect themselves from insurgent violence. In the diffusion literature, this mechanism is referred to as learning – the updating of previously held beliefs (Dobbin, Simmons, and Garrett Reference Dobbin, Simmons and Garrett2007, 462).Footnote ¹⁰

An empirical implication of this argument is that in order to learn from other communities, community residents have to define their situation as sufficiently similar to the one in which militias already exist to believe that a militia will also improve their situation (Tarrow and McAdam Reference Tarrow, McAdam, Porta and Tarrow2005). This may be influenced by the strategic situation in which the community finds itself. As outlined above, local military stalemates are necessary for militia formation; this implies that if the same conditions of local military stalemates exist in a neighboring community, then militias should spread.

Community linkages and militia success in neighboring communities are necessary for initial diffusion, but they are not sufficient for the sustained diffusion of community-initiated militias. A last condition thus relates to the characteristics of the receiving community. The institutional context of preexisting security forces and authority structures is crucial for understanding when sustained diffusion occurs. Divisions among social, political, and military elites may prevent militia formation in the long term. Militias may change the balance of power within a community and – when tied to certain elites – risk that one group exerts more power than another. This risk may lead some elites to prevent or stop the militias from forming or isolate the emerging militia. In contrast, when social, political and military elites are relatively unified, militias do not risk upsetting the balance of power among elites and can be incorporated into the existing security apparatus.

An empirical implication of this argument is that militias that change the balance of power among elites serve the interests of elites only and resemble a private army. Unless the commanding elites have independent income, these militias typically have access to few resources and recruits, which limits their activities and potentially their life span. In contrast, a militia with broad-based support from a range of different elites has access to more resources, including the most crucial one – new members.

In sum, I expect that initial diffusion of militias occurs in communities that are tied to each other by ethnic, ideological, cultural, and historical bonds, and where one community experiences militias improving stability for its residents. Diffusion agents who travel from one community to another reinforce such bonds and tell the story of militia success. Communities learn from each other when residents are convinced that their situation is sufficiently similar to that of a neighboring community, such as in the case of a local military stalemate. Sustained diffusion occurs when social, political, and military elites are relatively unified in the receiving community, which prevents militias from becoming private, semiprofessional armies.

2.3.3 The Process of Community-Initiated Militia Mobilization

How do militias mobilize their members? The arguments above account for the formation and diffusion of militias by identifying the factors that change people’s expectations about future violence and how they can respond to it. But once a community forms a militia, what makes the new organization grow? What are the mechanisms of social mobilization that operate in this instance? I suggest a causal path of militia mobilization that pays special attention to the social context. While the community level can explain militia formation and diffusion, it is at the individual and group levels that militia mobilization occurs (see Eck Reference Eck2010, 10). Militias successfully mobilize when the militia leadership offers potential members a means to manage uncertainty about future events – about how best to protect oneself and one’s family. Militias that appeal to social conventions that invoke collective meaning within the community help manage uncertainty and are therefore more successful in attracting members than other militias.

Many approaches to mobilization for violence rely on a rational cost–benefit approach, which implies that if the benefits of joining an armed group outweigh the cost, then joining is a likely outcome (Olson Reference Olson1965; Tullock Reference Tullock1971; Popkin Reference Popkin1979; Lichbach Reference Lichbach1995; Weinstein 2007; Humphreys and Weinstein Reference Humphreys and Weinstein2008).Footnote ¹¹ This assumes that probabilities of receiving such benefits and avoiding related costs can be estimated, and thus acted upon. However, the costs and benefits of participation in rebellion are often difficult to calculate due to the unpredictable nature of war. What dominates decision-making during war is uncertainty about future events. In economics, the distinction between risk and uncertainty goes back to Knight (Reference Knight1921) and Keynes (Reference Keynes1921) and implies that while estimates about future events are possible in risky environments based on known probability distributions, uncertain environments are too unstable and unpredictable to lend themselves to rational calculations (Nelson and Katzenstein Reference Nelson and Katzenstein2014).Footnote ¹² In irregular civil war, uncertainty is present when indiscriminate violence against civilians by insurgents and/or incumbents is the dominant form of violence. This is the case in contested areas in which no social contract between armed groups and civilians exists, or in areas of complete control in which the rival armed actor has no access to information (Kalyvas Reference Kalyvas2006; Arjona Reference Arjona2016). In such contexts, without the available information to calculate probable consequences of their actions, the question arises as to how people make decisions about how best to protect themselves. Accordingly, the focus here lies on the process of decision-making, not only the outcome of that process.

The process of decision-making under uncertainty is shaped by existing familiar repertoires and scripts. Applying the conceptual distinction between risk and uncertainty to decision-making in international relations, Nelson and Katzenstein (Reference Nelson and Katzenstein2014) argue that in the presence of uncertainty, actors’ decisions are rooted in social conventions – shared templates and understandings that coordinate action – rather than rational calculation. In situations of crises and social change, individuals’ decision-making process can no longer rely on established probability distributions. Rather, people rely on common knowledge from their immediate social environment to stabilize uncertainty and guide decision-making.

Regarding armed group mobilization, Nelson and Katzenstein’s theory suggests that social conventions help manage the uncertainty of war. This fits with two major lines of research in conflict studies. First, Gutiérrez and Wood’s (Reference Gutiérrez-Sanín and Wood2014) call for attention to immaterial and other-regarding incentives rather than self-regarding, material incentives for armed group mobilization in high-risk circumstances, including the noninstrumental, community-building use of ideology. Second, a growing number of researchers have emphasized that, in making their decisions under uncertain conditions, individuals are embedded in social contexts. Strong communities are important in facilitating recruitment through information, support networks, and normative commitments (Petersen Reference Petersen2001; Parkinson Reference Parkinson2013). Shesterinina (Reference Shesterinina2016, 417) shows how families and friends provide “access to information that is critical for making difficult choices about whether to fight for the group, escape the fighting, or defect to the other side.” But communities do not only provide crucial information but also access to a repertoire of collective action. Kaplan (Reference Kaplan2017, 34) argues that uncertainty during war encourages civilians to strive for “autonomy” and organize themselves. He observes in the context of unarmed civilians’ responses to civil war violence that “[a]n option to make daily life more certain and increase chances of survival is to turn to indigenous – meaning local – organizations.” Similarly, Arjona (Reference Arjona2016) has shown that preexisting community organizations that regulate civilian life are an important condition for civilians’ abilities to resist armed groups from intruding into their community.

My argument extends the second line of research, as I emphasize the social conventions – the “repertoire” or “script” – that communities can rely upon to respond to an uncertain situation. I bring together the question of individual motives with the nature of the group that seeks to attract participants to show that even though participants may be guided by various (non-)instrumental motives, it matters what kind of group they are asked to join.

Armed groups that rely on social conventions attract members through two main mechanisms: commonality and the context for self-empowerment. The first mechanism is commonality between the militia and potential participants: Building on social conventions, the militia is more familiar and less threatening to potential joiners. Such conventions are rituals and routines that transmit meaning to community residents and range from spiritual means to nonviolent forms of collective action. Relying on social conventions, militia institutions resonate with communities; which means that militias effectively frame their purpose of self-defense (Benford and Snow Reference Benford and Snow2000). Resonance occurs when a militia’s purpose is framed as credible and salient. Credible means that the framing reflects the militia’s actual activities and that the militia leader has expertise to direct these activities. Salient means that the militia’s purpose is central to the lives of the people and resonates with cultural narratives (Benford and Snow Reference Benford and Snow2000). For example, Ranger (Reference Ranger1985) argues that in southern Africa, respect for traditional religion made liberation movements’ ideology more relevant to peasant experiences and thus strengthened peasant support.Footnote ¹³ This implies that where militias bring with them “imported” ideologies and routines, successful mobilization is less likely. In addition to creating resonance, militias also need to provide innovation to motivate joiners. Community residents who neither feel protected by the state nor by the rebels “seek innovative courses of action” (Masullo Reference Masullo2015, 47) to overcome the limits of existing options. References to social conventions invite communities to adapt them to new social contexts, thereby creating a form of “invented tradition” (Hobsbawm and Ranger Reference Hobsbawm and Ranger1983) that legitimizes the new institution. The innovation may consist of new forms of legitimating leaders, new definitions of eligibility for membership, or new rules that govern participation.

The second mechanism, a context for self-empowerment, builds on the first. The reliance on repertoires rooted in the community facilitates people’s autonomy in shaping their own future, creating a belief in agency and therefore a context for self-empowerment. This idea builds on Wood’s (Reference Wood2003) concepts of “pleasure in agency” and “participation,” which explain peasant support for rural insurgency in El Salvador. “Pleasure in agency” refers to participants’ “ownership of successes of their collective actions” and expression of pride in the collective contribution to justice (Wood Reference Wood2003, 234–35), while “participation” emphasizes the expectation of success of collective action, which raises hope and a sense of purpose (Wood Reference Wood2003, 232). The concept of agency here refers to “protective strategies to retain autonomy, or self-rule” (Kaplan Reference Kaplan2017, 34), as it has been used in works on civilian agency in nonviolent communities. The commonality and the innovative character of the new institution foster the sense of purpose and the expectation of success as it improves upon existing social conventions.

I stipulate that social conventions are a necessary condition for militia mobilization by providing commonality between the militia and potential participants and a context for participants’ self-empowerment. A group that makes use of social conventions fares better in mobilizing members than a group that does not.

2.4.1 Case Selection and Data Collection

The goal of this book is to explain the formation of community-initiated militias across time, the diffusion of community-initiated militias across space, and the process by which such militias mobilize their members. To develop and explore the theory, I designed a subnational study of how community-initiated militias in Mozambique formed and diffused during the country’s civil war from 1976 to 1992 between the incumbent Frelimo and the insurgent Renamo. In particular, I study the largest militia, Naparama, and its spread across the northern territory.

Mozambique offers an excellent opportunity for study. The history of the war provides subnational variation on the variables of interest – military stalemates, levels of violence and territorial control, ethnic heterogeneity, and institutional (dis-)unity (Vines Reference Vines1991; Wilson Reference Wilson1992; Nordstrom Reference Raleigh and Kishi1997; Weinstein Reference Weinstein2007). Moreover, it is puzzling why community-initiated militias formed in Frelimo-held areas in Nampula and Zambézia provinces, as popular support for Renamo in these two provinces was stronger than in the southern region and the level of indiscriminate violence against civilians lower (Roesch Reference Roesch1992, 464; Finnegan Reference Finnegan1992, 72). In addition, militias in Mozambique’s war have not been studied in as much depth as the insurgent forces, even though their formation and activity provide important insights into why Mozambique experienced such a wide variation of violence. In the two provinces under study, Zambézia and Nampula, both community-initiated and state-initiated militias operated during the war. Thus, it is possible to compare how different types of militias form and mobilize.

In order to collect the necessary data, I conducted thirteen months of fieldwork in the capital and five districts in two provinces of Mozambique between 2010 and 2016.Footnote ¹⁴ I carried out more than 250 semi-structured interviews and oral histories with community members, former militia members, soldiers, and insurgents, public functionaries, and community leaders. I collected more than 10,000 pages of district and provincial government wartime reports and other government documents. In addition, I spoke to journalists, politicians, and researchers in various cities about the history of the war and its legacy.Footnote ¹⁵ From the interviews and archival material, I obtained detailed information on community histories; levels of violence and territorial control by insurgents and the government; motivations for joining the militia; the relationship between the militia, government, and the population; and current activities of former militia members who demand compensation from the government for their wartime contribution.

2.4.2 Analytical Strategy and Methods for Data Analysis

To study the formation and diffusion of militias, I combine within-case analysis over time with cross-sectional comparisons. Within-case analysis is suitable to study the process of the formation of Naparama in northern Mozambique over time (the militia as unit of analysis). I use controlled comparisons for the cross-sectional analysis of why Naparama diffused to certain communities but not to others (the militia in a certain community as unit of analysis). I also compare the formation and evolution of Naparama with that of the state-initiated part-time militia (the militia as unit of analysis) (see Table 2.2).

I use process tracing to construct theoretical narratives that identify causal pathways, thereby confirming or disconfirming alternative explanations (Hall Reference Hall, Mahoney and Rueschemeyer2003).Footnote ¹⁶ Process tracing has elements of both theory generating and theory testing (Bennett Reference Bennett, Box-Steffensmeier, Brady and Collier2008). It puts special emphasis on social processes by linking causal variables with different outcomes over time. It is therefore well suited for a theoretical framework that emphasizes causal mechanisms (Checkel Reference Checkel, Klotz and Prakash2008).

Process tracing has several advantages. First, process tracing is well suited to address the complex causes of militia formation and capture the endogenous relationships between different variables that account for how community-initiated militias form and diffuse. The method can help address complex causality, the problem of feedback loops, and endogenous relationships between variables (George and Bennett Reference George and Bennett2005). As a consequence, process tracing can reduce the gap that emerged between the ontology and the methodology in comparative politics – between the acknowledgment that the social world is characterized by a diversity of causal relationships and the methods that are unable to meet these challenges (Hall Reference Hall, Mahoney and Rueschemeyer2003).

Second, cross-sectional large-n studies often suffer from poor data quality. As a result, researchers may mis-categorize certain armed groups as militias, or they may not distinguish between different types of militias due to a lack of sufficient data on all cases. By focusing on few cases and studying them in depth, process tracing improves data quality and can better uncover omitted variables. Process tracing allows for causal inferences within the limits of a single case or few cases (Bennett and Elman Reference Bennett2006). To improve causal inference and explore external validity, I conduct out-of sample case studies for each part of the theory (Lyall Reference Lyall, Bennett and Checkel2015).

Third, identifying causal mechanisms is a crucial part of theory building and a necessary building block to advance the research agenda on militias and compare militias to other forms of violent collective action. Process tracing improves theory building by understanding how causal mechanisms work that connect different variables to each other while holding many variables constant (Checkel Reference Checkel, Klotz and Prakash2008; Gerring Reference Gerring2010; Kalyvas Reference Kalyvas, Kalyvas, Shapiro and Masoud2008b). Finally, process tracing is a suitable strategy for an approach that is grounded in methodological and theoretical pluralism (Checkel Reference Checkel, Klotz and Prakash2008). Contexts of complex causality are well suited for approaches that remain flexible with regard to their theoretical assumptions, as it is possible to explain different aspects of the phenomenon under study using theories with different assumptions about human behavior.

Process tracing, as I apply it, has two steps. First, I use the method to check which of the proposed values of the causal and outcome variables, as suggested by the alternative hypotheses, is congruent with the observed values of these variables. Second, I trace the mechanisms that link the causal variables with the outcome variables. For each case (each militia or each militia in a certain district), I map the potential factors that lead to community-initiated militias forming and diffusing and analyze the extent to which they affected the outcome.

I analyze the interview and archival data, as well as my field notes, with software for qualitative analysis (TAMS Analyzer, Weinstein Reference Weinstein2012), which helps code portions of the text files and then group text portions belonging to predefined codes. The qualitative data from my first field site, Nicoadala district in Zambézia province, informed the theory development. Data from the remaining field sites is used to further explore the theory.

The following chapter discusses in detail the opportunities and challenges of conducting fieldwork in such a polarized society as Mozambique, which is important to take into account when analyzing information collected about a war that ended more than twenty years ago. In the subsequent chapters, I present an overview of the history of the war and the empirical analyses of when, where, and how militias formed in Mozambique.