INTRODUCTION
Millions worldwide live under the influence of competing armed groups, including rebels, militias, and criminal organizations (Bakke, Cunningham, and Seymour Reference Bakke, Cunningham and Seymour2012; Lessing Reference Lessing2021). How do ordinary people in these contexts make judgments about those who aspire to rule them? In particular, what shapes civilian attitudes toward non-state governance where multiple armed groups compete? These questions are important not only because civilian attitudes affect broader conflict processes (Hirose, Imai, and Lyall Reference Hirose, Imai and Lyall2017; Wood Reference Wood2003), but also because their answers shed light on the sources of political legitimacy and the construction of durable political orders (e.g., Hobbes [Reference Hobbes1651] Reference Hobbes1968; Levi Reference Levi1989; Locke [Reference Locke1690] Reference Locke1961; Tilly Reference Tilly and Charles1975).
While foundational work has largely viewed territorial control as a prerequisite for armed group governance (e.g., Arjona Reference Arjona2016; Kasfir Reference Kasfir2005; Mampilly Reference Mampilly2011), political and criminal armed groups do frequently govern in areas they do not fully control (Börzel and Risse Reference Börzel and Risse2021; Loyle Reference Loyle2021; Loyle et al. Reference Loyle, Cunningham, Huang and Jung2023; Risse and Stollenwerk Reference Risse and Stollenwerk2018). They regulate civilian behavior and provide goods in contested areas of countries as diverse as Afghanistan (Jackson Reference Jackson2018), Brazil (Magaloni, Franco-Vivanco, and Melo Reference Magaloni, Franco-Vivanco and Melo2020), India (Waterman Reference Waterman2023), Mali (Weedon and Straus Reference Weedon and Straus2023), and South Africa (Cunningham, Huang, and Sawyer Reference Cunningham, Huang and Sawyer2021). Yet, how civilians respond to these efforts remains poorly understood.
We present results from a conjoint survey experiment conducted face-to-face with 2,397 respondents across 54 municipalities of Colombia recently contested by multiple armed groups. The survey asks respondents to evaluate pairs of hypothetical, competing armed groups engaging in forms of governance with randomly varied attributes.Footnote 1 We vary fundamental elements of non-state armed group governance, including rule-making and the provision of goods and services, as well as social network support, violence, and group ideology. Colombia is a fitting context in which to examine civilian attitudes toward armed group governance given the presence of large rebel armies and sophisticated criminal groups. These diverse armed actors frequently govern and regulate behavior in urban and rural areas, including where they compete for territorial control (Blattman et al. Reference Blattman, Duncan, Lessing and Tobón2024; De Bruin and Weintraub Reference De Bruin and Weintraub2023).Footnote 2
Our results show that the governance decisions that armed groups make in contested areas affect their legitimacy. We find that consulting with local leaders when making decisions about rules and establishing rules that reflect local traditions both decrease negative evaluations of armed groups. Groups that provide goods and services, including dispute resolution, infrastructure, and medicine, or that show restraint in the use of violence, are also judged less negatively. At the same time, we find only weak evidence for the notion that social support for armed groups from friends and family or the broader community decreases negative evaluations.
In assessing civilian responses to competitive armed group governance, our study contributes to scholarship on political legitimacy in contexts of limited statehood in several ways (Karim Reference Karim2020; Risse and Stollenwerk Reference Risse and Stollenwerk2018). First, existing scholarship on armed group governance tends to assume that civilians privilege security (e.g., Kalyvas Reference Kalyvas2006; Kalyvas and Kocher Reference Kalyvas and Kocher2007; Lyall, Blair, and Imai Reference Lyall, Blair and Imai2013), but does not directly test the consequences of armed group violence alongside other behaviors. We show that additional aspects of armed group interactions with civilians have important effects on civilian evaluations of such groups, even where they use high levels of violence to enforce their rules.
Second, while recent studies have begun to examine how armed groups govern absent territorial control (e.g., Jentzsch and Steele Reference Jentzsch and Steele2023; Keister Reference Keister2011; Loyle et al. Reference Loyle, Cunningham, Huang and Jung2023; Waterman Reference Waterman2023), how state presence affects criminal governance (Blattman et al. Reference Blattman, Duncan, Lessing and Tobón2024), and how residents compare rebel governance to that of the state (Revkin Reference Revkin2021), we still lack a clear understanding of how local residents assess the legitimacy of armed groups where they directly compete. It is precisely in these environments where civilian attitudes may be particularly important in shaping conflict dynamics (e.g., Condra and Wright Reference Condra and Wright2019; Kalyvas Reference Kalyvas2006; Rubin Reference Rubin2020). Given that criminal groups frequently compete against each other in areas where no single actor enjoys exclusive control, this study also advances our understanding of a broad class of phenomena in contexts of criminal governance (Barnes Reference Barnes2017; Lessing Reference Lessing2021; Moncada Reference Moncada2022; Skarbek Reference Skarbek2016).
Third, this article contributes to our understanding of how the distribution of benefits in violent spaces affects civilian attitudes. Prior research suggests that state efforts to win support by providing economic benefits may be ineffective (e.g., Lyall, Zhou, and Imai Reference Lyall, Zhou and Imai2020; Sexton and Zürcher Reference Sexton and Zürcher2024) and do not consistently reduce insurgent violence (e.g., Berman et al. Reference Berman, Felter, Shapiro and Troland2013; Lyall Reference Lyall2019; Sexton Reference Sexton2016; Weintraub Reference Weintraub2016). Yet these studies focus exclusively on economic interventions, and they consider only efforts by the state and its allies. In contrast, our findings suggest that the provision of goods and services by non-state armed groups—including dispute resolution, infrastructure, and healthcare—can improve civilian perceptions of these groups. Moreover, we demonstrate that efforts by armed groups to build legitimacy may benefit from tailoring governance approaches to locally salient traditions and from incorporating local leaders. This study is the first to theorize and experimentally demonstrate the impact of these factors on civilian attitudes toward armed groups.
THEORETICAL FRAMEWORK
In areas of limited statehood, where the state “lacks the ability to implement and enforce rules and decisions and/or lacks the legitimate monopoly over the means of violence” (Risse and Stollenwerk Reference Risse and Stollenwerk2018, 17), the exercise of governance by non-state armed groups frequently entails competition (Magaloni, Franco-Vivanco, and Melo Reference Magaloni, Franco-Vivanco and Melo2020; Mampilly Reference Mampilly2011; Weigand Reference Weigand2022).Footnote 3 We argue that civilians living in areas of limited statehood make comparative judgments about governing actors, even when they do not exhibit positive attitudes toward any armed group. Indeed, individuals need not view an armed group in a positive light nor actively support it in order to prefer it to another group. We therefore theorize about factors that reduce negative assessments of armed groups. In our theoretical framework, we focus on rules governing civilian behavior and the provision of goods and services, core features of armed group governance (Mampilly Reference Mampilly2011; Weinstein Reference Weinstein2007). Additionally, we hone in on networks of support for armed groups because attitudes may be shaped by the actions and ideas of relevant others (e.g., Kuran Reference Kuran1991; Petersen Reference Petersen2001).Footnote 4
Coherence with Community Norms
A central element of rebel and criminal governance is the establishment of rules that regulate the behaviors of community members (e.g., Arias Reference Arias2017; Arjona Reference Arjona2016). We argue that civilian attitudes are shaped by whether the rules that armed groups establish take into account community norms. We distinguish between two forms of coherence with community norms: procedural coherence, by which we mean the extent to which local leaders are included in decision-making processes for establishing rules, and conventional coherence, which denotes whether the rules armed groups adopt for the community conform to local traditions. Alternatively, armed groups may bypass local leaders and impose rules that are inconsistent with local traditions.
Non-state actors must decide how to incorporate local community perspectives into the rules they establish over civilian behaviors, including permissible conduct in public spaces, movement restrictions, or with whom civilians can have relationships (Arjona Reference Arjona2016; Mampilly and Stewart Reference Mampilly and Stewart2021; Weinstein Reference Weinstein2007). Some armed groups opt to make rules unilaterally, whereas others choose to incorporate local community representatives into decision-making about these rules (Murtazashvili Reference Murtazashvili2016; Steele and Weintraub Reference Steele and Weintraub2022; Vargas Castillo Reference Vargas Castillo2019). We expect that procedural coherence will reduce negative assessments of armed groups. Where local leaders are themselves perceived as legitimate, we argue that consultation with them will be seen as involving the community. Participation may be valued for its own sake (Wood Reference Wood2003) or it may be valued because procedures that allow individuals to provide input are generally perceived as fair, even when the outcome is unfavorable (Tyler Reference Tyler2006). Moreover, even where local leaders are not perceived as legitimate, consulting with them may signal consistency and mitigate uncertainty about the future form that rebel or criminal governance will take, thus reducing negative assessments of groups providing such governance.Footnote 5
We further argue that alignment of rules with local customs and limiting rule innovation will reduce civilians’ negative evaluations (Mampilly and Stewart Reference Mampilly and Stewart2021). This was the case for Naparama militia leaders in Mozambique, for example, who were more positively received when they appealed to local beliefs and values, incorporating the use of traditional medicine for personal defense (Jentzsch Reference Jentzsch2022). Armed group rules that draw upon shared community practices may be seen as less of an imposition, increasing public acceptance. In contrast, communities may resist armed group rules that do not align with local norms and traditions. In Peru, resistance to the insurgent group Shining Path grew once it “imposed its values and organizational structures, ignoring and negating communal traditions” (Del Pino Reference Del Pino and Stern1998, 179); for example, it banned longstanding Catholic festivals.Footnote 6
We also expect that governing processes inclusive of the local community (procedural coherence) will be more important to civilians than outcomes that reflect community traditions (conventional coherence).Footnote 7 Even a near-perfect match of rules with community norms and preferences that procedurally excludes community input can quickly unravel. In Syria, for example, while Jabhat al-Nusra reportedly respected local norms and helped enforce existing rules in communities it entered initially, it did not incorporate residents’ voices into governance decisions and eventually imposed rules increasingly out of step with the communities it aimed to control (Lister Reference Lister2016). In contrast, consultations with local leaders signal greater predictability and control over what rules might apply in the near future. Such predictability is likely to be particularly valuable in contested areas beset by uncertainty. Procedural coherence therefore can ameliorate commitment problems that armed groups face when seeking to govern local communities. We derive the following hypotheses:
-
H1 Procedural coherence: Consulting local leaders when establishing rules for the community reduces the extent to which an armed group is viewed negatively.
-
H2 Conventional coherence: Establishing rules for the community that reflect local traditions reduces the extent to which an armed group is viewed negatively.
-
H3 Procedural versus conventional coherence: Procedural coherence reduces the extent to which an armed group is viewed negatively more than conventional coherence does.
Social Networks
The social networks within which individuals are embedded shape their behaviors via “cue-taking” (e.g., Campbell Reference Campbell2013; Siegel Reference Siegel2009). Where publicly expressing political attitudes or engaging in political actions is high-risk, as in repressive environments, the incentives to engage in them may depend on the social context and conduct of others (Granovetter Reference Granovetter1978; Kuran Reference Kuran1991; Petersen Reference Petersen2001). We argue that private attitudes are likewise shaped by perceptions of what relevant others believe. These relevant others can be either from the broader community or from more intimate networks of friends and family.
Contexts with multiple armed actors are prone to high levels of uncertainty about what armed groups are doing and what they are like. Essential information about these armed actors can spread through broad community networks. For example, across multiple conflicts in Uganda and neighboring countries, Lewis (Reference Lewis2020) shows that rumors about the capabilities of nascent armed groups were transmitted through coethnic kinship networks. Social networks can also convey information about whether community members support armed groups. This information about the extent of an armed group’s perceived support among community members serves as a heuristic for its presumed nature and behavior. Evaluations of armed groups in contested areas should therefore depend on the extent to which individuals think that members of the broader community support them.
Information about support from intimate social networks might be even more valuable because, in high-risk environments, friends and family are often people’s most and perhaps only trusted connections (Parkinson Reference Parkinson2013). In the Philippines, word of mouth among friends and family was a crucial conduit through which information about the credibility of the New People’s Army was passed between villages (Haim Reference Haim2018). Likewise, during the civil war in Abkhazia, residents “sought information on the threat posed by the Georgian forces from the social networks they interacted with and trusted at the time” (Shesterinina Reference Shesterinina2021, 45), leaning particularly on relatives and friends. We expect that support from personal networks will be more important in shaping attitudes toward armed groups than support from other community members because it may be difficult to discern whether community members genuinely support an armed actor or are instead falsifying their preferences due to fear (Kalyvas Reference Kalyvas2006; Kuran Reference Kuran1991). Additionally, support from relatives and friends may be more informative for personal assessments about whether the armed group would generate benefits or harms. We therefore hypothesize:
-
H4 Community networks: Support from community members reduces the extent to which an armed group is viewed negatively.
-
H5 Personal networks: Support from personal networks reduces the extent to which an armed group is viewed negatively.
-
H6 Community versus personal networks: Support from personal networks reduces the extent to which an armed group is viewed negatively more than support from community members does.
Provision of Goods and Services
Armed groups regularly provide goods and services in contested areas (Loyle et al. Reference Loyle, Cunningham, Huang and Jung2023). In India, the United Liberation Front of Assam (ULFA) built roads, flood defenses, farming and irrigation projects, and schools in areas they did not fully control (Waterman Reference Waterman2023). Similarly, milícias in the favela of Rio das Pedras in Rio de Janeiro paved roads, increased access to electricity, and expanded water and sanitation services (Arias Reference Arias2017, 231). The provision of goods and services may improve the material welfare of individuals or communities, and may also increase perceptions of armed groups’ competence and resolve. We therefore expect that the provision of goods and services affects the legitimacy of armed groups.
Of course, armed groups face difficult trade-offs, particularly in contested areas, as they seek to cultivate support and build legitimacy: they must dedicate resources to countering other geographically proximate groups while simultaneously providing governance to civilians (Arias Reference Arias2017). This may mean that services and goods on offer are of lower quality than what residents would ultimately desire (Masullo Reference Masullo2021). In such cases, goods provision may be insufficient to reshape civilian attitudes. However, areas contested by armed groups are places where the state struggles to provide effective governance. As a result, even low-quality goods and services may generate changes in evaluations of armed groups. For example, in Afghanistan prior to the U.S. withdrawal, even among civilians who did not sympathize with the Taliban, the group’s system of dispute resolution was at times “more accessible, faster, and perceived as fairer than any alternative” (Jackson Reference Jackson2021, 129).
Residents may also respond negatively to armed groups’ provision of goods and services because they view any such efforts as attempts to exert control. Larkins (Reference Larkins2015, 186) describes how some residents of the Rio de Janeiro favela of Rocinha refused the dispute resolution mechanism provided by a criminal group because they rejected “the way they were co-opted as active participants in the maintenance of trafficker-backed public order.” However, in violent areas routinely characterized by limited state capacity and poverty, any goods or services provided by armed groups may be sufficient to reduce residents’ negative judgments about armed groups. Despite their brutality, criminal gangs in the Zona Norte of Rio de Janeiro managed to cultivate “criminal legitimacy” by providing gifts to children at holiday parties, maintaining public order, and stimulating the economy (Arias and Barnes Reference Arias and Barnes2017, 456). We therefore argue that, although residents may view goods and services provided by armed groups as attempts to exert control, their provision will tend to reduce negative perceptions of armed groups:
-
H7 Goods and services: Providing goods and services reduces the extent to which an armed group is viewed negatively.
We disaggregate different goods and services to understand how their provision shapes civilian attitudes. We focus on dispute resolution, infrastructure investments, and healthcare. All three types of goods can have a substantial impact on communities and are commonly provided by many armed groups globally (e.g., Albert Reference Albert2022; Huang Reference Huang2016). Dispute resolution and infrastructure investments may generate more long-term economic resilience for communities when compared to other kinds of goods and services (Ibáñez et al. Reference Ibáñez, Arjona, Arteaga, Cárdenas and Justino2024), while during public health crises medical care may be particularly valued (Breslawski Reference Breslawski2022; De Bruin and Weintraub Reference De Bruin and Weintraub2023).
We test three competing hypotheses related to the provision of goods and services. Respondents may value especially order-related goods and services such as dispute resolution, given that instability, impunity, and commitment problems are widespread in contested areas (Blair et al. Reference Blair, Moscoso, Castillo and Weintraub2022; Blattman, Hartman, and Blair Reference Blattman, Hartman and Blair2014). Indeed, armed groups often prioritize dispute resolution to consolidate power and become de facto administrators of justice (Arjona Reference Arjona2016, 11, 72). Yet, armed groups have limited capabilities to enforce decisions in competitive areas given the existence of multiple authorities that could potentially resolve quotidian disputes. Civilians therefore may perceive armed groups’ decisions as unenforceable over the medium-term and prefer welfare-related goods that bring concrete material benefits—as both roads and healthcare do. Finally, particularly amidst public health crises such as a global pandemic, civilians might value health care above other goods and services.
-
H8a Order-related goods and services: Providing dispute resolution reduces the extent to which an armed group is viewed negatively more than providing infrastructure or healthcare does.
-
H8b Welfare-related goods and services: Providing infrastructure and healthcare reduces the extent to which an armed group is viewed negatively more than providing dispute resolution does.
-
H8c Healthcare-related goods and services: Providing healthcare reduces the extent to which an armed group is viewed negatively more than providing dispute resolution or infrastructure does.
Scope Conditions
We expect our arguments to hold across a wide range of conflict, post-conflict, and otherwise violent contexts where multiple political or criminal armed groups capable of governing are present. Many contexts meet these criteria. For example, more than 70% of countries contain “significant” areas of limited statehood (Stollenwerk Reference Stollenwerk, Draude, Börzel and Risse2018). From 1989 to 2022, the Uppsala Conflict Data Program (UCDP) recorded over 900 conflicts between non-state armed actors (Davies, Pettersson, and Öberg Reference Davies, Pettersson and Öberg2023; Sundberg, Eck, and Kreutz Reference Sundberg, Eck and Kreutz2012). The arguments we develop may be most applicable where armed groups do not compete for support along ethnic, religious, or racial lines, as these cleavages may produce stronger a priori commitments to specific armed actors. Our scope conditions do, however, cover identity-based conflicts where multiple armed groups compete for support among the same constituencies.Footnote 8 As the sources of legitimacy for armed groups and state actors are likely to differ systematically, our arguments may not generalize to competition between state actors and armed non-state groups. We return to this question in the conclusion.
RESEARCH DESIGN
Case Selection
We test our hypotheses in contested areas of Colombia, which bear resemblance to many other territories worldwide characterized by armed group competition. Colombia has experienced a long-running conflict involving multiple political and criminal armed groups. The key historical cleavage in the civil war has been ideological, pitting left-wing rebels in favor of radical redistribution of wealth and political power against status-quo minded national and regional elites who mobilized paramilitaries to support their cause (Karl Reference Karl2017; Steele Reference Steele2017). Despite the 2016 signing of a peace agreement with the largest rebel group, the Fuerzas Armadas Revolucionarias de Colombia (FARC), strong rebel armies, including the Ejército de Liberación Nacional (ELN) and dissident factions of the FARC, remain active.Footnote 9 The lucrative drug trade, combined with illegal mining, has also caused criminal armed groups—many with no discernable political ideology—to proliferate throughout the country.
We study Colombia because it fits our scope conditions: it features areas of limited statehood where multiple non-state armed groups compete and seek to govern civilian populations. While the Colombian state is comparatively strong relative to other countries that have experienced civil war, and maintains a presence in all municipalities, state capacity varies geographically (e.g., Acemoglu, García-Jimeno, and Robinson Reference Acemoglu, García-Jimeno and Robinson2015; Ballvé Reference Ballvé2020; Holmes, Piñeres, and Curtin Reference Holmes, Piñeres and Curtin2010).Footnote 10 We focus on contested areas where multiple armed groups are or have recently been present, and where the state struggles to provide effective governance. A variety of armed groups have taken advantage of these conditions. For example, in each year between 2008 and the first half of 2022, paramilitary successor organizations had a presence in at least 25 of the country’s 32 departments, while FARC dissidents and the ELN each had a presence in at least 15 departments between 2017 and the first half of 2022 (Indepaz 2022). In other words, each day many civilians in Colombia are exposed to the actions of multiple non-state armed groups.
Many of these groups have demonstrated significant governance capacity, both alongside state efforts and in their absence (e.g., Arjona Reference Arjona2016; Daly Reference Daly2022; Otero-Bahamon, Uribe, and Peñaranda-Currie Reference Otero-Bahamon, Uribe and Peñaranda-Currie2022), including in areas in which multiple groups are present (De Bruin and Weintraub Reference De Bruin and Weintraub2023). For example, both rebel groups and criminal organizations in Colombia have regularly provided dispute resolution services (Arjona Reference Arjona2016; Blair et al. Reference Blair, Moscoso, Castillo and Weintraub2022). The FARC undertook extensive road construction (Currie, Otero-Bahamon, and Uribe Reference Currie, Otero-Bahamon and Uribe2021; Uribe, Otero-Bahamón, and Peñaranda Reference Uribe, Otero-Bahamón and Peñaranda2021), and many paramilitary groups and criminal organizations have also financed, built, or paved roads (Brewer Norman Reference Brewer Norman2012; Schmidt Reference Schmidt2023). Colombian armed groups ensured that government-issued quarantines were enforced and even provided local populations with masks and other protective gear during the COVID-19 pandemic (Angelo Reference Angelo2020; Idler and Hochmuller Reference Idler and Hochmuller2020). Public opinion data also indicate that armed groups enjoy some level of support (e.g., Rodríguez-Raga et al. Reference Rodríguez-Raga, Seligson, Donoso, Quiñones and Schwarz-Blum2006; Steele Reference Steele2017).Footnote 11
Sampling
We survey individuals who have recently experienced armed group competition or for whom such an experience likely would be realistic. We therefore administer the survey in 54 municipalities of Colombia characterized by the recent presence of multiple armed groups. To identify contested municipalities, we draw on internal data from the following organizations: the Colombian National Police, the Colombian Attorney General’s Office, and the Ideas for Peace Foundation (Fundación Ideas para la Paz, FIP). Each defines the presence of armed groups based on criminal investigation and/or fieldwork rather than measures of violence such as clashes. We define contested municipalities as those in which two or more armed groups were active according to any of the three sources between 2015 and 2018.Footnote 12 To be included, armed groups must have demonstrated capacity for sustained presence over some stretch of territory; we thus exclude neighborhood gangs in urban centers tied to small tracts of territory (e.g., pandillas in Cali or combos in Medellín).
We classify 103 municipalities as “contested” (Figure 1 and Supplementary Table A2), corresponding to approximately 9% of Colombia’s 1,123 municipalities. From this list, we exclude 10 large cities (e.g., Medellín and Santa Marta) because armed group behavior and governance vary across urban and rural areas—including in the profile of recruits, level of criminal know-how, and scale of rents that can be obtained via micro-trafficking and extortion (in cities) or transshipment of drugs and other contraband (in rural areas). Our sample includes respondents from 54 municipalities selected based on a stratified sampling procedure with five regional strata (Caribe, Pacífico, Territorios Nacionales, Antioquia, and Santanderes) and strata based on historic FARC control.Footnote 13 Municipalities are selected with a probability proportional to their 2022 projected population size as calculated by Colombia’s census agency (Departamento Administrativo Nacional de Estadística, DANE).
Contested Municipalities in Colombia
Within selected municipalities, our study covers 2,397 respondents surveyed by the Colombian firm Sistemas Especializados de Información (SEI) between July 3 and August 7, 2022. Respondents were selected both in municipal capitals (cabeceras municipales) and rural hamlets (centros poblados).Footnote 14 Hamlets within a municipality were selected with a probability proportional to their projected 2022 population as calculated by DANE. Within municipal capitals and selected hamlets, blocks were selected randomly. Enumerators determined the number of inhabited dwellings in each selected block, and software installed on the enumerators’ tablets randomly selected dwellings from this list.Footnote 15 Within each selected dwelling, the enumerator listed the number of households, and the program randomly selected one per dwelling. The enumerator then determined the number of individuals within the household, and the system randomly chose one person over the age of 18 who was capable of responding and present in the municipality during enumeration.Footnote 16
Survey Design and Analysis
We evaluate our hypotheses via a conjoint experimental design (Hainmueller, Hangartner, and Yamamoto Reference Hainmueller, Hangartner and Yamamoto2015; Hainmueller, Hopkins, and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014). Conjoint experiments reduce social desirability bias (Horiuchi, Markovich, and Yamamoto Reference Horiuchi, Markovich and Yamamoto2022), perform well in comparison to behavioral benchmarks (Hainmueller, Hangartner, and Yamamoto Reference Hainmueller, Hangartner and Yamamoto2015), and provide a partial solution to violations of information equivalence (Dafoe, Zhang, and Caughey Reference Dafoe, Zhang and Caughey2018). In our survey, respondents are presented with five pairs of hypothetical armed actors competing for territorial control. Armed groups’ attributes and their possible levels are listed in Table 1. The levels of all attributes are varied randomly from one profile to the next. To evaluate civilian attitudes, we use both forced choice questions in which respondents decide which armed group in each pair would be worse for the community, and rating questions in which respondents indicate how bad or negative a given armed group would be for the community (on a scale from 1 to 5).
Attributes and Levels
When introducing the profiles, the survey prompt emphasizes that we are asking about hypothetical groups in order to reduce the sensitivity of the question and to minimize respondents’ potential discomfort.Footnote 17 The prompt reads:
In some communities, illegal armed groups such as insurgents or drug trafficking organizations try to interact with the local population. This includes establishing rules, providing services, and asking for money. Sometimes these groups compete with other illegal armed groups within the same community.
Imagine that this were your community and that two armed groups were competing to influence it.
I am going to present you with five pairs of such imaginary armed groups so that you can, in each case, tell me which of these groups would be worse for the community, if the groups were real.
It is important to clarify that these examples are not real but rather fictitious. If you are not sure of your response, please try your best to choose one of the two options.
Now I am going to present the first pair of imaginary armed groups. Do you want to read it alone, read it together, or listen to it?
We include attributes to capture our core theoretical arguments concerning conventional and procedural coherence, social network support, and the kinds of goods and services that armed groups provide. Note that because it may be implausible for many armed groups to obtain the support of a majority of a community, we operationalize social network support as support from “several” members of the respondents’ community or their friends and family. We discuss the implications of this choice below. We also include the level of violence that armed groups use to enforce rules, as well as armed group ideology, two additional attributes prominent in prior literature (e.g., Costalli and Ruggeri Reference Costalli and Ruggeri2015; Keister Reference Keister2011; Lyall, Blair, and Imai Reference Lyall, Blair and Imai2013; Masullo Reference Masullo2021; Wood Reference Wood2003). Because most groups in contested areas cannot credibly promise to monopolize violence and provide security to the whole community, we focus on an aspect of violence that armed groups do have within their control: how much violence they use to punish deviations from their rules. We include political ideology as captured on the left–right spectrum because the main cleavage of the civil war in Colombia has been ideological (e.g., Gutierrez Sanín Reference Gutierrez Sanín2019; Ugarriza and Craig Reference Ugarriza and Craig2013); moreover, ideological identifications are relatively salient in Colombia (e.g., García-Sánchez and Plata-Caviedes Reference García-Sánchez and Plata-Caviedes2020; Levy Reference Levy2025).Footnote 18
The attributes are designed such that no combination of attribute levels is highly unrealistic, implausible, or likely to signal a specific armed group. For example, groups that provide goods and services may also wield high levels of violence to enforce their rules. Similarly, both left- and right-wing groups may incorporate local leaders into decision-making, even if they do not have significant support within a given community. An example pairing of hypothetical armed groups can be found in Table 2.
Example Pairing of Hypothetical Armed Groups
Following the presentation of the two profiles, we ask respondents to choose which group would be worse for the community, and then to rate how bad each group would be. We ask “how bad or negative” the armed group would be for the community to avoid eliciting positive assessments of armed groups. Even though we present hypothetical profiles, asking for positive assessments could produce confusion or discomfort. Moreover, we are interested in relative preferences, even if overall attitudes toward any armed group may be negative. The questions are:
-
Forced Choice: In your opinion, which of these two groups would be worse for the community?
-
Rating Group 1: We are going to think about the same groups. How bad or negative would group 1 be for the community? (1 not at all – 5 very)
-
Rating Group 2: We are going to think about the same groups. How bad or negative would group 2 be for the community? (1 not at all – 5 very)
Our principal quantity of interest is the average marginal component effect (AMCE) (Hainmueller, Hopkins, and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014). To calculate the AMCE, each outcome is regressed on several indicator variables, with each indicator representing an attribute level, and one omitted level per attribute constituting the baseline. Standard errors are clustered by respondent. Additionally, we rely on marginal means and omnibus F-tests when testing for differences across respondent subgroups (Leeper, Hobolt, and Tilley Reference Leeper, Hobolt and Tilley2022). There are two caveats regarding the interpretation of the results. First, AMCEs depend on the distribution of attributes used for averaging; hence, they are ideally derived from profile distributions mimicking those occurring in real life (de la Cuesta, Egami, and Imai Reference de la Cuesta, Egami and Imai2022). We were prevented from taking this approach by a lack of data. Second, we cannot interpret the results in terms of individuals’ “raw” or majority preferences (Abramson, Koçak, and Magazinnik Reference Abramson, Koçak and Magazinnik2022; Ganter Reference Ganter2023). Instead, the AMCE indicates the average causal effect of a certain characteristic, compared to the baseline, on an armed group’s evaluation among respondents (Bansak et al. Reference Bansak, Hainmueller, Hopkins and Yamamoto2023).
Protection of Enumerators and Respondents
In designing and fielding the survey, we took several steps to protect enumerators and respondents.Footnote 19 First, we prioritized physical safety. Any municipality or hamlet that local officials, community leaders, or the survey firm deemed unsafe was excluded; see Section B of the Supplementary Material for further details.Footnote 20 Protocols were also established to physically protect respondents and enumerators from COVID-19.Footnote 21 Second, we sought to minimize the potential for the survey to cause respondent distress. We alerted respondents to the survey’s content during the informed consent and indicated that they could skip any questions or end the survey at any time. As described above, the prompt emphasized to respondents that the questions concerned fictitious rather than real armed groups. The conjoint design also minimized the sensitivity of the core questions by preventing researchers and enumerators from identifying which attribute(s) shaped individuals’ responses. Third, to protect respondents’ anonymity, we did not collect names, birth dates, or government identification numbers that could identify respondents. Household addresses used for sampling and audio recordings of a random subsample of interviews for quality control were temporarily retained by the survey firm until quality checks were completed, and then were permanently deleted. We also excluded hamlets with fewer than 68 households to prevent potential identification of respondents by inference. All survey data collected were automatically encrypted, and no individually identifiable data were made accessible to the authors. Finally, we opted not to pay respondents to avoid subjecting them to increased attention from others, including armed groups. While unlikely, we sought to eliminate the possibility that respondents receiving payments could become targets of suspicion or crime.
RESULTS
Figure 2 displays our main results (data are available at De Bruin et al. Reference De Bruin, Gabriela, Livia I. and Michael2025). The dots with the horizontal lines represent the level-specific AMCEs and 95% confidence intervals, whereas the dots on the vertical middle line depict the baseline level of each attribute. The left panel shows the results for the forced choice outcome; the right panel shows the results for the rating outcome.Footnote 22 In line with our expectations, we find that civilian attitudes toward armed groups in contested areas are shaped by the choices that armed groups make. Armed groups that create rules reflecting local traditions, consult with local leaders, and provide goods and services are evaluated less negatively than those that do not, though the specific type of good presented does not appear to matter. We find limited support for the expectation that procedural coherence reduces negative evaluations more than conventional coherence. We also find only weak support for our hypotheses about social networks.Footnote 23
Average Marginal Component Effects (AMCEs) of Governance Attributes
Note: Numerical results are presented in Supplementary Table A4.
Regarding coherence with community norms, we find that, compared to rule-making that does not take the community into account, rules reflecting local traditions reduce the probability of an armed group being characterized as worse for the community by 5.1 percentage points (95% CI =−0.067,−0.035), while those made in consultation with local leaders reduce the probability by 6.0 percentage points (95% CI =−0.076,−0.044). The rating results echo these findings, providing support for H1 and H2.Footnote 24 Furthermore, in the rating analysis, we find that consultations with leaders have a slightly stronger effect than coherence with traditions, providing partial support for our argument that procedural coherence is more impactful than conventional coherence (H3). The difference between consultations and conformity with traditions, however, is not statistically significant with the forced choice outcome (Supplementary Figure A1).Footnote 25 Our hypothesis about the relative importance of procedural coherence over conventional coherence was motivated by the expectation that the former would serve as a more credible signal of armed groups’ intentions and predictability. That support for this hypothesis is limited might suggest that contested areas feature less uncertainty than anticipated; alternatively, armed groups may struggle to send credible signals in such places.
Our hypotheses about social networks receive weak support. In line with our expectations (H5), we find that support from a respondent’s friends and family reduces the probability that an armed group is judged as worse for the community by 1.6 percentage points (95% CI =−0.032,−0.001), compared to groups that receive support from only a few. Support from intimate social networks and from community members also reduce negative ratings of armed groups by 0.05 points on the five-point rating scale (95% CIs =−0.096,−0.010 and−0.095,−0.013). These effects are small. Moreover, while we find that support from the community matters in the rating outcome, it is not significant in the forced choice outcome (H4).Footnote 26 Further, there is no evidence for the notion that support from one’s personal network is more important than support from the broader community (H6) (Supplementary Figure A1).
In interpreting these results, one consideration is that we worded the attribute on community support in a relatively subtle way, shifting from armed groups as supported by “few people” to being supported by “several members of the community” or “several friends and family.” This wording ensures that all levels of social network support are compatible with other attributes; respondents may have thought it implausible for “most” or “all” of their community to support a group that used high levels of violence, for example. However, it also implies that differences between attribute levels may have been less apparent to respondents. The estimates on social network support are, therefore, likely conservative ones.
Turning to goods and service provision, Figure 2 shows that it reduces negative evaluations. Compared to delivering very little that benefits local residents, supplying medicines reduces the probability of being judged as worse for the community by 8.2 percentage points (95% CI =−0.101,−0.064); building roads by 10.0 percentage points (95% CI =−0.118,−0.081); and administering dispute resolution by 7.1 percentage points (95% CI =−0.089,−0.053). The related AMCEs are substantively similar and statistically significant for the rating outcome. These findings provide support for H7. If civilians in contested areas generally preferred being left alone, and thus viewed all goods supplied by armed groups with skepticism, we would expect to see no effects on attitudes here. The same would be true if goods provision were insufficient to shape civilian attitudes, as recent scholarship on economic aid provided by states suggests (e.g., Lyall, Zhou, and Imai Reference Lyall, Zhou and Imai2020; Sexton and Zürcher Reference Sexton and Zürcher2024). Instead, our results provide support for the argument that goods provision in contested areas affects the relative legitimacy of armed groups.
Interestingly, while providing goods and services significantly improves perceptions of armed groups, the type of service appears not to matter much: we find little difference between the effects of order-, welfare-, and healthcare-related goods (H8a–c). There is one exception: compared to dispute resolution, roads reduce the probability of a more negative assessment by 2.9 percentage points in the forced choice outcome (95% CI =−0.047,−0.011) yet this difference is not significant in the rating results (Supplementary Figure A1). We consider it unlikely that these goods have similar effects because they are perceived to be of low quality. If this were the case, we would not expect to see such a substantial impact of them on evaluations of armed groups. Alternatively, civilians may not distinguish between armed groups providing different goods because groups that provide one type of good also tend to provide others. While we cannot rule out this possibility, we note that evidence on this point is mixed. Cross-national data on rebel governance suggest that when rebels establish laws regarding civilian behavior, they also tend to provide health services, infrastructure, and mediation of civilian disputes (Albert Reference Albert2022). At the same time, work on armed group governance in contested areas of Colombia specifically does not find a strong correlation between the provision of dispute resolution and economic benefits (De Bruin and Weintraub Reference De Bruin and Weintraub2023). As a result, we interpret these findings as supportive of the notion that the supply of all three types of services and goods has payoffs for armed groups.
We also find that violence is important in shaping views of how bad a group would be for the community. Figure 2 shows that the use of “little violence” as opposed to “lots of violence” in enforcing rules reduces the probability of being judged more negatively by 13 percentage points (95% CI =−0.144,−0.117), with similar findings for the rating outcome. These effects are substantively large, as might be anticipated.Footnote 27 In terms of ideology, compared to groups that do not take a position on political issues, having a leftist ideology increases the probability of being judged more negatively by 1.8 percentage points (95% CI = 0.002, 0.034).Footnote 28 Finally, we test whether coherence with community norms and service provision matter even for those groups that use high levels of violence to enforce their rules. We find that they do.Footnote 29
In summary, our findings show that many of the governance decisions armed groups in contested areas make can succeed in shaping civilians’ views of them. In particular, rebel and criminal groups that establish rules that reflect community traditions, and groups that involve local leaders in decision-making are perceived to be less detrimental to the community than those that do not. The provision of goods and services, including dispute resolution, infrastructure investments, and healthcare, as well as the use of less violence when enforcing rules, also improves perceptions of armed groups.
CONCLUSIONS
Armed groups routinely compete with one another to control territory where the state’s capacity is limited. While a growing body of evidence shows that armed groups provide governance in contested areas, we know little about how civilians form comparative judgments about competing armed groups that seek to rule them. We present results from a conjoint survey experiment featuring hypothetical armed groups fielded in contested areas of Colombia, and show that armed groups’ governance choices shape civilian attitudes in significant and substantial ways. In particular, establishing rules that reflect local traditions, consulting with local leaders, and offering goods and services cause less negative evaluations of armed groups. However, we only find limited evidence that social network support matters or that inclusive governing processes reduce unfavorable assessments more than rules that reflect local traditions. Our study is the first to provide experimental evidence of how civilians respond to different elements of armed group governance in contested areas, where civilian support may be particularly impactful.
Our results hold broader implications for understanding armed group competition and efforts to build political legitimacy. First, while existing scholarship emphasizes the role of violence in shaping civilian attitudes, we find that civilians are also sensitive to the perceived match between armed group rules and local norms. This suggests that armed groups seeking legitimacy can benefit from tailoring their governance practices, either by incorporating local leaders into decision-making processes or establishing rules that reflect preexisting local traditions. Second, rebel and criminal groups may not have the ability or motivation to reduce overall violence in contested areas, but they can exercise restraint when enforcing rules governing civilian behavior; we show, indeed, that they have incentives to do so. Third, armed groups receive substantial dividends from offering dispute resolution, roads, and medicine. The finding that providing any of these goods or services reduces civilians’ rejection of the group suggests that the bar to “outcompete” rivals through service provision might be relatively low. This may help explain why rebel and criminal actors in contested areas continue to provide goods and services, despite the costs of doing so. Future work should explore a broader variety of relationships among armed groups, and between armed groups and the state. While our primary interest in this paper is contexts of competition, state and non-state armed actors can also coexist without competition or even collaborate (Barnes Reference Barnes2017; Magaloni, Franco-Vivanco, and Melo Reference Magaloni, Franco-Vivanco and Melo2020; Moncada Reference Moncada2022; Staniland Reference Staniland2012). The components of governance that we study may affect attitudes differently in such contexts.
More broadly, our results suggest potential differences in how governments and armed groups can build legitimacy in areas of limited statehood. While state-led efforts to provide economic benefits during armed conflict have shown mixed results in increasing support for the government (e.g., Lyall, Zhou, and Imai Reference Lyall, Zhou and Imai2020; Sexton and Zürcher Reference Sexton and Zürcher2024), our findings imply that non-state armed groups can generate returns from providing various types of benefits. This discrepancy between how the state and non-state armed groups can build legitimacy may reflect lower civilian expectations of armed groups than of the state. Future research should explicitly compare armed groups to the state to empirically assess this possibility. Yet given how widespread areas of weak state governance are, it is important to understand the sources of political legitimacy for competing non-state actors in these areas. This study finds that, even in contexts characterized by contestation, political and criminal armed groups have meaningful incentives to provide services and to incorporate the views of local residents into their governance.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424000789.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/NKPW79.
ACKNOWLEDGMENTS
Authors are listed alphabetically and contributed equally to the manuscript. For excellent feedback, we thank Mary Beth Altier, Kiran Auerbach, Kyle Beardsley, Joseph Huddleston, Cyanne Loyle, Romain Malejacq, Philip Martin, Almendra Ortiz de Zárate Béjar, Carlos Solar, Juan Tellez, and Alex Waterman as well as the audience of the Extra-Legal Governance Conference at Brown University and audiences at the annual meetings of the American Political Science Association, Conflict Research Society, European Political Science Association, International Studies Association, Midwest Political Science Association, and Public Choice Society. We also thank Tomás Alvarez Perez, Viviana Baraybar Hidalgo, Manuel Moscoso Rojas, and Nicole Ramirez for their excellent research assistance, as well as the editors and anonymous reviewers at the APSR.
FUNDING STATEMENT
This research was funded by the National Science Foundation (Awards #1558488 and #2038632).
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors declare the human subjects research in this article was reviewed and approved by Hamilton College, Duke University, Binghamton University, University of Oxford, and Universidad de los Andes; certificate numbers are provided in the text. The authors affirm that this article adheres to the principles concerning research with human participants laid out in APSA’s Principles and Guidance on Human Subject Research (2020).
Comments
No Comments have been published for this article.