THEORY

To explain the demand and supply of criminal governance, the theory centers on the needs and incentives of the population and criminal groups. The main argument claims that civilian demand for criminal governance arises from the population’s need to meet basic necessities, while security needs motivate criminals to supply governance to reduce disruptions from potential civilian resistance. The theory focuses on three main factors as depicted in Figure 1: economic difficulties, articulation/resistance capacity, and government response. For clarity, the diagram presents each factor as “low” or “high,” though they should be understood as a continuum of varying intensity. While the factors are presented sequentially, we assume no specific order in which civilians or criminals consider them. We first outline these factors, their assumptions, and scope conditions before illustrating how they shape the demand and supply of criminal governance.

Figure 1. Demand and Supply of Criminal Governance

First, the theory focuses on individuals experiencing economic difficulties.Footnote ² Severe economic crises have several implications: they motivate individuals to seek help; represent the extent of government assistance required; and indicate the investments required from criminals if they provide aid.

The second factor is the civilian capacity to articulate demands and resist organized crime. The articulation/resistance capacity depends on both collective action and mobilization resources. High collective action enables individuals to articulate their demands to the state and, if necessary, to criminal groups. Collective action also facilitates peaceful or violent resistance against criminal groups (Arjona Reference Arjona2016; Reference Arjona2017; Kaplan Reference Kaplan2017; Moncada Reference Moncada2022; Osorio, Schubiger, and Weintraub Reference Osorio, Schubiger and Weintraub2020). In addition, as McCarthy and Zald (Reference McCarthy and Zald1977) indicate, sustaining collective resistance requires considerable mobilization resources.Footnote ³ Individuals with collective action skills with sufficient resources are more likely to sustain their demands. Civilian articulation/resistance capacity is also relevant to criminal groups as they are concerned about civilians resisting their illicit activities.

Finally, the model considers government responses to the economic shock. If the government addresses the population’s needs, individuals are less likely to seek help from criminals. Conversely, if authorities neglect the population, people may turn to criminals for help. Additionally, government responses influence criminal behavior in terms of governance supply, as outlined in the theory.

Following the Empirical Implications of Theoretical Models (EITM) approach (de Marchi et al. Reference de Marchi, Gailmard, Guggenheim, Lupia and Penn2017; Granato, Lo, and Wong Reference Granato, Lo and Sunny Wong2021), the theoretical development explicitly outlines the scope conditions and assumptions, and derives multiple observable implications in terms of hypotheses and scenarios. The following scope conditions circumscribe the explanatory limits of the theory. First, this criminal governance model focuses on economic aid, thus leaving out a broader range of governance activities, such as dispute resolution, public goods, or taxation (Arjona Reference Arjona2016; Blattman et al. Reference Blattman, Duncan, Lessing and Tobón2019; Lessing Reference Lessing2021). Second, the theory primarily applies to economic crises as criminal governance may be more salient during severe economic difficulties than under economic growth or stability (Dix-Carneiro, Soares, and Ulyssea Reference Dix-Carneiro, Soares and Ulyssea2018; Dube and Vargas Reference Dube and Vargas2013). Third, the theory requires pre-existing criminal groups with violence specialization (Bates, Greif, and Singh Reference Bates, Greif and Singh2002; Skaperdas Reference Skaperdas2002). Fourth, the theory explains better nonlethal forms of coercion (threats, lockdowns, or curfews) than intense violence (torture or mass killings) or performative violence (Lantz Reference Lantz2016; Fujii Reference Fujii2021). Fifth, departing from the Weberian ideal state, the theory operates in contexts of partial state presence due to uneven subnational state capacity, political interest, corruption, or institutional development (Arias Reference Arias2017; Astorga Reference Astorga2005; Bull and Hoelscher Reference Bull and Hoelscher2023; Lessing Reference Lessing2021; O’Donnell Reference O’Donnell1993; Olson Reference Olson2000; Tilly Reference Tilly, Evans, Rueschemeyer and Skocpol1985). Finally, although some criminals engage in transnational activities or have international connections (Brown et al. Reference Brown, Montero, Schmidt-Padilla and Sviatschi2025; Cruz Reference Cruz2010; Paarlberg Reference Paarlberg2022; Varese Reference Varese2013), the theory focuses exclusively on the domestic governance activities they deploy in their home communities.

The theory rests on the following assumptions. A1: Civilians and criminals are rational actors maximizing their economic well-being. Most civilians primarily rely on licit income sources, while criminals thrive in illicit markets.Footnote ⁴ To maximize their income, criminals prefer to operate in territories with cooperative populations and try to prevent civilian resistance.Footnote ⁵ A2: Civilians prefer asking the government for help before asking criminals for support. This assumes higher government legitimacy among the population than criminals. However, this assumption may not hold under high state repression and predation where criminals serve as protectors (Magaloni, Franco-Vivanco, and Melo Reference Magaloni, Franco-Vivanco and Melo2020; Wolff Reference Wolff2015) or for rebel governance where people may have an affinity (ideological, ethnic, and religious) toward rebels and animosity toward the government. A3: Criminal groups exercise governance with mixed portfolios comprising a combination of material aid and violent coercion. Although some criminals may procure a variety of goods and services to the population (Bull and Hoelscher Reference Bull and Hoelscher2023; Lessing and Willis Reference Lessing and Willis2019; Magaloni et al. Reference Magaloni, Robles, Matanock, Diaz-Cayeros and Romero2020; Olson Reference Olson2000), they are primarily violence specialists (Bates, Greif, and Singh Reference Bates, Greif and Singh2002; Volkov Reference Volkov2002) and may find it cheaper to use coercion than to provide aid.

EXPERIMENTAL DESIGN

This study uses a set of list experiments conducted in Mexico, Honduras, Guatemala, and El Salvador to estimate the bottom-up demand for criminal governance and the top-down supply of aid and lockdowns during the pandemic. List experiments are often used to unobtrusively ask about sensitive topics when people are expected to systematically misrepresent their responses (Blair and Imai Reference Blair and Imai2012; Blair, Coppock, and Moor Reference Blair, Coppock and Moor2020; Imai Reference Imai2011; Lyall, Blair, and Imai Reference Lyall, Blair and Imai2013; Magaloni et al. Reference Magaloni, Franco-Vivanco and Melo2020). In this case, misrepresentation could stem from fear in areas controlled by criminal groups or social desirability related to illegality or poverty.

Table 1 presents the three list experiments. Experiment 1 estimates the civilian demand for aid (DA), while Experiments 2 and 3 estimate the supply of aid (SA) and supply of criminal lockdowns (SL), respectively. All three experiments use a 1×2 design, a list of $ J=4 $ placebo items, and one sensitive item. We use the same list experiments in the four countries but vary the terms used to refer to “organized criminal groups” (grupos del crimen organizado) in Mexico, and “street gangs” (maras) in Honduras, Guatemala, and El Salvador, to reflect colloquial language.

Table 1. List Experiments Design

Note: Text in italics represents the sensitive item in the treatment group.

The statistical assessment uses two strategies. First, the univariate difference-in-means estimator stated in Equation 1 calculates the average treatment effect of the sensitive item:

⁽¹⁾ $$ \begin{array}{rl}{y}_{cei}={\alpha}_{ce}+\beta {T}_{cei}+{\epsilon}_{cei},& \end{array} $$

where $ {y}_{cei} $ is the item count by individual i in experiment e conducted in country c, where $ e=\{DA,SA,SL\} $ represents the experiments on the demand for aid, supply of aid, and supply of lockdowns, and $ c=\{M,H,G,S\} $ for Mexico (M), Honduras (H), Guatemala (G), and El Salvador (S). Variable $ {T}_{cei} $ indicates the random treatment assignment for individuals in each experiment-country; $ {\alpha}_{ce} $ is the intercept; and $ {\epsilon}_{cei} $ are the disturbances. We analyze each list experiment and country separately.

Following Imai (Reference Imai2011), Equation 2 uses a multivariate regression including respondent characteristics:

(2) $$ \begin{array}{rl}{y}_{cei}={\alpha}_{ce}+{\beta}_{cek}{T}_{cei}{X}_{ceik}+\gamma {X}_{ceik}+{\epsilon}_{cei},& \end{array} $$

where $ {y}_{cei} $ is the number of items reported by individual (i) in each experiment (e) in any given country (c); $ {X}_{ceik} $ is a matrix of individual characteristics (k); the coefficients $ {\beta}_{cek} $ represent the interaction between the treatment and individual-level characteristics; $ \gamma $ is the individual-level effects; and $ {\epsilon}_{cei} $ are the error terms. The multivariate regression analyzes each experiment-country separately rather than pooling the data.Footnote ¹¹

Note that the individual-level correlates are not causally identified. Section A2 of the Supplementary Material presents the methodological details of the list experiments and Section A3 of the Supplementary Material presents the power analysis. Following Blair and Imai (Reference Blair and Imai2012), we evaluate the list experiment assumptions and identify the presence of floor effects (see Section A7 of the Supplementary Material), which consist of respondents reporting a low number of items in both treatment and control groups. To address this issue, we implement Blair, Chou, and Imai’s (Reference Blair, Chou and Imai2019) recommendation and control for floor effects in all the models. In line with Blair and Imai (Reference Blair and Imai2012), we use maximum likelihood rather than the standard linear model estimator in both Equation 1 and Equation 2 to generate more efficient estimates, which is a preferred strategy for modeling floor effects. Note that while experiments 1 and 3 ask about respondents’ personal engagement with criminal governance, experiment 2 inquires about respondents’ witnessing governance behaviors in their neighborhoods. This issue is discussed in the “Limitations” section and in Section A2 of the Supplementary Material.

SURVEY DATA

The study analyzes the demand and supply of criminal governance using nationally representative surveys conducted face-to-face in Mexico, Honduras, Guatemala, and El Salvador using the same structured closed-ended questionnaire across countries. The samples comprise 1,387 interviews in Mexico, 1,234 in Honduras, 1,212 in Guatemala, and 1,200 in El Salvador (Section A4 of the Supplementary Material). The case selection used two criteria. First, we selected countries affected by endemic high levels but distinct types of organized criminal activity with the expectation that the survey instrument be able to detect the demand and supply of criminal governance. Second, we considered theoretically relevant characteristics, such as the COVID pandemic, economic conditions, social resilience, and state capacity (see Section A1 of the Supplementary Material for details).

The list experiments in Table 1 are the study’s three dependent variables. First, the Demand for Aid represents the proportion of the population asking criminal groups for help to solve a problem during the pandemic. Second, the Supply of Aid indicates the proportion of the population who saw criminal groups delivering money, medicine, or food during the pandemic. Finally, the Supply of Lockdowns indicates the share of the population who decided to stay at home in lockdown due to criminal threats.

The key independent variable of each experiment is the treatment manipulation $ {T}_{cei} $ indicated in Equations 1 and 2, taking the values of $ {T}_i=(0,1) $ for individuals in the control and treatment groups, respectively. For the multivariate model in Equation 2, the regression includes the respondent’s economic characteristics, social capital, perception of government authorities, degree of criminal control, and other individual characteristics.

To assess the economic effects of the pandemic, the Economic Difficulties index captures the sum of dichotomous responses to the following questions: “During the pandemic, has anyone in your family lost their job or experienced significant financial hardship?”; “Could you tell me if during the pandemic you had to cover hospital or funeral expenses for a family member or acquaintance?”; and “During the pandemic, due to lack of money or other resources, did your household ever run out of food?”

Civilian articulation/resistance capacity comprises the degree of collective action and the availability of mobilization resources. To measure collective action, variable Social Organization takes the value of 1 when respondents answer affirmatively to the question “Are you an active member of any type of social, religious, political, sports, business, or labor organization?” and 0 otherwise. In addition, Trust is a Likert indicator of social capital based on the question “Speaking of people in your community, would you say that the people in your community are very trustworthy, somewhat trustworthy, not very trustworthy, or not trustworthy at all?” To measure the availability of mobilization resources, we use a specific measure of income. Given the prevailing commensurability challenges of measuring income in cross-national surveys and missing data problems (Donnelly and Pop-Eleches Reference Donnelly and Pop-Eleches2018; Korinek, Mistiaen, and Ravallion Reference Korinek, Mistiaen and Ravallion2006), we use the number of light bulbs as a proxy for income as recommended by the Mexican Statistics Bureau (INEGI 2020). Under the intuition that wealthy households have more light bulbs than poor ones, this measure reduces systematic nonresponses and facilitates cross-country comparisons.Footnote ¹²

The study includes three measures of government performance. First, Government Performance is a pandemic-specific Likert measure responding to the question “Thinking about the beginning of the pandemic, how would you rate the services that the government offered to you or your family to face the pandemic?” and ranges from “not effective at all” to “very effective.” Second, the Government Effectiveness index captures the respondents’ perception that the government can help them solve a problem. To measure this, the survey asked “If you had a problem or necessity, how effective do you think (insert institution) would be in helping you resolve your situation?” and inquired about the Army, the Police, the Judicial System, and the Municipal government. Each institution is evaluated with a Likert scale ranging from “not effective at all” to “very effective.” The Government Effectiveness Index calculates the average responses for these institutions. Third, the Welfare Index reflects the extent to which individuals benefit from welfare programs. This index is the sum of the dichotomous responses to the question “Could you tell me if you or any member of your family is a beneficiary of any of the following types of welfare programs?” which include four different programs: food stamps, housing, education, and medicine and health services.

The survey also measures respondent’s perceptions about the level of Territorial Control by criminal groups through the question “Thinking about the degree of territorial control that criminal groups could have, which of the following scenarios best describes the situation in your community?”; responses range from “absent,” “emerging,” “competitive,” and “dominant.” By measuring perceptions of criminal territorial control directly in the survey, the study avoids imputing municipal-level measures of criminal violence as proxies of territorial control (Magaloni et al. Reference Magaloni, Franco-Vivanco and Melo2020), which could lead to ecological fallacies. Finally, the analysis includes individual characteristics, such as Age, measured as an ordinal variable, and Female, taking the value of 1 when respondents self-identify as female, and 0 as male. Section A5 of the Supplementary Material presents the descriptive statistics and Section A6 of the Supplementary Material presents the balance test.

The researchers gave paramount consideration to a variety of ethical issues and followed the guidelines of the University of Arizona Institutional Review Board (Section A12 of the Supplementary Material). Regarding public health, the researchers deployed the surveys only after government authorities had lifted most COVID-related restrictions, and enumerators in the field maintained strict health and security protocols. Since the research involved some high crime areas, enumerators requested permission from the government and local stakeholders with de facto authority to enter their communities. To minimize discomfort or risks to respondents, the survey used list experiments to inquire about sensitive topics. Due to ethical considerations, the survey did not inquire about the names of specific criminal groups or their activities.

EXTENT OF THE DEMAND AND SUPPLY OF CRIMINAL GOVERNANCE

To evaluate the overall levels of demand and supply of criminal governance, the empirical strategy relies on the difference-in-means estimator indicated in Equation 1. Figure 2 presents the coefficient plots derived from the list experiments. Section A8.1 of the Supplementary Material reports the detailed results. According to the estimates, Guatemala and Mexico show lower demand for criminal assistance than Honduras and El Salvador during the pandemic. In particular, the percentage of the population asking for support from criminal groups in Guatemala is 7.2% and 8.8% in Mexico. In contrast, the bottom-up demand for criminal aid in Honduras and El Salvador is substantially larger, with 14.1% and 14.9%, respectively. Section A11 of the Supplementary Material presents additional descriptive patterns.

Figure 2. Extent of Demand and Supply of Criminal Governance

Note: Predicted average treatment effects from Table 13 in the Supplementary Material.

The pandemic-induced economic crisis hit hardest the most vulnerable sectors, primarily those living on day-to-day income. In Guadalajara, Mexico, a man left out from government support said that “he has no other option than asking the cartels for groceries” (Mesas Reference Mesas2020). However, not all aid demands came from individuals, sometimes groups advanced articulated demands. San Salvador’s downtown concentrates about 40,000 street vendors and their leaders have developed close connections with the gangs (Papadovassilakis and Dudley Reference Papadovassilakis and Dudley2020), which allowed them to request a reduction or a pause in the extortion charges, known as la renta (“the rent”), due to the pandemic. As a Salvadorian journalist puts it, “the gang dies without the community. But in times like this, the community needs the gang” (Avelar Reference Avelar2020).

Regarding the supply of aid, Mexico and Guatemala experience the highest levels of aid supply, while Honduras and El Salvador report a lower supply of aid. The share of the population observing criminal groups deliver money, medicines, or food in Mexico is 22% and 20.3% in Guatemala; in contrast, only 13.6% of respondents in Honduras and 11.2% in El Salvador report witnessing aid from criminals.

There are innumerable reports of criminal groups delivering aid during the pandemic. Alejandra Guzmán, daughter of Joaquín “El Chapo” Guzmán, leader of the Sinaloa Cartel in Mexico, delivered despensas (grocery bags) to nursing homes and impoverished communities in Guadalajara (Reza Reference Reza2020). Several other groups, such as the Jalisco Cartel New Generation (CJNG), the Gulf Cartel, Los Zetas, and Los Viagras, publicized videos and pictures of fully armed sicarios (hitmen) delivering food, medicines, and money (Dittmar Reference Dittmar2020). In Guatemala, mareros (gang members) of the Barrio 18 delivered face masks to hospitals (Gutiérrez Reference Gutiérrez2020). Mareros of the Mara Salvatrucha 13 (MS13) delivered grocery bags in a churchyard North of San Salvador in El Salvador (Avelar Reference Avelar2020), and food and medicine in bordos (slums) in San Pedro Sula, Honduras.

The analysis of coercive governance indicates that criminal groups imposed relatively similar levels of lockdowns during the pandemic. In particular, criminals forced about 17.9% of the population in Mexico to stay in lockdown, while 16.8% of the population in El Salvador, 14.1% in Guatemala, and 12.1% in Honduras experienced similar situations.

News articles report criminal groups imposing lockdowns in their communities. In Mexico, criminal organizations, such as the Sinaloa Cartel, the Gulf Cartel, Grupo Sombra, and CJNG, imposed curfews throughout the country (Lara Reference Lara2020). In Ciliacán, Sinaloa, Los Chapitos, sons of El Chapo made a recording telling the population “no estamos jugando” (“we are not playing”), threatening to punish those who defied the curfew (Redacción 2020). An MS13 ranflero (gang leader) in El Salvador made the lesson clear by posting a video that went viral beating a man in the back of the legs with a baseball bat for violating the lockdown (Martinez, Martinez, and Lemus Reference Martinez, Martinez and Lemus2020).

These list experiments provide three valuable contributions. First, results reveal a considerable bottom-up demand for criminal aid in Mexico, Honduras, Guatemala, and El Salvador. This finding helps to advance research on civilian-agency in armed governance beyond community resistance, while recognizing civilian agency to participate in the development of local governance. Second, results also present systematic evidence about the magnitude of criminal aid and lockdowns beyond anecdotal reports. Instead of being isolated incidents, the estimates show that criminal aid and lockdowns are prevalent dimensions of criminal governance in these countries. Finally, criminal groups use both carrots and sticks in mixed portfolios of aid and lockdowns that cannot be easily classified as purely benevolent or entirely coercive, a topic discussed in further detail below. The following section uses multivariate analysis to identify the conditions associated with the demand and supply of criminal governance.

DETERMINANTS OF THE DEMAND AND SUPPLY OF CRIMINAL GOVERNANCE

To assess the individual-level correlates influencing the demand and supply of criminal governance, the study uses the multivariate maximum likelihood estimator from Equation 2. Since the list experiment coefficients are not intuitively interpretable (see full results in Section A8.2 of the Supplementary Material), Table 2 presents the predicted marginal effects as the difference between the percentages predicted using each variable’s maximum and minimum values, with the p-value of their respective difference.

Table 2. Multivariate Predicted Marginal Effects

Note: Coefficients indicate predicted marginal effects as the difference of percentages derived from the maximum and minimum values of each variable. Full estimates in Section A8.2 of the Supplementary Material. p-values: $ {}^{+}p< $ 0.1; $ {}^{*}p< $ 0.05; $ {}^{**}p< $ 0.01; $ {}^{***}p< $ 0.001.

Demand for Aid

Panel a in Table 2 presents the characteristics shaping the demand for criminal aid. Based on the human needs hierarchy, results suggest that the demand for aid stems from an individual’s urge to appease their primary necessities. In line with hypothesis HDA $ {}_1 $ , results present robust evidence that high economic difficulties increase bottom-up demands for criminal assistance from the population (Figure 3a). As the economic crisis ravages the most vulnerable, those facing acute economic needs are likely to ask criminals for help, which is consistent with the hierarchy of primary needs. In addition, results show that civilian articulation capacity shapes demands for aid. According to this finding, individuals active in social organizations are more likely to demand criminal aid in most countries, as expected from hypothesis HDA $ {}_2 $ (Figure 3b). However, interpersonal trust is not a strong predictor of aid demands. Mobilization resources also seem to drive demand for criminal governance as indicated in hypothesis HDA $ {}_2 $ , but only in Mexico and Guatemala (Figure 3c). In contrast to hypothesis HDA $ {}_3 $ , the government’s response is not a key determinant of aid demands. According to the model, regardless of the government’s performance during the pandemic, perceptions of government effectiveness, or welfare provision, the bottom-up demands for criminal aid are primarily driven by economic difficulties and civilian articulation capacity. Section A9.1 of the Supplementary Material evaluates the conditional expectations of hypotheses HDA $ {}_2 $ and HDA $ {}_4 $ .

Figure 3. Demand for Criminal Aid

Note: Predicted marginal effects of selected variables from panel a in Table 2.

Contrary to the territorial control paradigm, results show that the degree of gang territorial control only explains the demand for criminal aid in Honduras, but not in other countries (Section A10.1 of the Supplementary Material). Additional analysis also shows that criminal competition does not drive demand either (Section A10.2 of the Supplementary Material). Finally, results show that age is an important predictor of aid demand with younger people being more inclined to ask criminals for help. This relationship could be explained by young people’s economic vulnerability and recruitment risks (Baron Reference Baron2004; Densley Reference Densley2012), which is consistent with hypothesis HDA $ {}_1 $ (see Section A10.3 of the Supplementary Material for further examination on age).

Supply of Aid

Panel b in Table 2 presents the results of the supply of aid. Results indicate that high economic difficulties do not drive the provision of criminal aid. Although hypothesis HSA $ {}_1 $ expects a negative relationship between economic difficulties and criminal aid supply, the null result indicates that criminals primarily neglect people’s economic needs, thus challenging claims portraying criminals as benefactors or loyalty seekers.

In line with the argument that security needs prompt criminal governance, results show that civilian resistance capacity shapes the supply of aid. Related to collective action, social organization increases the provision of criminal aid in three out of the four countries as stated in hypothesis HSA $ {}_2 $ (Figure 4a), and high levels of interpersonal trust increase the supply of aid in Guatemala and Honduras (see Figure 4b). The availability of resources from hypothesis HSA $ {}_2 $ also increases criminal aid supply in three countries (Figure 4c). These results suggest that organized criminals tend to strategically supply aid to communities that have the potential to resist their illicit activities regardless of the population’s economic needs. This suggests evidence of criminals buying the hearts and minds of the population cheaply.

Figure 4. Supply of Criminal Aid

Note: Predicted marginal effects of selected variables from panel b in Table 2.

Results also provide evidence of criminal groups competing with the state in providing aid to the population. Based on hypothesis HSA $ {}_2 $ , high government pandemic performance is associated with a high supply of aid in three countries (Figure 4c). This evidence suggests criminal groups are trying to compete against the state in providing aid when it is cheap to do so. Section A9.2 of the Supplementary Material evaluates the conditional expectations of hypotheses HSA $ {}_2 $ and HSA $ {}_3 $ .

The theoretical expectation of criminals benefiting the population in areas where they have tight control over territory only finds support in Mexico and Honduras. However, societal and governmental explanations have more leverage than territorial control to account for the supply of aid. Section A10.1 of the Supplementary Material further explores territorial dynamics and Section A10.2 of the Supplementary Material analyzes criminal competition. Finally, results also indicate that age is a relevant factor in explaining the supply of aid. This is not surprising as young people are common targets of criminal and rebel recruitment (Beber and Blattman Reference Beber and Blattman2013; Densley Reference Densley2012). See Section A10.3 of the Supplementary Material for more details about age. Overall, the analysis shows that criminals largely neglect the economic needs of the population and they primarily supply aid in a strategic effort to neutralize potential pockets of resistance or to compete against the state in the provision of aid.

Supply of Lockdowns

Finally, panel c in Table 2 reports the results of coercive criminal governance. As stated in hypothesis HSL $ {}_1 $ , individuals facing economic difficulties are more likely to experience criminal lockdowns (see Figure 5a). This harsh criminal coercion in the context of high economic difficulties sharply contrasts with the null relationship between economic difficulties and criminal aid, thus suggesting that criminals revert to coercion when it becomes too expensive to assist the population. This also reveals the indifference of criminal groups coercing the population when they are going through economic crisis.

Figure 5. Supply of Criminal Lockdowns

Note: Predicted marginal effects of selected variables from panel c in Table 2.

Civilian resistance capacity also helps to explain criminal lockdowns. However, the availability of mobilization resources seems more relevant to account for this dimension of criminal governance than the respondent’s collective action capacity. The analysis only finds support in Mexico and Guatemala for hypothesis HSL $ {}_2 $ related to social organizations increasing criminal lockdowns (see Figure 5b). In contrast, the availability of resources stated in hypothesis HSL $ {}_2 $ is a robust predictor of criminal lockdowns across cases (see Figure 5c). These results seem to support the argument that criminals use lockdowns to curtail the resistance capacity of the population.

According to the results, there is moderate support for the role of the government in shaping the supply of lockdowns. As expected in hypothesis HSL $ {}_3 $ , positive perceptions of government pandemic performance are associated with high levels of criminal coercion in Mexico and Guatemala, and the association is barely statistically significant in El Salvador (see Figure 5d). In addition, the overall perception of government effectiveness is only relevant in the Mexican case. Section A9.3 of the Supplementary Material evaluates the conditional expectations of hypotheses HSL $ {}_2 $ and HSL $ {}_3 $ .

Contrary to the expectation of criminals exercising their comparative advantage in the use of violence in territories where they hold monopolistic control, results find weak support for this relationship and only in two cases (Section A10.1 of the Supplementary Material). In addition, criminal competition does not seem to explain lockdowns (Section A10.2 of the Supplementary Material). Finally, the model indicates that young people are consistently more likely to suffer criminal coercion than older respondents (Section A10.3 of the Supplementary Material).

In general, results suggest that criminal organizations are more likely to impose lockdowns on individuals experiencing economic difficulties and those who have a high capacity to resist their illicit activities. In contrast, the role of the government in explaining criminal coercion is only relevant in half of the cases. These results indicate that criminals primarily use coercion to suppress potential pockets of resistance, while there is only partial evidence of criminals suppressing state-civilian collaborations.

MIXED PORTFOLIOS OF CRIMINAL GOVERNANCE

Other areas of research have shown that political actors not only provide benefits or coercion as mutually exclusive strategies, but sometimes mix them. For example, research on electoral mobilization shows that political parties sometimes use vote buying (González-Ocantos et al. Reference González-Ocantos, De Jonge, Meléndez, Osorio and Nickerson2012) or engage in electoral violence (Birch, Daxecker, and Höglund Reference Birch, Daxecker and Höglund2020), but they also combine both strategies (González-Ocantos et al. Reference González-Ocantos, de Jonge, Meléndez, Nickerson and Osorio2020). Similar mixed strategies of assistance and predation have also been observed in criminal groups (Magaloni, Franco-Vivanco, and Melo Reference Magaloni, Franco-Vivanco and Melo2020). This section explores a set of mixed portfolio scenarios (MPSs) of criminal governance integrating the supply of aid and lockdowns. Based on the key factors outlined in the theoretical section (see Figure 1b), the following MPS explores the joint supply of aid and lockdowns under different combinations of economic conditions, resistance capacity, and government support:

• • MPS $ {}_1 $ Neglect: Regardless of the economic crisis severity and the effectiveness of government support, individuals with low resistance capacity are likely to receive low levels of criminal aid and lockdowns.

• • MPS $ {}_2 $ Cheap hearts and minds: Individuals going through low economic difficulties, embedded with high resistance capacity, and receiving low government assistance likely receive a high supply of aid and low lockdown levels.

• • MPS $ {}_3 $ Outcompeting the state: Individuals experiencing low economic difficulties with a high resistance capacity and receiving high government support are likely to receive high levels of criminal aid and low lockdown levels.

• • MPS $ {}_4 $ Curtailing potential resistance: Individuals suffering from high economic difficulties with a high resistance capacity and receiving low government assistance, likely receive high criminal lockdowns and low aid.

• • MPS $ {}_5 $ Suppressing civilian–state cooperation: Individuals suffering high economic difficulties with a high resistance capacity and receiving high government support are likely to receive higher levels of lockdowns and low criminal aid.

To analyze the combination of criminal aid and lockdowns, Figure 6 illustrates these MPS by presenting the supply of lockdowns on the vertical axis and the supply of aid on the horizontal axis. The dotted diagonal indicates an equal lockdown-aid allocation at different supply levels. The upper left triangular area above the diagonal represents more coercive portfolios, including a higher proportion of lockdowns relative to aid, while the lower right triangular area shows supportive portfolios containing more aid than lockdowns. Figure 6a presents the supply of governance during low economic difficulties and Figure 6b under high economic difficulties. Using results from panels b and c of Table 2, the point estimates along the arrow segments in Figure 6 represent the predicted combinations of lockdowns and criminal for each MPS derived from setting the variables of economic difficulties, articulation/resistance capacity, and government response according to the characteristics of each scenario outlined above.

Figure 6. Mixed Portfolio of Aid and Lockdowns

Note: Predicted effects from panels b and c in Table 2.

Based on these MPS, the statistical analysis shows that criminals deploy different lockdown-aid allocations depending on the conditions where they operate. By focusing first on Figure 6a where respondents undergo low economic difficulties, the first node in each country-segment in Figure 6a represents an MPS of neglect where individuals undergo low economic difficulties and have low resistance capacity. As expected from MPS $ {}_1 $ , this scenario reports the lowest lockdown-aid levels, with a slightly more coercive portfolio in Mexico, Honduras, and Guatemala, and a slightly more supportive portfolio in El Salvador. The second node in Figure 6a shows evidence of criminals buying the hearts and minds of the population cheaply. Here, civilians have a high resistance capacity and low government support which, according to the theoretical expectations, triggers security concerns for criminals. As a reaction, criminals substantially increase the governance supply as shown by the large diagonal of the first arrow segments. However, the steep inclination of these segments indicates more coercive portfolios than the ones predicted in MPS $ {}_2 $ . This suggests that security concerns may motivate criminals to substantially increase the use of sticks to send a clear deterrence signal while increasing the use of carrots at slightly lower rate to co-opt potential pockets of resistance. Finally, the last node in Figure 6a shows the scenario where criminals try to outcompete the state. Here, civilians have high resistance capacity and receive high government support. The low inclination of the second arrow segment indicates that criminals further increase the aid supply in a larger ratio than lockdowns, which is consistent with criminals outcompeting the state in MPS $ {}_3 $ . Results indicate that criminals in Mexico, Honduras, and Guatemala act according to this expectation, but that is not the case in El Salvador, where criminals become more coercive without increasing the levels of aid.

Figure 6b analyzes different lockdown-aid allocations under high economic difficulties. In general, results indicate that economic distress increases the supply of criminal coercion against the population. Evidence of this comes from the general upward displacement of all arrows in panel b as compared to panel a, which indicates a higher lockdown allocation relative to aid. This is consistent with the theoretical expectation that severe economic difficulties increase the population’s needs, making it too expensive for criminals to provide assistance and forcing them to rely on their coercive advantage as violent specialists by imposing lockdowns.

The first nodes in Figure 6b indicate a context of high economic difficulties and low resistance capacity. In this scenario, criminals largely neglect the population, as expected from MPS $ {}_1 $ , but with a more coercive tone. The second set of nodes reflects the reaction of criminals to communities endowed with high resistance capacity but neglected by the government. As expected from MPS $ {}_4 $ , security concerns motivate criminals to increase the lockdown-aid supply with a higher proportion of coercive lockdowns to curtail potential civilian resistance. This is indicated by the pronounced inclination of the first arrow segments in Figure 6b. Finally, the last nodes in this panel portray the mixed portfolio under high economic difficulties affecting individuals endowed with high resistance capacity that receive high government support. Results support the expectation of MPS $ {}_5 $ predicting the suppression of state-civilian cooperation with a high allocation of lockdowns and a lower proportion of aid. Results indicate that suppression of state-civilian cooperation is particularly harsh in Mexico and El Salvador, while communities in Honduras and Guatemala receive slightly more aid. In particular, the almost-flat inclination of the second segment in Honduras reveals some efforts of criminal groups to outcompete the state in this country as in MPS $ {}_3 $ .

Overall, the mixed portfolio analysis of lockdowns and aid shows the complex decision-making behind the supply of criminal governance. In line with the theoretical framework, results suggest that criminals tend to strategically provide aid in order to neutralize potential resistance when it is cheap to do so. In a similar way, criminals tend to compete against the state in the provision of aid when it is relatively cheap to do it. However, as the severity of the economic crisis increases, the cost of assisting the population, criminals use their comparative advantage in the use of violence to increase the imposition of lockdowns.

LIMITATIONS

Despite its contributions, this research has important shortcomings. The relevance of civilian articulation/resistance capacity focuses at the individual level. However, the study does not explain the aggregation of behaviors from individual to collective action. Future research should study these aggregation mechanisms by shifting from individual level responses to network analysis or qualitative analysis at the community level. Such approach would help to understand how individual characteristics shape community behavior and how criminals respond to it.

The study recognizes that not all criminal groups are created equal. Important variations in leadership styles, organizational structures, income sources, and illicit market characteristics likely shape the supply of governance by different criminal groups. Unfortunately, the empirical strategy does not permit a disaggregated analysis of specific criminal groups operating within each country. Additionally, the empirical strategy does not disaggregate other types of armed actors that may coexist with criminals in the territory, such as self-defense forces or rebel organizations. Unfortunately, investigating the behavior of specific armed groups by identifying their names and activities poses considerable ethical challenges. Future work should analyze governance behaviors across different organizations in an empirically rigorous and ethically sound way.

The narrow focus on criminal or government assistance downplays other potential providers of community support. Although the list experiments consider support from religious groups, civil society organizations, or international actors in the placebo items, these providers are not theorized or empirically analyzed. Future works should go beyond the civilians–criminals–state triad and take into account additional actors.

Despite the list experiments eliciting more truthful responses related to criminal governance, the factors associated with these behaviors are based on statistical correlation rather than causal identification. Future research should rigorously explore the causal determinants of criminal governance using rigorous causal inference research designs.

While the questions about the demand for aid and supply of lockdowns provide direct measures of behaviors conducted by the respondents, the question measuring the supply of aid reports behaviors that respondents observed in their neighborhoods, but not direct actions conducted by them. This opens the potential for ecological fallacy in the interpretation of results for the supply of aid. This implies that respondents with certain characteristics (individual-level features) could be more likely to observe criminal handouts in their neighborhood (group-level features), but such characteristics may not necessarily make them individual targets of criminal aid. This challenge could be addressed by directly asking respondents about their personal benefit from criminal aid. However, the authors considered such approach too obtrusive and potentially dangerous. Directly asking individuals about personally receiving gifts or favors from criminals is likely to indicate some sort of association with criminal groups. Given that beneficiaries of criminal gifts are expected to pay back in kind or in loyalty (Varese Reference Varese2020), providing evidence of this type of linkage with a direct question may put respondents and interviewers at risk from law enforcement or rival criminal groups.Footnote ¹³ Based on these ethical concerns, the experimental design focuses on measuring the supply of aid at the neighborhood level.

Finally, future works should go beyond the scope conditions of this study and analyze other governance behaviors, such as dispute resolution or public goods provision. Natural disasters and climate change could also help to analyze criminal governance beyond the pandemic or economic shocks.