Gangs, labor mobility, and development

#672 | 2025 Gangs, labor mobility, and development NIKITA MELNIKOV, CARLOS SCHMIDT-PADILLA, MARÍA MICAELA SVIATSCHI Submitted to Econometrica GANGS, LABOR MOBILITY, AND DEVELOPMENT NIKITA MELNIKOV Nova School of Business and Economics CARLOS SCHMIDT-PADILLA Goldman School of Public Policy, University of California, Berkeley MARÍA MICAELA SVIATSCHI Department of Economics, Princeton University We study how criminal organizations affect economic development. We exploit a natural experiment in El Salvador, where these criminal organizations emerged due to an exogenous shift in American immigration policy that led to the deportation of gang leaders from the United States to El Salvador. Using a spatial regression discontinuity design that focuses on the gang-created system of borders, we find that individuals in gang-controlled neighborhoods have less material well-being, income, and education than individuals living only 50 meters away but outside of gang territory. None of these discontinuities existed before the arrival of the gangs. A key mechanism behind the results is that gangs restrict individuals’ mobility, affecting their labor-market options by preventing them from commuting to other parts of the city. The results are not determined by high rates of selective migration, differential exposure to extortion and violence, or differences in public goods provision. KEYWORDS: gangs, development, mobility, crime. 1. INTRODUCTION How do nonstate armed actors affect economic development? On the one hand, they can impede the state from providing public goods, enforcing property rights and contracts, and Nikita Melnikov: [email protected] Carlos Schmidt-Padilla: [email protected] María Micaela Sviatschi: [email protected] We thank Alicia Adsera, Cevat Giray Aksoy, Alberto Alesina, Sofia Amaral, Oriana Bandiera, Samuel Bazzi, Chris Blattman, Leah Boustan, Timothy Besley, Eli Berman, Ethan Bueno de Mesquita, Filipe Campante, Doris Chiang, Abby Córdova, Raúl Sanchez de la Sierra, Melissa Dell, Jennifer Doleac, Patricio Dominguez, John J. Donohue, Oeindrila Dube, Thad Dunning, Stefano Fiorin, Thomas Fujiwara, Tarek Ghani, Edward Glaeser, Jeff Grogger, Guy Grossman, Sergei Guriev, Gaurav Khanna, Asim Khwaja, Tom Kirchmaier, Ilyana Kuziemko, Leah Lakdawala, Horacio Larreguy, Benjamin Lessing, Nicola Limodio, Sarah Lowes, Stephen Machin, Luis Martínez, Atif Mian, Magne Mogstad, Chris Neilson, Sam Norris, Ben Olken, Daniel Ortega, Emily Owens, Rohini Pande, Paolo Pinotti, Oscar Pocasangre, Nishith Prakash, Stephen Redding, James Robinson, Mark Rosenzweig, Matteo Sandi, Rogério Santarrosa, Jacob Shapiro, Santiago Tobón, Daniel Treisman, Oliver Vanden Eynde, Juan Vargas, Leonard Wantchekon, Austin Wright, Nathaniel Young, Ekaterina Zhuravskaya, Owen Zidar, Fabrizio Zilibotti, and the participants of seminars and conferences at the AEA, AL CAPONE, APPAM, APSA, Berkeley, Bocconi, CERP, Conference on the Economics of Crime and Justice, the EBRD, ESOC, Harvard, the IDB, ifo Institute, LACEA, LSE, MIEDC, MIT, NBER SI, PSE, Princeton, Sciences Po, Stanford, University of Chicago, University of Connecticut, University of Munich, University of Passau, World Bank Group, and Yale for helpful comments and suggestions. We also thank the International Crisis Group for helping us access parts of the data. Carlos Aguilar, Bruno Jiménez, Paulo Matos, Sarita Oré Quispe, Graciela Saca, and Édgar Sánchez-Cuevas provided excellent research assistance. 2 preventing violence (Acemoglu et al.,2001,Michalopoulos and Papaioannou,2013). On the other hand, if the state is weak and unable to control parts of its territory, nonstate armed actors may take on the role of the state in fulfilling essential institutional functions, potentially enabling economic growth (Tilly,1985,Olson,1993,Bates et al.,2002,Ibáñez et al.,2019, De la Sierra,2020) and competing for the “hearts and minds” of civilians (Ibáñez et al.,2019, De la Sierra,2020,Blattman et al.,2022). Overall, how and why nonstate armed actors affect development remains an open question. In this paper, we study how a specific type of nonstate armed actor—namely, criminal organizations—affects socioeconomic development. In urban areas in the developing world, millions of people live under some form of criminal governance (Lessing,2021,Blattman et al., 2022). Criminal organizations function mainly in urban centers, often controlling parts of the city, while other parts are controlled by the state. In particular, this paper analyzes how two of the world’s most prolific gangs—MS-13 (Mara Salvatrucha) and 18th Street (Barrio 18)— affected socioeconomic development in El Salvador.1 We exploit a natural experiment that took place in El Salvador. Before the mid-1990s, El Salvador had no significant criminal organizations. However, in 1996, after a shift in American immigration policy that made it easier to deport individuals—especially those with criminal backgrounds—back to their country of origin, many Salvadoran migrants who were members of California-based gangs (specifically, MS-13 and 18th Street) were deported back to El Salvador. These deported gang members reestablished their gangs in El Salvador and quickly gained control over certain parts of the country. To protect their territory from outsiders, the gangs also re-created a system of borders and checkpoints that they used to establish territorial dominance in California (Nuño and Maguire,2021), resulting in the division of urban areas between the gangs and the state. To estimate the effects of gangs’ territorial control, we use the boundaries of gang-controlled neighborhoods in El Salvador’s capital, San Salvador, to implement a spatial regression discontinuity design. These territorial demarcations were formed soon after the gang leaders arrived in 1996, and they roughly coincide with existing natural barriers, such as boulevards and highways. We measure the outcome variables using the 2007 census and our own geocoded survey, which we conducted in 2019 in both gang and nongang neighborhoods. Our results indicate that residents of gang-controlled neighborhoods in San Salvador have worse dwelling conditions, less income, and a lower probability of owning durable goods compared to individuals living just 50 meters away but outside of gang territory. They are also less likely to work in large firms. For instance, we find that residents of gang areas have $350 less monthly household income (the sample mean is $625) compared to individuals living in neighboring nongang locations and have a 12-percentage-point lower probability of working 1Both MS-13 and 18th Street also have a major presence in Honduras, Guatemala, and parts of Italy, Mexico, Spain, and the United States. Similar criminal organizations are also present in many other countries (e.g., Brazil, Colombia, Jamaica, and South Africa). GANGS, LABOR MOBILITY, AND DEVELOPMENT 3 in a firm with at least 100 employees. The results are highly robust to the choice of empirical specifications. These differences in living standards did not exist before the gang leaders arrived. We replicate the regression discontinuity design with data from the 1992 census to show that, before the gangs emerged, areas on both sides of the gang borders had similar socioeconomic and geographic characteristics, as well as similar levels of crime. These results are consistent with the fact that the boundaries of gang territory were not formed based on preexisting socioeconomic differences, but rather on the availability of natural barriers (i.e., major roads). We also show that the natural barriers are not associated with differences in socioeconomic conditions when they do not determine gang territorial control. An important mechanism through which gangs affect socioeconomic development in the neighborhoods they control is related to restrictions on individuals’ mobility. The gangs’ longterm survival depends on their ability to secure the borders of their territory and prevent the police and rival gang members from arresting or killing them. Therefore, to maintain secure control over their territory, both MS-13 and 18th Street instituted a system of checkpoints, not allowing individuals to freely enter or leave gang-controlled neighborhoods (ICG, 2018). The security of their territory also allows the gangs to use it as a bridgehead from which they conduct extortion raids to neighboring areas. Using the data from our geocoded survey, we perform a spatial regression discontinuity design to document the presence of restrictions on individuals’ mobility. We show that residents of gang areas are 50 percentage points more likely to work in gang territory compared to individuals living only 50 meters away but on the nongang side of the boundaries. They are also more likely to say that gang-imposed borders prevented them from getting jobs in large firms in other parts of the city, less likely to say that there is freedom of movement in the neighborhood where they live, and less likely to have been to places outside of San Salvador. However, those individuals do not have lower levels of mobility per se. Using cell phone ping data, we show that, while residents of gang-controlled neighborhoods are largely confined in their movements to gang territory, they travel the same distance as their peers on the other side of the gang boundaries. These mobility restrictions affect labor-market outcomes: residents of gang territory end up working in smaller firms and earning lower wages because they cannot commute to the areas where the largest and best-paying firms are located. Notably, labor-market conditions do not change directly at the boundaries of gang territory (i.e., there is no change in firm size, wages, profitability, or the number of business establishments). Instead, we show that, after the emergence of the gangs, new business establishments increasingly opened in areas far away from gang territory. Nonetheless, residents of nongang neighborhoods close to the boundaries were able to take advantage of these new labor-market opportunities by commuting to parts of the city where the largest firms are located, whereas individuals living in gang areas were prevented from doing so by the restrictions on their mobility. 4 Another factor limiting socioeconomic development in gang-controlled neighborhoods is related to educational attainment. Using school census data, we show that the annual school dropout rate is 2 percentage points higher in gang territory than in nongang areas. The differences in educational attainment contribute to further widening the income gap between gang and nongang territories. We also examine other potential determinants of lower socioeconomic development in gangcontrolled neighborhoods, but we find that, in this context, they cannot explain the results. In particular, we demonstrate that individuals and firms on both sides of the boundaries are equally exposed to extortion and other violent crimes. This result is explained by the fact that, since gang members are not subject to the same mobility restrictions as the other people living on their territory, they conduct regular raids into neighboring areas outside their immediate control. This result is fully consistent with the finding that labor-market conditions do not change directly at the boundaries of gang territory. Similarly, we find no differences in the availability and quality of public goods provision (e.g., schools and hospitals), consistent with the qualitative evidence suggesting that the government has been willing to provide public goods in gang areas to avoid ostracizing the residents of those locations.2In turn, because the gangs benefit from public goods provision in their neighborhoods, they have been willing to allow the government to provide (nonpolicerelated) services in the areas they control.3Finally, we show that the results are not driven by higher levels of unemployment (or informal employment) in gang-controlled neighborhoods and that selective migration of individuals across the boundaries of gang territory can explain no more than 14% of the gap in socioeconomic development between the gang and nongang neighborhoods. Finally, we use data from all of El Salvador to perform a difference-in-differences design that analyzes how gang presence affected the spatial allocation of economic activity in the country. We find that, after the arrival of the gangs, municipalities least exposed to gang activity experience significantly more openings of new business establishments, as well as higher growth in nighttime light density and household income. These results highlight how the economic costs of mobility restrictions increased over time: as employment opportunities improved in places 2In addition, the government and other political actors are motivated by electoral considerations: without providing public goods in gang-controlled neighborhoods, political parties would likely have been unable to campaign in those areas (e.g., see Córdova,2019). This stems from the client-broker relationship between the political parties and the gangs, particularly during elections. To campaign in gang-controlled neighborhoods, political parties need to provide public goods in those areas. 3We find that the gangs themselves provide very limited public services, the probability of which does not change at the boundaries of gang territory. This result may be different in other settings where nonstate actors have the resources and incentives to co-opt the population under their control (e.g., Magaloni et al.,2020b,Blattman et al., 2022). In particular, in San Salvador, the gangs might not provide more public services in their territories because the government has been willing to provide them. Salvadoran gangs also have limited financial resources (Martínez et al., 2016), making it difficult for them to compete for hearts and minds. However, in settings where the government is not present (e.g., in rural areas) and criminal organizations have the resources to provide services to the public (e.g., drug cartels), territorial control by nonstate actors may result in more public goods provision. GANGS, LABOR MOBILITY, AND DEVELOPMENT 5 without gang activity, it became increasingly important to be able to commute to work in those areas. Our paper is related to several strands of the existing literature. First, it contributes to the literature studying the origins and consequences of organized crime and other nonstate armed actors (e.g., Gambetta,1996,Frye and Zhuravskaya,2000,Bandiera,2003,Daniele and Marani, 2011,Acemoglu et al.,2013,Daniele and Geys,2015,Buonanno et al.,2015,2016,Dell,2015, Pinotti,2015,Daniele and Dipoppa,2017,De Feo and De Luca,2017,Acemoglu et al.,2019, Alesina et al.,2019,De la Sierra,2020,Murphy and Rossi,2020,Mirenda et al.,2022,Sviatschi,2022a,b). Most of this literature has focused on violence, or the potential thereof, as the channel behind the effects of organized crime on politics, investment, migration, and other aspects of socioeconomic development. We complement this literature by presenting novel evidence on one specific aspect of criminal organizations that is increasingly prevalent in the developing world: territorial control in urban settings. By looking at urban areas where the territory is divided between the state and the gangs, we document a previously ignored mechanism through which criminal organizations affect socioeconomic development: restrictions on mobility. As Glaeser and Sims (2015) point out, little is known about the consequences of crime in the urbanized, developing world. In these contexts, because criminal organizations constantly face the potential for territorial challenges both from rival criminal groups and from the state, they need to implement stringent security measures to protect the borders of the neighborhoods they control (e.g., imposing restrictions on individuals’ mobility). As a result, residents of these neighborhoods end up having significantly worse labor-market outcomes because of their inability to work in other parts of the city. Second, our paper is related to the literature on criminal governance and the organizational structure of criminal enterprises (Levitt and Venkatesh,2000,Skarbek,2011,Carvalho and Soares,2016,Ibáñez et al.,2019,Lessing and Willis,2019,Magaloni et al.,2020a,Lessing, 2021,Blattman et al.,2022). Much of the existing literature has shown how nonstate armed actors emerge to fill the void left by the state and provide security and other public goods to the local population in exchange for political influence (e.g., Blattman and Miguel,2010), taxation (e.g., Olson,1993,De la Sierra,2020), and the opportunity to conduct their illegal activities. Our paper analyzes how these relationships are altered in an urban context, where the proximity of the state, on the one hand, poses a threat to the gangs’ territorial control but, on the other hand, allows the gangs to rely on the provision of most public goods by the government.4 Third, our paper contributes to the literature studying the causes and consequences of the formation of extractive institutions, which can have a long-lasting impact on socioeconomic 4In particular, while the literature on stationary bandits would imply that armed actors have incentives for maximizing residents’ incomes—including through some public goods provision—to maximize extortion rents in the territory they control (e.g., Olson,1993,De la Sierra,2020), we provide novel evidence that this incentive can be undermined in an urban context where labor-market mobility is needed to maximize income. 6 development (e.g., Acemoglu et al.,2001,2002,Dell,2010,Michalopoulos and Papaioannou, 2013,Dell et al.,2018,Dell and Olken,2020,Lowes and Montero,2021). Specifically, we show how the deportation of criminal leaders from the United States to El Salvador has resulted in their establishing extortionary gangs that significantly limited socioeconomic development in El Salvador. It also contributes to a long-standing debate on whether individual leaders—in this case, gang leaders—affect economic growth in developing countries (Jones and Olken,2005). Finally, our work is related to the literature analyzing the economic effects of barriers to geographical mobility. The existing literature has focused on the effects of international borders (e.g., Clemons et al.,2008,McKenzie et al.,2010,Mergo,2016,Calì and Miaari,2018, Alsawady et al.,2022) and the absence of transportation infrastructure (e.g., Donaldson,2018, Asher and Novosad,2020). We complement this work by showing how gang-imposed restrictions on individuals’ freedom of movement can significantly affect socioeconomic development, even within an integrated metropolitan area and in the absence of direct transportation costs and legal borders. Given the global prevalence of similar intracountry barriers to mobility, our results provide important policy implications for many developing countries. In particular, nonstate armed actors restrict individuals’ freedom of movement in Brazil, Colombia, Guatemala, and Honduras (e.g., Ibáñez et al.,2019,Magaloni et al.,2020a); many other countries, too, experience various forms of mobility restrictions (e.g., see Walther et al.,2020). The rest of this paper is structured as follows. Section 2describes the rise and organization of criminal groups in El Salvador. Section 3describes the main data sources. Section 4presents the identification strategy and the main results. Section 5examines the mechanisms driving the results. Section 6analyzes the aggregate effects of gang presence. Section 7concludes. 2. HISTORICAL BACKGROUND In this section, we present an overview of how MS-13 and 18th Street developed in Salvadoran migrant communities in the United States and how criminal capital was exported from these communities to El Salvador following a shift in American immigration policy in 1996. We then describe how, once in El Salvador, the gangs quickly reestablished their criminal structures, began recruiting, and gained territorial control over many urban neighborhoods throughout the country, most notably in the capital, San Salvador. We also provide qualitative evidence on how the boundaries of gang territory were formed soon after the arrival of the criminal deportees, based on the system of territorial control that the gangs had developed in the United States. 2.1. The Origins of MS-13 and 18th Street Southern California, especially Los Angeles, became home for thousands of Salvadorans fleeing the country’s descent into civil war in the 1980s (Stanley,1987). Lacking an established support network, Salvadoran migrants lived in poor, overcrowded neighborhoods and often faced discrimination from other migrant groups (Brettell,2011). In a typical family, both parents worked, often leaving the children unsupervised (Savenije,2009). GANGS, LABOR MOBILITY, AND DEVELOPMENT 7 Left on their own and facing prejudice from other migrant groups and their gangs, some Salvadoran youth formed the precursors to MS-13—self-defense groups that were initially better known for petty crimes and for their affinity to cannabis and heavy metal, rather than for brutal violence—while others joined 18th Street, an existing Mexican gang (Dunn,2007,Cruz, 2010,Martínez and Martínez,2018). As membership in MS-13 and 18th Street grew across Salvadoran immigrant communities, the gangs became known to the local authorities. Some of their members were sent to prison, where they gained criminal capital and social connections that helped them solidify their structures (Womer and Bunker,2010,Martínez and Martínez, 2018). By the mid-1980s, both MS-13 and 18th Street had developed independent identities, organizational structures revolving around territory-based cliques (clicas), and a fierce rivalry that continues to this day (Ward,2013). Many gangs in 1980s Los Angeles shared a noteworthy trait: they precisely demarcated their territory, which greatly contributed to their identity and development (Coughlin and Venkatesh, 2003). For example, they used graffiti to define the territories under their control and to project authority over their rivals and the local population (Tita et al.,2005,Artsy,2018). This demarcation had a profound impact on the mobility and decisions of individuals living in gang territories: “One of the really important things to think about is how the invisible borders [...] add costs we often don’t think about. If I’m a young person growing up in a particular neighborhood [in Los Angeles] and the closest movie theater or the closest shopping mall is claimed by a rival gang, [...] I’m going to have to spend more time on a bus, put more gas in my car, to travel to other areas” (Artsy,2018). In an observational study of incarcerated MS-13 gang members in Los Angeles County, Nuño and Maguire (2021) highlight how “most MS-13 members are involved in cliques that claim certain turf or territory (96.3%) and would be willing to use violence to defend it against others (92.6%),” relying on graffiti and outposts to mark and control their territories.5This facet of gang culture became a fundamental trait of gang structures in El Salvador. 2.2. American Immigration Policy and the Emergence of Gangs in El Salvador In 1996, to reduce crime in urban areas and address the surge in irregular migration, the United States passed the Illegal Immigration Reform and Immigration Responsibility Act (IIRIRA) (Chacón,2009,Abrego et al.,2017). IIRIRA drastically increased immigration enforcement, creating procedures for expedited removal, adding new grounds for deportation, and increasing the number of border patrol agents. This shift in American policy had a profound impact on El Salvador. During the first wave of deportations in 1996, over 500 Salvadoran gang members were deported from the United States, leading to devastating changes in Salvadoran communities (Sviatschi,2022b). 5The territorial identity is so important that, when MS-13 and 18th Street expanded to El Salvador, many of the cliques there adopted names that referenced the locations where their gang leader commenced their illicit careers in the United States (e.g., Hollywood Locos Salvatruchos). 8 Given that they did not have criminal records in El Salvador, the repatriated gang members— many of whom were serving or had previously served sentences in the United States—gained their freedom after returning to their home country (Ward,2013). El Salvador was still recovering from its civil war, which ended in 1992, and the Salvadoran state did not have the resources to prevent the gangs from expanding. The 1992 Peace Accords mandated the creation of a new police force—the National Civil Police (Policía Nacional Civil, PNC)—and at the time of the repatriations, the structure of the PNC was still being defined (e.g., no rural police units existed until 2004). The repatriated gang leaders exploited this low level of state capacity and expanded their operations to many urban areas. Most of the repatriated MS-13 and 18th Street gang members had lived in the United States since a young age and knew little about their home country. For this reason, most of them returned to their birth municipalities, relying on their family networks to resettle in a new environment (DeCesare,1998,Sviatschi,2022b). Seeking social acceptance and status, the gang deportees banded together and tapped into local youth groups to replicate the gang structures they had in California. Even though only a few hundred gang members were repatriated from the United States in 1996, they quickly expanded their ranks, recruiting new members from the local population. Many locals were attracted by the camaraderie and respect that the gangs offered, others sought more tangible material gains such as money and drugs (Cruz and Portillo Peña,1998,Martínez and Martínez,2018). Sviatschi (2022b), in particular, shows how, after the MS-13 and 18th Street gang members arrived, and began recruiting adolescents to join their structures, El Salvador experienced an immediate increase in gang-related activities. According to the local authorities, by the end of 1996, at least 20 thousand individuals had joined the two gangs (Cruz and Portillo Peña,1998). 2.3. The Formation of Gang Territory in El Salvador Taking advantage of the postwar environment and widespread destitution, both MS-13 and 18th Street quickly expanded their influence over many neighborhoods, particularly in the capital, San Salvador, and other urban areas, “gain[ing] complete control of [certain] localities” (Zoethout,2015). This rapid formation and enforcement of boundaries was possible due to four main factors: (i) the gangs’ experience in implementing a system of territorial control in California, (ii) the importance of territorial control for the gangs’ identity and long-term survival, (iii) the gangs’ ability to recruit new members from the local population, and (iv) El Salvador’s low state capacity in the 1990s. The system of territorial control built upon the strategy the gangs honed in California, where demarcation, largely based on natural barriers, split urban areas into small geographical confines known as cliques (Miguel Cruz,2010). In El Salvador, the gangs also defined their territory based on natural barriers such as major roads and boulevards (Tenorio,2002,Vega,2015). We identify and take advantage of three such major roads (see Figure 1)—Bulevar Venezuela, 49 Avenida Sur, and Autopista Comalapa, all of which existed in 1996—that largely determined GANGS, LABOR MOBILITY, AND DEVELOPMENT 15 where the locations of the boundaries were determined by the presence of natural barriers that prevented the gangs from expanding further. We then use these natural boundaries of gang territory to verify that our results are not driven by the potential endogeneity of some of the other boundaries. The second assumption is that residents of gang territory did not selectively migrate from those areas to neighboring locations in the control group. Subsection 4.3 and Appendix Subsection A.2 provide a detailed discussion of this assumption, showing that selective migration can explain no more than 14% of the socioeconomic gaps between gang and nongang areas. 4.2. Main Results Table Ipresents the results of estimating Specification (1) using the 2007 census data. It shows that, after experiencing gang rule, individuals living in gang-controlled neighborhoods have significantly worse dwelling conditions, lower levels of education, and are less wealthy than their peers on the other side of the boundaries. For instance, residents of gang territory are estimated to have a 21-percentage-point lower probability of owning a car, a 15-percentagepoint lower probability of having a high school degree, and a 5-percentage-point lower probability of living in a house with concrete walls than individuals living less than 50 meters away but not under the control of gangs.20 The results for the other measures of socioeconomic development present the same pattern. Figure 2illustrates the findings from Table Ifor the first principal components of the dwelling, household, and individual characteristics. The vertical axis represents the average value of the outcomes variables; the horizontal axis represents distance (in meters) to the boundaries of gang territory. Areas to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. For all the outcome variables, there is a clear discontinuity at the boundaries of gang-controlled neighborhoods.21 Overall, the results suggest that gangs have had a significant negative effect on socioeconomic development in the neighborhoods they control. To estimate the total monetary cost of this effect, we consider a variable that potentially aggregates all the effects of living under gang control into one—household income—the data for which come from the 2019 survey. The left part of Appendix Figure A.4 presents the regression discontinuity plot for this variable. The results suggest that residents of gang neighborhoods earn approximately $350 less each month compared to residents of nongang areas. Given that the average monthly income in our sample is $625, this discontinuity implies a substantial reduction in earnings. Table A.I in the Ap20In the individual-level regressions, the sample consists of the entire population. The results are very similar if, instead, we analyze just the adult population. 21In the Appendix, we illustrate the results for all the other outcome variables from Table I. Figure A.1 presents the results for dwelling characteristics, Figure A.2 for individual characteristics, and Figure A.3 for household characteristics. 16 TABLE I SOCIOECONOMIC CONDITIONS AFTER EXPOSURE TO GANG CONTROL Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.047 0.026 -0.050 -0.079 0.006 -0.131 (0.015)*** (0.010)** (0.021)** (0.021)*** (0.002)*** (0.029)*** [0.017]*** [0.010]** [0.027]* [0.027]*** [0.003]** [0.039]*** Mean of dep. var. 0.932 0.028 0.941 0.108 0.005 0.181 Observations 72,087 60,675 62,169 62,169 62,169 59,776 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.013 -0.207 -0.135 -0.021 -0.173 -0.693 (0.006)** (0.046)*** (0.033)*** (0.006)*** (0.035)*** (0.195)*** [0.005]*** [0.057]*** [0.040]*** [0.008]** [0.045]*** [0.203]*** Mean of dep. var. 0.033 0.429 0.697 0.952 0.346 3.093 Observations 59,096 60,045 60,168 60,384 60,020 62,169 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.032 -0.153 -0.121 -0.036 -0.089 -0.101 (0.007)*** (0.029)*** (0.026)*** (0.012)*** (0.019)*** (0.020)*** [0.008]*** [0.033]*** [0.030]*** [0.013]*** [0.024]*** [0.023]*** Mean of dep. var. 0.928 0.449 0.208 0.952 0.378 0.522 Observations 208,416 202,935 202,935 60,675 58,293 202,935 Note: *** p<0.01, ** p<0.05, * p<0.1. After experiencing gang control, gang-controlled areas have worse socioeconomic conditions than neighboring areas that were not under the control of gangs. The table presents the results of estimating Specification (1) for the variables from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to gang territory (separately for each side of the boundaries). Standard errors in brackets are adjusted to allow for spatial correlation within a 100 meter radius (Conley correction). pendix presents the regression estimates for household income and the other socioeconomic characteristics from the 2019 survey. 4.3. Addressing Identification Challenges In this subsection, we analyze the assumptions that need to be satisfied for the estimates in Table Ito represent the causal effect of gang control on socioeconomic development. Conditions before the arrival of the gangs. To ensure that nongang areas close to the boundaries of gang territory are the appropriate counterfactual for gang-controlled neighborhoods, we check whether, before the arrival of the gangs, those locations had any preexisting differences in geography, socioeconomic development, or crime. First, we estimate Specification (1) for potentially important neighborhood characteristics (e.g., elevation, access to waterways, road density) and the socioeconomic characteristics from GANGS, LABOR MOBILITY, AND DEVELOPMENT 17 FIGURE 2.—Socioeconomic Conditions After 10 Years of Gang Control Note: By 2007, socioeconomic conditions had become significantly worse in gang-controlled areas. The figure illustrates the results for the 1st principal components of the dwelling, household, and individual characteristics from Table I. All the variables come from the 2007 census. The unit of observation is a dwelling, a household, and an individual, depending on the specification. All the variables are normalized to vary between zero and one with higher values representing better outcomes. The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. the 1992 census (e.g., dwelling conditions, having a TV).22 Columns 1–24 of Table II present the results. There are no discontinuities in any of the variables, confirming the notion that, initially, the locations on opposite sides of the boundaries were not different from one another. Appendix Figure A.5 illustrates these results for the first principal components of dwelling, household, and individual characteristics.23 Next, we estimate Specification (1) for the level of crime prior to the arrival of the gangs, measured by the number of people incarcerated in different parts of the city. Using the incarceration records from San Salvador’s prisons, we geocode the residential addresses of the 4,726 individuals who had been incarcerated prior to 1997. Then, we calculate the number of incarcerations per square kilometer for each 10-meter bin, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries).24 Columns 25–30 of Table II present 22Some neighborhood characteristics (e.g., elevation or access to waterways) are time-invariant. Other neighborhood characteristics may change over time. For all the variables except for road density, we use the data from either before the arrival of the gangs or soon after their arrival. For road density, the data reflect 2020 infrastructure, making the pretreatment balance test for this variable valid only under the assumption that road density is practically timeinvariant. However, given the difficulty of constructing new roads in the center of a large city, this assumption is likely to be satisfied. We describe the data in detail in the Supplementary Materials. 23Figures S.1–S.4 in the Supplementary Materials present the results for each of the neighborhood, dwelling, household, and individual characteristics from Table II. 24We perform the calculation as follows. First, we divide the map of San Salvador into zones, denoting every 10 meters from the boundaries of gang territory, separately for gang and nongang areas (all nongang locations within 10 meters of the boundaries of gang territory, all nongang locations 10–20 meters away from gang territory, and so on). Then, for each of the zones, we calculate the number of geocoded addresses within it and divide that number by the area of the zone. We employ the same procedure for other outcome variables with the same unit of analysis. 18 TABLE II GEOGRAPHIC AND SOCIOECONOMIC CHARACTERISTICS BEFORE THE ARRIVAL OF THE GANGS Neighborhood characteristics Urban territory Road density Has access to Elevation Territory used for Tree coverage the waterways coffee production (1) (2) (3) (4) (5) (6) Gang territory -0.018 -0.571 0.027 -0.193 0.008 -0.005 (0.060) (0.953) (0.069) (16.599) (0.019) (0.026) [0.052] [1.849] [0.095] [17.439] [0.023] [0.027] Mean of dep. var. 0.814 17.84 0.326 720.4 0.049 0.029 Observations 476 476 476 476 476 476 Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Shared bathroom of concrete infrastructure lighting and cooking (7) (8) (9) (10) (11) (12) Gang territory -0.015 -0.003 -0.032 -0.036 -0.007 0.021 (0.036) (0.028) (0.047) (0.039) (0.017) (0.032) [0.035] [0.030] [0.046] [0.030] [0.013] [0.029] Mean of dep. var. 0.813 0.010 0.816 0.182 0.030 0.142 Observations 64,899 64,899 64,899 64,899 64,899 64,899 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a blender Number of rooms (13) (14) (15) (16) (17) (18) Gang territory -0.004 -0.049 -0.030 0.009 0.014 -0.069 (0.009) (0.051) (0.054) (0.019) (0.032) (0.170) [0.007] [0.043] [0.049] [0.019] [0.034] [0.172] Mean of dep. var. 0.034 0.285 0.320 0.860 0.625 2.670 Observations 64,899 64,899 64,899 64,899 64,899 64,899 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (19) (20) (21) (22) (23) (24) Gang territory -0.000 -0.014 -0.019 -0.005 -0.016 -0.013 (0.011) (0.028) (0.017) (0.031) (0.030) (0.018) [0.009] [0.028] [0.017] [0.031] [0.026] [0.018] Mean of dep. var. 0.904 0.314 0.112 0.863 0.525 0.380 Observations 234,749 227,281 227,281 64,899 64,899 227,281 Number of incarcerations per km2prior to 1997: All crimes Homicide Robbery Sex crimes Assault Other violent crimes (25) (26) (27) (28) (29) (30) Gang territory -2.096 1.548 -0.640 -1.404 -0.823 -1.777 (18.200) (1.291) (3.979) (1.321) (3.400) (1.873) Mean of dep. var. 114.6 4.476 21.61 6.147 19.78 9.275 Observations 86 86 86 86 86 86 Note: *** p<0.01, ** p<0.05, * p<0.1. Before the arrival of the gangs, locations on either side of the boundaries of gang territory had similar geographic and socioeconomic characteristics. The table presents the results of estimating Specification (1) for the neighborhood characteristics and the variables from the 1992 census. The unit of observation is a census tract, dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). Standard errors in brackets are adjusted to allow for spatial correlation within a 100 meter radius (Conley correction). In columns 25–30, the Conley standard errors are not reported because there the location of the observations is not defined (the unit of observation is a 10 meter bin, denoting the distance to the boundaries of gang territory). the results of estimating Specification (1) for different types of crimes, showing that locations on both sides of the boundaries had similar levels of crime prior to the arrival of the gangs.25 25As we explain in footnote 18, we cannot report Conley standard errors in these specifications, because the unit of analysis includes areas from different parts of the city. GANGS, LABOR MOBILITY, AND DEVELOPMENT 19 Overall, before the mid-1990s, gang and nongang locations had similar levels of socioeconomic development and crime, allowing us to conclude that nongang areas close to the boundaries are the appropriate counterfactual for gang neighborhoods in the absence of gang control. Boundaries of gang territory from geographical barriers. To address any remaining concerns regarding the potential endogeneity of the boundaries, we perform the following analysis. We identify three major multilane roads—Bulevar Venezuela,49 Avenida Sur, and Autopista Comalapa—which together form more than 45 kilometers of natural barriers that largely determined the southern and western boundaries of gang territory.26 Table III reports the results of estimating Specification (1) using these three roads, rather than the actual boundaries of gang territory, to predict the location of the borders. The results remain highly significant, demonstrating that they are not driven by the potential endogeneity of some gang-territory boundaries. We also perform a placebo analysis in which we use major multilane roads that did not define the boundaries of gang territory to ensure that these geographical barriers did not affect socioeconomic development through factors unrelated to the gang boundaries. The analysis focuses on a series of consecutive roads, ranging from Redondel Masferrer in the west to Avenida Independencia in the east, that split San Salvador into two similar-size parts (see Figure 1). We then estimate whether the level of socioeconomic development changes at the placebo boundaries.27 Appendix Table A.II presents the results, confirming the notion that major roads do not affect development outcomes through factors unrelated to the gang boundaries. Stability of the boundaries of gang territory. A potential concern is that the boundaries of gang territory may not have remained stable between the time they were formed (soon after the gangs emerged) and 2015, when EDH published the map of gang territory. If the EDH map does not accurately reflect which areas were controlled by the gangs in 2007, the estimates in Table Iwould be biased toward zero (i.e., against finding an effect).28 Thus, the results in Table Ishould be interpreted as the lower bound of the effects of gang control. 26To ensure comparability of the census tracts on both sides of the regression discontinuity threshold, we exclude 25% of the largest census tracts, which are disproportionately present outside gang territory, and include dummies for the three remaining quartiles of the census tract size distribution. Table S.II in the Supplementary Materials reports the results of estimating the same regression specification without excluding the largest census tracts and, instead, including dummies for all four quartiles of the census tract size distribution. 27Specifically, we estimate the regression specification defined below, where north is a dummy variable for a census tract being to the north of the placebo boundaries. The coefficient of interest is ω4, which estimates the change in socioeconomic conditions at the placebo boundaries. Similarly to the other regression specifications, we limit the sample to observations within 420 meters of the (placebo) discontinuity threshold. In addition, similarly to Table III, we exclude 25% of the largest census tracts, which are predominantly present outside gang territory, and include dummies for the three remaining quartiles of the census tract size distribution. Table S.III in the Supplementary Materials reports the results of estimating the same regression specification without excluding the largest census tracts, and instead, including dummies for all four quartiles of the census tract size distribution. yic =ω0+ω1distancec+ω2distancec→northc+ω3gang territoryc+ω4northc+εic.(2) 28For instance, if, in reality, the gangs controlled more neighborhoods than suggested by the map, then, under the assumption that the gangs have a homogeneous effect on socioeconomic development in all the areas they control, living conditions in the control group would be underestimated. In turn, the difference in living conditions between the gang and nongang areas would also be underestimated. Similarly, if the gangs actually controlled fewer neighbor- 20 TABLE III BOUNDARIES OF GANG TERRITORY FROM GEOGRAPHICAL BARRIERS Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.096 0.047 -0.064 -0.226 0.004 -0.287 (0.014)*** (0.009)*** (0.014)*** (0.056)*** (0.002)** (0.029)*** [0.021]*** [0.012]*** [0.022]*** [0.102]** [0.002]** [0.101]*** Mean of dep. var. 0.947 0.021 0.966 0.050 0.002 0.097 Observations 7,424 6,312 6,348 6,348 6,348 6,056 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.021 -0.606 -0.385 -0.045 -0.556 -2.061 (0.009)** (0.052)*** (0.032)*** (0.014)*** (0.062)*** (0.321)*** [0.013]* [0.155]*** [0.053]*** [0.014]*** [0.118]*** [0.398]*** Mean of dep. var. 0.033 0.305 0.671 0.957 0.256 2.814 Observations 6,021 6,080 6,098 6,119 6,086 6,348 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.167 -0.406 -0.233 -0.071 -0.246 -0.266 (0.039)*** (0.032)*** (0.054)*** (0.009)*** (0.025)*** (0.028)*** [0.038]*** [0.049]*** [0.107]** [0.014]*** [0.059]*** [0.051]*** Mean of dep. var. 0.926 0.406 0.146 0.964 0.335 0.486 Observations 21,488 20,722 20,722 6,312 5,933 20,722 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1), using the locations of major roads and boulevards (geographical barriers) as the predicted boundaries of gang territory. To ensure comparability of the census tracts on both sides of the regression discontinuity threshold, we exclude 25% of the largest census tracts, which are disproportionately present outside gang territory. We also include dummies for the three remaining quartiles of the census tract size distribution. Table S.II in the Supplementary Materials reports the results of estimating the same regression specification without excluding the largest census tracts and, instead, including dummies for all four quartiles of the census tract size distribution. All the variables come from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). Standard errors in brackets are adjusted to allow for spatial correlation within a 100 meter radius (Conley correction). Nevertheless, in Appendix Subsection A.1, we demonstrate that the gang-territory boundaries have remained stable since they were first formed. Specifically, we exploit the fact that most gang-related homicides take place precisely at the boundaries of gang territory because of people attempting to enter or leave gang-controlled neighborhoods without permission.29 As a result, by showing that, throughout the years, gang-related homicides consistently take place right at the boundaries from the EDH map, we are able to confirm the validity of that map and demonstrate the stability of those boundaries. In addition, in 2023, we conducted a new survey of individuals from gang and nongang neighborhoods, in which, among other questions, the respondents were asked whether their hoods than suggested by the map, then living conditions in the treatment group would be overestimated, which would also lead to a smaller difference between the treatment and control groups. 29This phenomenon has also been documented for the 1970s through the 1990s in gang neighborhoods in Los Angeles, where most of the violence took place right at the entrance to these neighborhoods (Artsy,2018). GANGS, LABOR MOBILITY, AND DEVELOPMENT 21 neighborhood had been controlled by gangs 20 years ago, during the presidency of Francisco Flores Pérez (President of El Salvador in 1999–2004). As shown in Appendix Figure A.6, the share of respondents answering in the affirmative significantly increases at the boundaries of gang territory, suggesting that the borders have remained stable over time. Selective migration: in-sample migration. Another assumption that needs to be satisfied for our estimates to be interpreted as causal is that there has been no selective migration of individuals across the regression discontinuity threshold. Selective migration can affect our results in two ways. The first is what we call in-sample migration: individuals moving from a neighborhood on one side of the boundaries to an area on the other side of the boundaries while remaining in San Salvador and, thus, in our sample. This type of migration would be a direct threat to identification because it would imply that individuals can manipulate their treatment status. The second is what we call out-of-sample migration: individuals moving from San Salvador to a different municipality in El Salvador or abroad. This type of migration does not invalidate the identification strategy, but it changes the interpretation of the mechanism through which the gangs affect local socioeconomic conditions (i.e., that gang control makes wealthy, educated individuals leave San Salvador). In this subsection, we consider the direct threat to identification that comes from in-sample migration. To show that in-sample migration is not driving our findings, we leverage our 2019 survey, where, among other questions, we asked individuals whether they had lived in the exact same place their entire life: 77% of respondents said they had. This information allows us to compare the results for the full sample and for the subsample of respondents for whom we know the ex ante treatment status (i.e., that they lived in the location before the arrival of the gangs). In the absence of in-sample migration, the two sets of results would be quite similar, whereas, if the results are determined by in-sample migration, the discontinuities would appear only in the full sample. Notably, this exercise also allows us to determine that the results are not driven by wealthy and educated newcomers choosing to settle in nongang parts of San Salvador. By restricting the sample to individuals who have lived in the same neighborhood their entire life, by definition, we exclude all newcomers. When we limit the sample this way, the results of the regression discontinuity analysis are practically unchanged. Appendix Figure A.4 illustrates this fact by showing the two regression discontinuity plots for household income. The left-hand side of the figure presents the results for the full sample; the right-hand side presents the subsample of never-movers. The two plots are quite similar, suggesting that the results are not driven by selective in-sample migration. Table A.I in the Appendix reports the regression estimates for the socioeconomic characteristics from our 2019 survey, both for the full sample and for the sample of never-movers; Figure S.5 in the Supplementary Materials illustrates these results.30 30In the 2007 census, individuals were also asked whether they had lived in the same municipality their entire life. Since individuals who answered in the affirmative could still have moved within the municipality, this question is 22 For a detailed discussion of out-of-sample migration (i.e., individuals moving from San Salvador to a different municipality or abroad), see Appendix Subsection A.2. Absence of pretrends. In Section 6, we also demonstrate the absence of pretrends in socioeconomic development between areas with and without gang presence. Specifically, we perform a difference-in-differences analysis using nighttime light density, household income, and firm openings to show that these variables only started to change after the deportation of the gang leaders from the United States to El Salvador. Other robustness checks. In Appendix Subsection A.3, we also present a wide range of additional robustness checks to ensure that the estimates in Table Irepresent the causal effect of gang control on socioeconomic development. 5. MECHANISMS In this section, we explore the mechanisms behind the negative effects of gangs’ territorial control on development outcomes. In particular, we provide novel evidence on how gangimposed mobility restrictions affect individuals’ labor-market choices by preventing them from commuting to areas outside of gang territory, where the largest and best-paying firms are located. We also show that the differences in educational attainment between gang and nongang areas can be explained by higher dropout rates in gang-controlled neighborhoods. Finally, we investigate alternative mechanisms and find that the regression discontinuity results cannot be explained by differences in crime (i.e., homicides, extortion) or the composition of firms at the boundaries of gang territory. In Appendix Subsection A.2, we show that our results are not driven by selective migration of individuals out of gang territory. Specifically, we estimate the rates of selective out-of-sample migration by considering the relationship between household wealth and the probability of a family member migrating abroad from 1997 through 2007, finding that selective migration accounts for no more than 14% of the gaps in socioeconomic development between gang and nongang areas.31 In Appendix Subsections A.4 and A.5, we also demonstrate that the regression less precise at determining the ex ante treatment status of the respondents. Coincidentally, however, the 2007 survey found that the share of population that had always lived in San Salvador municipality was 77%, the same percentage as the share of population that had always lived in the same location according to the 2019 survey. Thus, it appears that, in this context, individuals primarily move across municipalities and not within the same municipality. Under this assumption, we estimate Specification (1) for the variables from the 2007 census for the subsample of individuals who had always lived in the same municipality. Table S.IV in the Supplementary Materials presents the results, which are very similar to those presented in Table I, confirming that in-sample migration is not likely to be driving the results. In addition, for the two main outcome variables that are potentially affected by in-sample migration (i.e., the first principal components of household and individual characteristics), we perform a test in the spirit of Lee (2009). Focusing on individuals living within 100 meters of the gang boundaries, we calculate the bounds of the treatment effects under the very strong assumption that all individuals who have ever changed their municipality of residence did so in a way that biases our estimates. Even under this strong assumption, we find that the effects of gang control are bounded between -0.127 and -0.07 for household characteristics and -0.124 and -0.048 for individual characteristics. In the 2019 survey, a similar exercise bounds the effects on household income between -497.7 and -299.1. All the bounds are statistically different from zero. 31In Panel B of Appendix Table A.XIV, we also show that our results are not determined by positive selective internal migration within El Salvador. Specifically, we find that individuals who have previously lived in San Salvador GANGS, LABOR MOBILITY, AND DEVELOPMENT 23 discontinuity results cannot be explained by differences in public goods provision or occupational structure (i.e., unemployment, informal employment, or hours worked), respectively.32 5.1. Restrictions on Mobility The presence of mobility restrictions. To document the presence of restrictions on individuals’ mobility, we estimate Specification (1) for mobility questions from three different sources: the 2019 survey (columns 1–5), a follow-up survey that we conducted in 2023 (column 6), and cell phone ping data from early 2022 (columns 7–8). Table IV presents the results. The estimates in column 1 suggest that the share of population working in gang-controlled neighborhoods dramatically increases by almost 50 percentage points (from 5.7% to 55.2%) at the boundaries of gang territory. Residents of gang territory are also more likely to work in the same neighborhood where they live and are less likely to have traveled outside of San Salvador: the share of individuals who have ever been to the beach or visited Santa Ana department, which are both 30 to 60 kilometers away, discontinuously decreases at the boundaries of gang territory. To further demonstrate the salience of restrictions on individuals’ mobility, we show that residents of gang areas acknowledge the presence of these restrictions. First, in 2019, they were significantly less likely to say that there is freedom of movement in the neighborhood where they live (column 5 of Table IV). Second, in 2023, we conducted a new survey in San Salvador, in which the respondents were asked whether, in the past, the gangs’ “invisible borders” prevented them from finding jobs in large firms in other parts of the city. While individuals outside of gang territory were also affected (e.g., due to gang areas blocking the routes between some nongang parts of the city), the impact was significantly stronger for the residents of gang neighborhoods (column 6 of Table IV). Finally, in columns 7 and 8 of Table IV, we use cell phone ping data from early 2022 to illustrate that residents of gang territory are not generally less mobile than individuals living in other parts of the city, but that their movements are confined to gang-controlled areas. We begin with dividing the map of San Salvador into 100→100-meter grid cells and using the prevalence of pings during the night hours (from 9 p.m. to 7 a.m.) to identify the grid cell where an individual lives. Then, for each individual, we calculate the share of pings inside gang territory during the daytime (from 9 a.m. to 7 p.m.), excluding pings in their home grid cells.33 Similarly, we calculate the average distance that an individual travels away from their home during the daytime. Columns 7 and 8 of Table IV present the results of estimating Specification (1) for but now reside in a different municipality have similar socioeconomic characteristics to individuals living in gangcontrolled neighborhoods of San Salvador (see columns 5 and 7 of Appendix Table A.XIV). 32Notably, there is no discontinuity in the probability of being employed. The results of estimating Specification (1) suggest that residents of gang territory aged between 18 and 65 are only 0.2 percentage points less likely to be employed than individuals from nongang areas (with the standard error of 1.1 percentage points). 33Home grid cells are intentionally excluded to ensure that the results are not driven by individuals spending time at home. If home grid cells are not excluded, the estimate in column 7 of Table IV becomes three times larger. 24 these outcome variables. They confirm that residents of gang-controlled neighborhoods spend a substantially larger share of their time in gang territory than individuals on the other side of the regression discontinuity cutoff. However, within the areas to which they are confined, both groups of individuals travel the same distance throughout the day, suggesting that residents of gang neighborhoods do not have a lower capacity to travel away from home. Figure 3presents the regression discontinuity plots for the four most important variables in Table IV: the share of people working in gang territory, the share of time individuals spend in gang territory, the share of people who think there is freedom of movement in the area where they live, and the share of people who say that gang-imposed restrictions on mobility prevented them from finding jobs in large firms in other parts of the city. TABLE IV GANG CONTROL AND RESTRICTIONS ON INDIVIDUALS’MOBILITY Works in Works in neighborhood Has been to Has been to gang territory where they live Santa Ana the beach (1) (2) (3) (4) Gang territory 0.495 0.111 -0.277 -0.064 (0.039)*** (0.031)*** (0.043)*** (0.031)** [0.042]*** [0.050]** [0.052]*** [0.032]** Mean of dep. var. 0.334 0.302 0.495 0.872 Observations 1,738 2,071 2,314 2,314 Freedom of Gang borders prevented you Share of time spent in Mean distance away movement from getting jobs in large firms gang areas, excluding from home during where they live in other parts of the city time spent at home the day (in meters) (5) (6) (7) (8) Gang territory -0.097 0.100 0.213 -52.82 (0.039)** (0.041)** (0.039)*** (154.82) [0.039]** [0.046]** [0.034]*** [147.55] Mean of dep. var. 0.811 0.407 0.222 1955.62 Observations 2,314 2,313 9,605 9,605 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the questions related to mobility. The outcome variables in columns 1–5 come from the 2019 survey, the outcome variable in column 6—from the 2023 survey, and the outcome variables in columns 7–8 are based on cell phone ping data from SDK. Santa Ana is a neighboring municipality, which is approximately 60 kilometers away from San Salvador. The beach is approximately 30 kilometers away from San Salvador. The unit of observation is an individual. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). Standard errors in brackets are adjusted for spatial correlation within a 100 meter radius (Conley correction). We additionally investigate whether the results in Table IV can be determined by residents of gang areas having less access to personal transportation, such as cars and motorcycles. To address this question, we consider the heterogeneity of the effects of gang control on people who get to work by car or motorcycle and people who get to work in another way, controlling GANGS, LABOR MOBILITY, AND DEVELOPMENT 31 they should be interpreted to denote the effects of gangs’ territorial control and accompanying restrictions on mobility. In Section 6, we discuss this latter implication in much more detail and quantify the effects of exposure to overall gang presence, not just gang territorial control. 6. AGGREGATE EFFECTS OF GANG PRESENCE As we previewed at the end of Subsection 5.3, the regression discontinuity results largely represent the socioeconomic costs of full territorial control by the gangs and not necessarily other forms of gang activity. In this section, we use data for all of El Salvador to analyze the broader consequences of gang presence on economic activity in the country. 6.1. Theoretical Framework To clarify the distinction between gang presence and gang territorial control, we begin with presenting a simple conceptual framework for the interpretation of our results. Appendix Subsection A.6 provides a more detailed discussion of the model behind this conceptual framework, as well as an analysis of two counterfactual scenarios: the removal of restrictions on individuals’ mobility and the full removal of gang presence. We consider a one-dimensional city on a unit interval, where locations are characterized by their proximity to the gangs. Overall, the city is divided into three qualitatively different areas. Places in [0,b]are fully controlled by the gangs, and individuals living there cannot work in other parts of the city. These locations are the equivalent of gang territory in the regression discontinuity design. Places in [b, b +ϑ](ϑ>0) are not controlled by the gangs. Individuals living there are free to work in any nongang part of the city, but firms in [b, b +ϑ]are still exposed to extortion and other gang-related activities. Together, places in [0,b+ϑ]comprise what we refer to as areas with gang presence. Finally, places in [b+ϑ,1] do not have any gang presence. The differences in labor-market conditions between these three areas are determined by the production technology used by firms in that location. All firms can choose between two options: a simple technology that does not require any investment from the firms and a more productive technology that requires an initial investment at a fixed cost. In the absence of gangs, firms benefit from the adoption of the productive technology. However, productive firms in areas with gang presence face a high risk of their output being extorted, which makes them choose the simple technology instead. Thus, only firms in areas without gang presence choose to increase their productivity. In turn, as we show in Appendix Subsection A.6, under a realistic set of parameters, this results in increased employment and higher wages in those firms. Despite labor-market conditions only being better in areas without gang presence, individuals living in [b, b +ϑ]are still able to take advantage of them because of their ability to commute to [b+ϑ,1], whereas people living in [0,b]cannot do so. This part of the mechanism highlights the importance of restrictions on mobility for people living in gang territory (i.e., [0,b]). At the 32 same time, restrictions on mobility only matter due to higher economic growth in areas without gang presence: if firms in all parts of the city were the same, there would have been no need to commute to [b+ϑ,1]. Based on this conclusion, we now analyze whether, after the arrival of the gangs, locations without gang presence indeed experienced more economic growth than places exposed to gang activity. 6.2. Difference-in-Differences: Empirical Strategy To analyze the aggregate impact of gang activity, we use data from all of El Salvador to perform a difference-in-differences analysis, comparing the evolution of economic activity in areas with varying levels of gang activity after 1996. Our analysis exploits two sources of variation: the timing of gang members’ deportation from the United States, which led to the emergence of gangs in El Salvador, and the geographic differences in exposure to organized crime. Our hypothesis is that prior to 1996, the year of the first wave of deportations from the United States, locations that would later experience different levels of gang activity had similar rates of economic development. However, after 1996, we expect to see higher rates of economic growth in areas with low levels of gang presence. We exploit the fact that, after being deported, many gang members who were born in El Salvador returned to their municipality of birth (Sviatschi,2022b). Thus, we use the municipalities of birth of known gang leaders as a treatment variable for whether the municipality became exposed to gang activity.43 We then estimate the following event study model (Specification 3) to measure the effect of gang presence on economic growth. Econ. growthi,t =gi+ϖt+! j→=1995 ϱjgang presencei→1{Year=j}t+εi,t.(3) Econ. growth represents various measures of economic growth in municipality iat time t. gang presence is a dummy for whether a gang leader was born in municipality i;giand ϖt represent municipality and year fixed effects, respectively. Standard errors are estimated using Conley standard errors with spatial correlation within a 5 km radius. The coefficients of interest are ϱj, which represent the differences in economic growth between locations with and without gang presence relative to 1995—the year before the change in the United States immigration policy. We use three outcome variables to measure municipality-level growth in economic activity. The first one is the opening of new business establishments. Specifically, we use data from 43We verify the relevance of this treatment variable by showing that gang leaders’ birth municipalities were 80 percentage points more likely to experience a gang-related homicide in 2003–2004, the first years for which geocoded homicide data are available. GANGS, LABOR MOBILITY, AND DEVELOPMENT 33 the 2005 economic census, which includes information on when the firms were opened.44 The second outcome variable is nighttime light density (or luminosity) which recent studies have found to be a good proxy for local-level economic activity (Chen and Nordhaus,2011,Henderson et al.,2012). Finally, we use data on household income, which is based on annual household surveys conducted in 1992–2007 by DIGESTYC. In all three cases, the outcomes are measured in percentage points, normalized to be 100 percent in 1995–1996, both in areas with and without gang presence. In addition, given that the gangs were primarily attracted to large cities, to avoid the comparison between urban and rural locations, we limit our analysis to urban municipalities. 6.3. Difference-in-Differences: Results Figure 4presents the results of estimating Specification (3) for the three outcome variables.45 It shows that, before the 1996 change in United States immigration policy, areas with and without future gang presence experienced similar growth in economic activity. However, after the arrival of the gangs in 1996–1997, municipalities with gang presence experienced significantly lower economic growth. FIGURE 4.—Gang Presence and Economic Activity Note: The figure presents an event study graph for the differences in economic growth between municipalities with and without gang presence. For all three outcome variables, the data are in percentage points, normalized to be equal to 100 percent in 1995–1996, before the change in the United States immigration policy. The magnitude of the effects is substantial. For example, by 2005, municipalities without gang presence had experienced a 105 percentage point higher rate of new business openings. Additionally, after 1997, on average, these areas had an 82 percentage point higher growth in nighttime light density and a 28.5 percentage point higher growth in household income.46 44Given that the economic census was conducted in 2005, to be included in our sample, the firm must not have closed until that time. Thus, our results should be interpreted as illustrating an effect on the opening of moderately successful business establishments. 45The corresponding regression coefficients are reported in Table A.XII in the Appendix. 46According to Henderson et al. (2012), a one percentage point change in luminosity corresponds to a 0.28 percentage point change in GDP. Thus, between 1998 and 2013, on average, areas without gang presence had approximately 34 Overall, these results confirm the notion that, after the arrival of the gangs, most economic growth has taken place in areas far away from gang territory, plausibly due to business owners’ desire to avoid extortion and other forms of gang activity. We also complement the difference-in-differences results by using household income data from our 2019 survey in San Salvador and performing a back-of-the-envelope calculation that compares locations without gang presence separately to fully gang-controlled neighborhoods and places with only some gang activities.47 We find that, after 1997, areas with no gang presence experienced approximately 50 percentage points higher growth in household income than the former and approximately 9 percentage points higher growth than the latter. Thus, while proximity to places with the highest growth of employment opportunities positively affected individuals’ earnings, it was residents of gang-controlled neighborhoods who were particularly negatively affected due to their inability to commute across the boundaries of gang territory. We also analyze whether the effects of gang presence are different in the largest urban centers (e.g., San Salvador) and in the rest of the country. To address this question, we follow Arkhangelsky et al. (2021) and implement two types of synthetic difference-in-differences analyses. The first one defines the treatment variable in the same way as in the baseline differencein-differences estimation. The second one narrows the treatment group to the four largest cities in El Salvador, all of which have had a substantial gang presence since the late 1990s: San Salvador, Soyapango, Santa Ana, and San Miguel. Appendix Figure A.9 presents the two sets of results. In general, we find the two specifications to be quite similar, suggesting that the largest cities were not differentially affected compared to other places with gang presence. 7. CONCLUDING REMARKS Overall, the results presented in this paper indicate that, via a combination of restrictions on individuals’ mobility and displacement of economic activity, nonstate armed actors can have a considerable negative impact on socioeconomic development. These findings have broad policy implications, shedding light on the long-term consequences of deporting individuals with criminal records to a country with low state capacity, suggesting that improvements in state capacity can significantly improve economic growth, and highlighting the importance of freedom of movement for socioeconomic development. 82→0.28 = 23 percentage points higher GDP growth compared to areas with gang activity. This estimate closely aligns with the one for household income. 47According to the 2007 census, 60% of San Salvador’s population lived in gang-controlled neighborhoods. Thus, the average household income in San Salvador is equal to A=0.4GC +0.6GP , where GC and GP represent household income from the 2019 survey in fully gang-controlled parts of the city and places with only some gang presence, respectively. Then, without gang presence, San Salvador would have been expected to have an average household income of 1.285A, which, in turn, allows for a comparison with GC and GP . GANGS, LABOR MOBILITY, AND DEVELOPMENT 35 REFERENCES ABREGO,LEISY,MAT COLEMAN,DANIEL MARTÍNEZ,CECILIA MENJÍVAR,AND JEREMY SLACK (2017): “Making immigrants into criminals: Legal processes of criminalization in the post-IIRIRA era,” Journal on Migration and Human Security, 5, 694–715. [7] ACEMOGLU,DARON,GIUSEPPE DEFEO,AND GIACOMO DAVIDE DELUCA (2019): “Weak States: Causes and Consequences of the Sicilian Mafia,” Review of Economic Studies, 87, 537–581. [5] ACEMOGLU,DARON,SIMON JOHNSON,AND JAMES A. ROBINSON (2001): “The Colonial Origins of Comparative Development: An Empirical Investigation,” American Economic Review, 91, 1369–1401. [2,6] ——— (2002): “Reversal of Fortune: Geography and Institutions in the Making of the Modern World Income Distribution,” Quarterly Journal of Economics, 117, 1231–1294. [6] ACEMOGLU,DARON,JAMES A. ROBINSON,AND RAFAEL J. SANTOS (2013): “The Monopoly of Violence: Evidence from Colombia,” Journal of the European Economic Association, 11, 5–44. [5] ALESINA,ALBERTO,SALVATORE PICCOLO,AND PAOLO PINOTTI (2019): “Organized crime, violence, and politics,” Review of Economic Studies, 86, 457–499. [5] ALSAWADY,YOSEPH A., M. KABIR HASSAN,AND ARJA H. TURUNEN-RED (2022): “Limiting The Mobility of Migrant Workers in The Saudi Labor Market,” International Journal of Economics and Management, 16, 45–64. [6] ARKHANGELSKY,DMITRY,SUSAN ATHEY,DAVID A. HIRSHBERG,GUIDO W. IMBENS,AND STEFAN WAGER (2021): “Synthetic Difference-in-Differences,” American Economic Review, 111, 4088–4118. [34] ARTSY,AVISHAY (2018): “Gang borders create invisible walls in Los Angeles,” Tech. rep., KCWR, https://www.kcrw.com/culture/shows/design-and-architecture/gang-borders-create-invisible-walls-in-los-angeles. [7,20] ASHER,SAM AND PAUL NOVOSAD (2020): “Rural Roads and Local Economic Development,” American Economic Review, 110, 797–823. [6] BANDIERA,ORIANA (2003): “Land Reform, the Market for Protection, and the Origins of the Sicilian Mafia: Theory and Evidence,” Journal of Law, Economics, and Organization, 19, 218–244. [5] BATES,ROBERT,AVNER GREIF,AND SMITA SINGH (2002): “Organizing violence,” Journal of Conflict Resolution, 46, 599–628. [2] BLATTMAN,CHRISTOPHER,GUSTAVO DUNCAN,BENJAMIN LESSING,AND SANTIAGO TOBON (2022): “Gang rule: Understanding and Countering Criminal Governance,” NBER Working Paper 28458.[2,4,5] BLATTMAN,CHRISTOPHER AND EDWARD MIGUEL (2010): “Civil war,” Journal of Economic literature, 48, 3–57. [5] BOERMAN,THOMAS AND ADAM GOLOB (2020): “Gangs and modern-day slavery in El Salvador, Honduras and Guatemala: A non-traditional model of human trafficking,” Journal of Human Trafficking, 1–17. [9,10] BRETTELL,CAROLINE B (2011): “Experiencing everyday discrimination: A comparison across five immigrant populations,” Race and Social Problems, 3, 266–279. [6] BROWN,ZACH Y., EDUARDO MONTERO,CARLOS SCHMIDT-PADILLA,AND MARÍA MICAELA SVIATSCHI (2020): “Market Structure and Extortion:Evidence from 50,000 Extortion Payments,” NBER Working Paper.[73] BUONANNO,PAOLO,RUBEN DURANTE,GIOVANNI PRAROLO,AND PAOLO VANIN (2015): “Poor Institutions, Rich Mines: Resource Curse in the Origins of the Sicilian Mafia,” Economic Journal, 125, F175–F202. [5] BUONANNO,PAOLO,GIOVANNI PRAROLO,AND PAOLO VANIN (2016): “Organized crime and electoral outcomes. Evidence from Sicily at the turn of the XXI century,” European Journal of Political Economy, 41, 61–74. [5] CALÌ,MASSIMILIANO AND SAMI H. MIAARI (2018): “The Labor Market Impact of Mobility Restrictions: Evidence from the West Bank,” Labour Economics, 51, 136–151. [6] CARVALHO,LEANDRO S. AND RODRIGO R. SOARES (2016): “Living on the edge: Youth entry, career and exit in drug-selling gangs,” Journal of Economic Behavior & Organization, 121, 77–98. [5] CHACÓN,JENNIFER M. (2009): “Managing migration through crime,” Columbia Law Review, 109, 135–148. [7] CHEN,XIANDWILLIAM D. NORDHAUS (2011): “Using luminosity data as a proxy for economic statistics,” PNAS, 108, 8589–8594. [33] CLEMONS,MICHAEL,CLAUDIO MONTENEGRO,AND LANT PRITCHETT (2008): “The great discrimination: borders as a labor market barrier,” Mimeo.[6] CÓRDOVA,ABBY (2019): “Living in Gang-Controlled Neighborhoods: Impacts on Electoral and Nonelectoral Participation in El Salvador,” Latin American Research Review, 54, 201–221. [4,11,44] COUGHLIN,BRENDA CAND SUDHIR ALLADI VENKATESH (2003): “The urban street gang after 1970,” Annual Review of Sociology, 41–64. [7] CRUZ,JOSÉ MIGUEL (2010): “Central American maras: From youth street gangs to transnational protection rackets,” Global Crime, 11, 379–398. [7,9] CRUZ,JOSÉ MIGUEL AND NELSON PORTILLO PEÑA (1998): Solidaridad y violencia en las pandillas del gran San Salvador, UCA Editores. [8] 36 DANIELE,GIANMARCO AND GEMMA DIPOPPA (2017): “Mafia, elections and violence against politicians,” Journal of Public Economics, 154, 10–33. [5] DANIELE,GIANMARCO AND BENNY GEYS (2015): “Organised Crime, Institutions and Political Quality: Empirical Evidence from Italian Municipalities,” The Economic Journal, 125, F233–F255. [5] DANIELE,VITTORIO AND UGO MARANI (2011): “Organized crime, the quality of local institutions and FDI in Italy: A panel data analysis,” European Journal of Political Economy, 27, 132–142. [5] DEFEO,GIUSEPPE AND GIACOMO DAVIDE DELUCA (2017): “Mafia in the Ballot Box,” American Economic Journal: Economic Policy, 9, 134–67. [5] DELASIERRA,RAÚL (2020): “On the Origins of the State: Stationary Bandits and Taxation in Eastern Congo,” Journal of Political Economy, 128, 32–74. [2,5] DECESARE,DONNA (1998): “The children of war street gangs in el salvador,” NACLA Report on the Americas, 32, 21–29. [8] DELL,MELISSA (2010): “The Persistent Effects of Peru’s Mining Mita,” Econometrica, 78, 1863–1903. [6] ——— (2015): “Trafficking Networks and the Mexican Drug War,” American Economic Review, 105, 1738–1779. [5] DELL,MELISSA,NATHAN LANE,AND PABLO QUERUBIN (2018): “The Historical State, Local Collective Action, and Economic Development in Vietnam,” Econometrica, 86, 2083–2121. [6] DELL,MELISSA AND BENJAMIN OLKEN (2020): “The Development Effects of the Extractive Colonial Economy: The Dutch Cultivation System in Java,” Review of Economic Studies, 87, 164–203. [6] DONALDSON,DAVE (2018): “Railroads of the Raj: Estimating the Impact of Transportation Infrastructure,” American Economic Review, 108, 899–934. [6] DUNN,WILLIAM CARL (2007): The Gangs of Los Angeles, iUniverse. [7] FRYE,TIMOTHY AND EKATERINA ZHURAVSKAYA (2000): “Rackets, Regulation, and the Rule of Law,” Journal of Law, Economics, & Organization, 16, 478–502. [5] GAMBETTA,DIEGO (1996): The Sicilian Mafia: The Business of Private Protection, Cambridge MA: Harvard University Press. [5] GANONG,PETER AND SIMON JÄGER (2018): “A Permutation Test for the Regression Kink Design,” Journal of the American Statistical Association, 113, 494–504. [14] GLAESER,EDWARD AND HELEN SIMS (2015): “Contagion, crime, and congestion: overcoming the downsides of density,” IGC Growth Brief.[5] HENDERSON,J.VERNON,ADAM STOREYGARD,AND DAVID N. WEIL (2012): “Measuring Economic Growth from Outer Space,” American Economic Review, 102, 994–1028. [33] IBÁÑEZ,ANA MARÍA,JULIAN ARTEAGA,JUAN CAMILO CÁRDENAS CAMPO,ANA ARJONA,AND PATRICIA JUSTINO (2019): “The Effects of Wartime Institutions on Households’ Ability to Cope with Shocks: Evidence for Colombia,” WIDER Working Paper №2019/84.[2,5,6] INSIGHT CRIME AND CLALS (2018): “MS13 in the Americas. How the World’s Most Notorious Gang Defies Logic, Resists Destruction,” Tech. rep., Insight Crime and The Center for Latin American & Latino Studies (CLALS). [10] INTERNATIONAL CRISIS GROUP (2017): “El Salvador’s Politics of Perpetual Violence,” Latin America Report №64. [9,11] ——— (2018): “Life Under Gang Rule in El Salvador,” Latin America & Caribbean Commentary.[3,9,10,11] JONES,BENJAMIN FAND BENJAMIN AOLKEN (2005): “Do leaders matter? National leadership and growth since World War II,” The Quarterly Journal of Economics, 120, 835–864. [6] KULISH,NICHOLAS (2018): “What It Costs to Be Smuggled Across the U.S. Border,” The New York Times.[40] LEE,DAVID S. (2009): “Training, Wages, and Sample Selection: Estimating Sharp Bounds on Treatment Effects,” The Review of Economic Studies, 76, 1071–1102. [22] LESSING,BENJAMIN (2021): “Conceptualizing Criminal Governance,” Perspectives on Politics, 19, 854–873. [2,5] LESSING,BENJAMIN AND GRAHAM DENYER WILLIS (2019): “Legitimacy in Criminal Governance: Managing a Drug Empire from Behind Bars,” American Political Science Review, 113, 584–606. [5] LEVITT,STEVEN AND SUDHIR VENKATESH (2000): “An Economic Analysis of a Drug-Selling Gang’s Finances,” Quarterly Journal of Economics, 115, 755–789. [5] LIBREROS,VICTORIA,JULIÁN ANTONIO,AND ROBERTO CARBAJAL (2010): El PAEBA de El Salvador. La experiencia y el aporte de la Universidad Francisco Gavidia, San Salvador: UFG - Editores. [28] LOWES,SARA AND EDUARDO MONTERO (2021): “Concessions, violence, and indirect rule: evidence from the Congo Free State,” The Quarterly Journal of Economics, 136, 2047–2091. [6] MAGALONI,BEATRIZ,EDGAR FRANCO VIVANCO,AND VANESSA MELO (2020a): “Killing in the Slums: Social Order, Criminal Governance, and Police Violence in Rio de Janeiro,” American Political Science Review, 114, 552–572. [5,6] MAGALONI,BEATRIZ,GUSTAVO ROBLES,AILA MMATANOCK,ALBERTO DIAZ-CAYEROS,AND VIDAL ROMERO (2020b): “Living in Fear: the Dynamics of extortion in Mexico’s Drug War,” Comparative Political Studies, 53, 1124–1174. [4] GANGS, LABOR MOBILITY, AND DEVELOPMENT 37 MARTÍNEZ,JUAN JOSÉ AND ÓSCAR MARTÍNEZ (2018): El niño de Hollywood, Penguin Random House Grupo Editorial. [7,8] MARTÍNEZ,OSCAR (2016): “Living Within the Boundaries of El Salvador’s Gang ‘War’,” Tech. rep., Insight Crime. [10] MARTÍNEZ,OSCAR,EFREN LEMUS,CARLOS MARTÍNEZ,AND DEBORAH SONTAG (2016): “Killers on a Shoestring: Inside the Gangs of El Salvador,” The New York Times.[4,11] MCCRARY,JUSTIN (2008): “Manipulation of the running variable in the regression discontinuity design: A density tes,” Journal of Econometrics, 142, 698–714. [40] MCKENZIE,DAVID,STEVEN STILLMAN,AND JOHN GIBSON (2010): “How Important is Selection? Experimental vs. Non-Experimental Measures of the Income Gains from Migration,” Journal of the European Economic Association, 8, 913–945. [6] MELNIKOV,NIKITA,CARLOS SCHMIDT-PADILLA,AND MARIA MICAELA SVIATSCHI (2024): “Gang Crackdowns and Development,” Tech. rep., Princeton University. [25,48] MERGO,TEFERI (2016): “The Effects of International Migration on Migrant-Source Households: Evidence from Ethiopian Diversity-Visa Lottery Migrants,” World Development, 84, 69–81. [6] MICHALOPOULOS,STELIOS AND ELIAS PAPAIOANNOU (2013): “Pre-Colonial Ethnic Institutions and Contemporary African Development,” Econometrica, 81, 113–152. [2,6] MIGUEL CRUZ,JOSÉ (2010): “Central American maras: from youth street gangs to transnational protection rackets,” Global Crime, 11, 379–398. [8] MIRENDA,LITTERIO,SAURO MOCETTI,AND LUCIA RIZZICA (2022): “The Economic Effects of Mafia: Firm Level Evidence,” American Economic Review, forthcoming. [5] MURPHY,TOMMY E. AND MARTÍN A. ROSSI (2020): “Following the poppy trail: Origins and consequences of Mexican drug cartels,” Journal of Development Economics, 143. [5] NUÑO,LIDIA E. AND EDWARD R. MAGUIRE (2021): “The Nature and Structure of MS-13 in Los Angeles County,” Criminal Justice Review, 07340168211029990. [2,7] OLSON,MANCUR (1993): “Dictatorship, Democracy, and Development,” American Political Science Review, 87, 567–576. [2,5] PALMA,EDGARD (1999): “Pandillas y Delincuencia en El Salvador,” Asociacíon Salvadoreña de Socioloía.[10] PINOTTI,PAOLO (2015): “The Causes and Consequences of Organised Crime: Preliminary Evidence Across Countries,” Economic Journal, 125, F158–F174. [5] SAVENIJE,WIM (2009): Maras y barras: pandillas y violencia juvenil en los barrios marginales de Centroamérica, FLACSO—Programa El Salvador. [6] SKARBEK,DAVID (2011): “Governance and Prison Gangs,” American Political Science Review, 105, 702–716. [5] STANLEY,WILLIAM DEANE (1987): “Economic migrants or refugees from violence? A time-series analysis of Salvadoran migration to the United States,” Latin American Research Review, 22, 132–154. [6] SVIATSCHI,MARÍA MICAELA (2022a): “Making a Narco: Childhood Exposure to Illegal Labor Markets and Criminal Life Paths,” Econometrica, forthcoming. [5,27] ——— (2022b): “Spreading Gangs: Exporting US Criminal Capital to El Salvador,” American Economic Review, 112, 1985–2024. [5,7,8,27,32,41] TENORIO,OSCAR (2002): “Tierra de todos, tierra de nadie,” Tech. rep., El Diario de Hoy. [8] TILLY,CHARLES (1985): “War Making and State Making as Organized crime,” in Collective Violence, Contentious Politics, and Social Change, New York: Routledge, 121–139. [2] TITA,GEORGE E., JACQUELINE COHEN,AND JOHN ENGBERG (2005): “An ecological study of the location of gang “set space”,” Social problems, 52, 272–299. [7] VEGA,MARIO (2015): “El mapa de las pandillas,” Tech. rep., El Diario de Hoy, https://historico.elsalvador.com/historico/145152/el-mapa-de-las-pandillas.html. [8] WALTHER,OLIVIER J., LAWALI DAMBO,MOUSTAPHA KONÉ,AND MICHIEL VAN EUPEN (2020): “Mapping travel time to assess accessibility in West Africa: The role of borders, checkpoints and road conditions,” Journal of Transport Geography, 82, 10–1016. [6] WARD,THOMAS W (2013): Gangsters without borders: An ethnography of a Salvadoran street gang, Oxford: Oxford University Press. [7,8,9] WOMER,SARAH AND ROBERT J. BUNKER (2010): “Sureños gangs and Mexican cartel use of social networking sites,” Small Wars & Insurgencies, 21, 81–94. [7] ZOETHOUT,MARGRIET ANTOINETTE (2015): “Recovering Government Control over Mara Salvatrucha territory: Analysis based on the ‘Santa Tecla, a Municipality Free of Violence’ Agreement,” Police and Public Security Journal, 5, 179–246. [8,11] ——— (2016): “Consolidate State control in ‘The Mara Salvatrucha (Salvatrucha gang)’ territory: Lessons from community work in high-risk communities,” Police and Public Security Journal, 6, 115–204. [44] 38 ONLINE APPENDIX A. ADDITIONAL INFORMATION AND ROBUSTNESS CHECKS A.1. Stability of the Boundaries of Gang Territory To the best of our knowledge, the boundaries of gang territory have remained stable throughout the sample period. In particular, we contacted the PNC, inquiring about this issue, and multiple PNC officials confirmed that the boundaries of gang territory had had no significant changes since they were initially formed in the late 1990s and early 2000s. This information has also been confirmed by informal conversations with residents of San Salvador. To provide additional evidence that the boundaries of gang territory did not change in time, we take advantage of the following fact. As described in Subsection 2.4, both MS-13 and 18th Street consider outsiders a threat to their security. Thus, a disproportionate number of gangrelated homicides take place at the boundaries of gang territory (both between the gangs and the state and between the two gangs) because of outsiders attempting to enter gang neighborhoods without permission (Martínez, 2016). Leveraging this fact, we consider geocoded data on all gang-related homicides that were committed in San Salvador in 2003-2014 and split it into two subsamples: those that took place in the first six years of the sample period (2003-2008) and those that took place in the latest six years of the sample period (2009-2014). For each of the homicides, we identify whether it took place in a gang location and calculate the distance to the boundaries of gang territory (either between the gang and the state or between the two gangs). Panel A of Appendix Figure A.10 presents the number of gang-related homicides that took place in 2003-2008 by 10-meter bins on either side of the boundaries of gang territory; Panel B of Figure A.10 provides a similar illustration for gang-related homicides in 2009-2014. In both cases, the number of homicides was particularly high in areas close to the boundaries of the gang neighborhoods from the EDH map, confirming that the map correctly identifies the boundaries of gang territory in the two periods.48 In turn, the fact that the highest number of gang-related homicides took place in the same locations both in 2003-2008 and 2009-2014 suggests that the boundaries of gang territory have remained stable during this period. A.2. Selective Migration: Out-of-Sample Migration In Subsection 4.3, we demonstrated that our main results are not driven by selective insample migration: individuals moving to or from gang-controlled neighborhoods, while remaining in San Salvador municipality. Another type of selective migration that can potentially affect the interpretation of our results is out-of-sample migration: individuals moving from San 48Notably, as shown in Figure A.10, there are multiple gang-related homicides outside of gang territory. We provide a detailed discussion of this fact in Section 5. Also, as we show in Section A.3, the results in Table Iare robust to excluding observations from neighborhoods close to the regression discontinuity cutoff (see Table A.III). Thus, while the location of the gang-related homicides allows us to validate the boundaries of gang territory from the EDH maps, the results in Table Iare not driven by areas with the highest numbers of gang-related homicides. GANGS, LABOR MOBILITY, AND DEVELOPMENT 39 Salvador to a different municipality or abroad. In particular, if rich, educated individuals who initially lived in gang-controlled neighborhoods were more likely to move out of San Salvador than poor and uneducated individuals from the same areas, it could imply that the results in Table Iare partly determined by this change in the composition of the population. We analyze this mechanism in the following ways. First, we calculate the rates of selective out-of-sample migration from gang-controlled neighborhoods that would be required to generate the discontinuities from Table I. For each of the binary household-level characteristics, we define a household to be “rich” if it has that characteristic (e.g., a phone, a computer) and “poor” if it does not. The only exception is the variable for not having a bathroom, which is defined in the opposite way. Similarly, for each of the individual-level characteristics, we define an individual to be “educated” if they have that characteristic (e.g., a high school degree, a university degree) and “uneducated” if they do not. We make the conservative assumption that outside of gang territory, the probability of moving out of San Salvador is the same for all individuals and that in gang neighborhoods, poor and uneducated individuals migrate out of sample with probability ϱ.49 Then, for a given ϱ, we calculate the share of rich households and educated individuals from gang territory that needed to move out of San Salvador to generate the discontinuities for each of the outcome variables. We use the example of the share of households with a computer to show how these rates were calculated. From the regression output, we get the predicted share of households with a computer for observations zero meters away from the boundaries of gang territory, separately for locations inside and outside of gang territory. We denote those numbers as Gand NG, respectively. We further denote the number of “rich” households (i.e., those that have a computer) in gang-controlled areas before any migration took place as xand the share of “poor” households (i.e., those that do not have a computer) as 1↓x. Next, we assume that a fraction ωof the “rich” households and a fraction ϱof the“poor” households migrated out of sample. Thus, in the data, we observe the following relationship. (1 ↓ω)x (1 ↓ω)x+(1↓ϱ)(1 ↓x)=G.(4) Then, assuming different values of ϱ, we calculate the value of ωthat would make this relationship hold if, in the absence of migration, there would not have been any difference in the outcome variable between gang and nongang locations (i.e., x=NG). Appendix Table A.XIII presents the results of these calculations for ϱequal to 0%, 10%, and 20%. Even if we unrealistically assume ϱ=0% (i.e., that poor and uneducated individuals from gang areas do not have a chance to move out of San Salvador), on average, the rate of out-of-sample migration for rich, educated individuals would have to be as high as 51.7% to generate the discontinuities from Table I. For higher values of ϱ, this rate is even higher. 49If rich, educated individuals from nongang areas are more likely to migrate out of sample, that would make the required rates of selective out-of-sample migration from gang territory even higher. 40 Can the rate of out-of-sample migration for rich individuals be that high? To address this question, we take advantage of the fact that, until the mid-2010s, international migration of entire families had been very rare.50 International migration is expensive: e.g., the costs of migrating from El Salvador to the United States—the most popular destination among Salvadoran migrants—are approximately $12,500 (Kulish,2018). In turn, the average monthly household income in San Salvador is only $625. Thus, even to send one family member abroad, Salvadoran households have to save up for a long time, and migration of entire families is incredibly rare. This fact allows us to estimate the rate of out-of-sample migration by considering whether a household has a family member who moved abroad in 1997-2007 (the 2007 census contains this information). In addition, by looking at the correlation between the probability of a family member moving abroad and the first principal component of the household characteristics, we are able to estimate the extent to which individuals from rich households were more likely to migrate out of San Salvador. Appendix Table A.XIV presents the results of estimating Specification (1) for the probability of a household having a family member who moved abroad in 1997-2007. On average, only 6% of the households have a family member who moved abroad, and this rate does not change at the boundaries of gang territory. We also find that rich households both inside and outside of gang territory are more likely to have a family member living abroad. However, the correlation between wealth and out-of-sample migration in gang and nongang areas are not statistically different from one another. Moreover, although rich households are more likely to have a family member who moved abroad, the magnitude of that effect is much smaller than the rates of selective out-of-sample migration from Appendix Table A.XIII that are required to generate the discontinuities. In gang territory, an increase in the first principal component of the household characteristics from zero to one (i.e., the difference between the poorest and richest household) increases the probability of the household having a family member move abroad by only 7.1%, whereas the estimates from Table A.XIII suggest that, even under the unrealistic assumption of ϱ=0%, the rate of out-of-sample migration for rich households needs to be at least 51.7% to explain the discontinuities. Therefore, out-of-sample migration can account for no more than 100 →7.1/51.7 =13.7 percent of the effects in Table I.5152 50For instance, according to United States Customs and Border Protection, in 2012, the number of apprehensions of individuals in family units constituted less than 3% of all apprehensions of Salvadoran citizens at the Southwest border of the United States. In previous years, that number was even smaller. 5113.7% should be interpreted as the upper bound for the share of the results that can be explained by out-ofsample migration for the following reasons. First, the 7.1% number assumes that there is no selective out-of-sample migration outside of gang territory. If there is selective out-of-sample migration from nongang areas, as suggested by the results in Table A.XIV, then this number should be lower. Second, it is possible that some households with a family member abroad have increased their wealth because of that fact (e.g., because of receiving remittances). If that is the case, the results from Table A.XIV would overestimate the probability of individuals from rich households migrating out of sample. Finally, the 50% number required to generate the discontinuities in Table Iis calculated under the assumption that poor individuals are unable to migrate out of sample at all. If poor individuals also have a chance of migrating out of sample, this number should be higher. 52We also perform a test in the spirit of McCrary (2008) to check whether, at the boundaries of gang territory, there is a discontinuous change in population density for various groups of the population. If individuals from gang- GANGS, LABOR MOBILITY, AND DEVELOPMENT 47 The solution to the profit maximization problem of these firms yields L(wl)="ω¯ K wl# 1 1→ω .(10) ς(wl)=wl ω ω→1¯ K1 1→ω$ωω 1→ω↓ω1 1→ω%↓c↓r. (11) The free entry condition implies that, in equilibrium, the firms’ profits are equal to zero. w↑↑ l ω ω→1¯ K1 1→ω$ωω 1→ω↓ω1 1→ω%↓c↓r=0 =↗w↑↑ l ω 1→ω= ¯ K1 1→ω$ωω 1→ω↓ω1 1→ω% c+r=↗ (12) =↗w↑↑ l=  ¯ K1 1→ω$ωω 1→ω↓ω1 1→ω% c+r  1→ω ω =↗L↑↑ =ω1 1→ω(c+r)1 ω ¯ K1 ω(ωω 1→ω↓ω1 1→ω)1 ω . Empirically we observe that firms far away from gang territory (i.e., in (b+ϑ,1]) pay higher wages and hire more workers. For the model to match these empirical findings, the following conditions need to be satisfied.   ¯ K1 1→ω$ωω 1→ω↓ω1 1→ω% c+r  1→ω ω >&ωω 1→ω↓ω1 1→ω c' 1→ω ω =↗¯ K>(1+r c)1↓ω(13) ω1 1→ω(c+r)1 ω ¯ K1 ω(ωω 1→ω↓ω1 1→ω)1 ω >ω1 1→ωc1 ω (ωω 1→ω↓ω1 1→ω)1 ω =↗¯ K<1+r c(14) Together, these two conditions yield ¯ K↔(1+r c1↓ω,1+r c).65 Intuitively, condition (13) states that ¯ Kneeds to be sufficiently high for the productivity gains from ¯ Kto offset the decrease in demand due to the necessity of paying rto adopt the technology. In turn, condition (14) states that firms will only hire more workers if ¯ Kis sufficiently low that the increase in wages does not negate the increase in labor demand due to higher productivity. Finally, to ensure that the described situation represents an equilibrium in this market, we need to ensure that firms far away from gang territory (i.e., in (b+ϑ,1]) have incentives to purchase the productivity-enhancing equipment instead of using the labor-only technology. Given (12), this condition requires that w↑↑ l ω ω→1$ωω 1→ω↓ω1 1→ω%↓c<0=↗¯ K>(1+r c)1↓ω .(15) 65For instance, this condition would be satisfied for ¯ K=2,r=2,c=1,and ϖ=0.5. 48 Thus, (13) guarantees that the firms do not have incentives to change their production behavior. Counterfactual analysis. Using the theoretical framework outlined above, we now provide a counterfactual analysis of two hypothetical scenarios: the elimination of gang-imposed restrictions on individuals’ mobility and the full removal of gang presence. In the setting of our model, under the former scenario, firms in [0,b+ϑ]would still be unwilling to adopt the productive technology. As a result, both aggregate production and aggregate employment in productive firms, which is determined by the free entry condition, would remain the same. Individuals living in [0,b]would still benefit from their ability to commute to work in high-productivity firms, but given the zero-sum nature of this economy, these benefits would come at the expense of workers living in [b, 1]. On the other hand, if gang presence were fully eliminated, firms in [0,b+ϑ]would find it profitable to adopt the productive technology, leading to both higher aggregate production and higher demand for labor. Consequently, benefits would accrue to individuals living in all parts of the city, although residents of former gang-controlled neighborhoods would gain the most due to their previous inability to work in high-productivity firms. In mid-2022, El Salvador witnessed unprecedented government-led crackdowns on criminal activity, leading to the de facto implementation of this latter scenario. In a separate project, we provide a detailed analysis of the consequences of these crackdowns, showing how former gang-controlled neighborhoods experienced significant improvements in mobility (Melnikov et al.,2024). ONLINE APPENDIX TABLES GANGS, LABOR MOBILITY, AND DEVELOPMENT 49 TABLE A.I SOCIOECONOMIC CHARACTERISTICS FROM THE 2019 SURVEY Has a high Has a university Household Works in a firm with Works in a firm with school degree degree income ↘100 employees ↘200 employees (1) (2) (3) (4) (5) Panel A: All survey respondents Gang territory -0.311*** -0.254*** -352.60*** -0.123*** -0.115*** (0.057) (0.062) (112.22) (0.019) (0.028) Mean of dep. var. 0.508 0.180 625.0 0.169 0.133 Observations 2,275 2,275 2,314 2,071 2,071 Panel B: Respondents who have lived in the same location their entire life Gang territory -0.281*** -0.173*** -271.05** -0.114*** -0.104** (0.061) (0.056) (118.14) (0.033) (0.041) Mean of dep. var. 0.474 0.149 602.3 0.155 0.123 Observations 1,757 1,757 1,787 1,589 1,589 Note: *** p<0.01, ** p<0.05, * p<0.1. After years of gang control, gang-controlled areas have worse socioeconomic conditions than neighboring areas that were not under the control of gangs. The table presents the results of estimating Specification (1) for the variables from the 2019 survey. Panel A presents the results for the full sample; Panel B—for the subsample of respondents who have always lived in the same location. For household income, the unit of observation is a household; for all the other variables—an individual. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 50 TABLE A.II PLACEBO:EFFECTS OF MAJOR ROADS THAT DID NOT DEFINE THE BORDERS OF GANG TERRITORY Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Placebo treatment group -0.147 -0.018 0.026 0.033 -0.008 0.010 (0.151) (0.044) (0.036) (0.059) (0.028) (0.076) Mean of dep. var. 0.883 0.033 0.979 0.064 0.012 0.074 Observations 6,716 5,623 5,714 5,714 5,714 5,374 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Placebo treatment group 0.007 -0.048 -0.091 -0.022 -0.016 -0.130 (0.014) (0.147) (0.150) (0.035) (0.106) (0.377) Mean of dep. var. 0.026 0.222 0.528 0.936 0.171 2.427 Observations 5,308 5,411 5,434 5,475 5,419 5,714 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Placebo treatment group -0.034 0.022 0.007 -0.035 -0.009 0.001 (0.028) (0.080) (0.059) (0.081) (0.054) (0.050) Mean of dep. var. 0.892 0.299 0.0920 0.930 0.313 0.360 Observations 19,130 18,563 18,563 5,623 5,238 18,563 Demographic characteristics Neighborhood characteristics Female Age Urban territory Road density Elevation Tree coverage (19) (20) (21) (22) (23) (24) Placebo treatment group 0.016 -0.741 -0.002 -9.353 -51.593 -0.010 (0.017) (1.920) (0.062) (9.426) (52.059) (0.016) Mean of dep. var. 0.541 29.63 0.954 19.54 676.8 0.008 Observations 20,967 20,967 47 47 47 47 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (2), using the locations of major roads that did not contribute to the formation of the boundaries of gang territory as a placebo. Similarly to Table III, we exclude 25% of the largest census tracts, which are predominantly present outside gang territory. We also include dummies for the three remaining quartiles of the census tract size distribution. Table S.III in the Supplementary Materials reports the results of estimating the same regression specification without excluding the largest census tracts, and instead, including dummies for all four quartiles of the census tract size distribution. The unit of observation is a dwelling, household, individual, or census tract, depending on which characteristics are being considered. In the individuallevel regressions, the sample consists of the entire population. Omitted controls include a dummy for gang territory as well as a linear trend in distance to the placebo boundaries, separately for locations on each side of the placebo boundaries and on each side of the boundaries of gang territory. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the placebo boundaries (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 51 TABLE A.III EXCLUDING OBSERVATIONS WITHIN 100 METERS OF THE BOUNDARIES OF GANG TERRITORY Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.067*** 0.054*** -0.092*** -0.101*** 0.002 -0.176*** (0.019) (0.011) (0.028) (0.017) (0.003) (0.024) Mean of dep. var. 0.936 0.026 0.943 0.116 0.004 0.194 Observations 50,183 42,287 43,258 43,258 43,258 41,726 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.032*** -0.288*** -0.204*** -0.036*** -0.239*** -1.006*** (0.008) (0.048) (0.052) (0.008) (0.045) (0.235) Mean of dep. var. 0.034 0.456 0.708 0.954 0.362 3.179 Observations 41,205 41,911 41,964 42,108 41,860 43,258 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.040*** -0.208*** -0.163*** -0.058*** -0.126*** -0.136*** (0.009) (0.029) (0.028) (0.013) (0.022) (0.021) Mean of dep. var. 0.931 0.464 0.223 0.955 0.388 0.533 Observations 144,977 141,210 141,210 42,287 40,651 141,210 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the variables from the 2007 census after excluding observations within 100 meters of the boundaries of gang territory. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 52 TABLE A.IV CONTROLLING FOR 300→300 METER FIXED EFFECTS Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.052* 0.023*** -0.073*** -0.097*** 0.006*** -0.160*** (0.030) (0.007) (0.026) (0.025) (0.002) (0.028) Mean of dep. var. 0.932 0.028 0.941 0.108 0.005 0.180 Observations 72,087 60,675 62,169 62,169 62,169 59,776 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.010* -0.224*** -0.135*** -0.019 -0.190*** -0.641*** (0.006) (0.047) (0.032) (0.011) (0.037) (0.207) Mean of dep. var. 0.033 0.428 0.697 0.952 0.346 3.093 Observations 59,096 60,045 60,168 60,384 60,020 62,169 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.031*** -0.137*** -0.101*** -0.040** -0.100*** -0.089*** (0.006) (0.031) (0.032) (0.017) (0.021) (0.023) Mean of dep. var. 0.928 0.449 0.208 0.952 0.378 0.522 Observations 208,416 202,935 202,935 60,675 58,293 202,935 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the variables from the 2007 census, controlling for 300↔300 meter fixed effects. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include 300↔300 meter fixed effects and a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 53 TABLE A.V RESTRICTIONS ON INDIVIDUALS’MOBILITY AND LABOR-MARKET OUTCOMES,BY CAR AVAILABILITY Works in Works in same Has been to Has been to Freedom of gang territory neighborhood Santa Ana the beach movement where they live where they live (1) (2) (3) (4) (5) Gang territory →0.238*** 0.102** -0.228*** -0.090** -0.105* →Gets to work by car or motorcycle (0.059) (0.044) (0.052) (0.042) (0.055) Gang territory →0.532*** 0.062* -0.214*** -0.044 -0.105** →Gets to work in another way (0.041) (0.035) (0.044) (0.034) (0.040) Mean of dep. var. 0.334 0.302 0.511 0.876 0.809 Observations 1,738 2,071 2,071 2,071 2,071 Household Works in a Works in a Gang borders prevented income firm with firm with you from finding jobs ↘100 employees ↘200 employees in large firms in other parts of the city (6) (7) (8) (9) (10) Gang territory →-401.78** -0.144*** -0.153*** 0.132** →Gets to work by car or motorcycle (165.87) (0.037) (0.047) (0.050) Gang territory →-257.37*** -0.080*** -0.073** 0.091** →Gets to work in another way (92.81) (0.024) (0.031) (0.042) Gang territory →Has a car 0.089* (0.048) Gang territory →Does not have a car 0.105** (0.046) Mean of dep. var. 629 0.169 0.133 0.407 0.407 Observations 2,071 2,071 2,071 2,313 2,313 Note: *** p<0.01, ** p<0.05, * p<0.1. Columns 1–9 of the table present the results of estimating Specification (1) for mobility and labor-market outcomes with the dummy for gang territory replaced with dummies for individuals in gang territory who get to work by car or motorcycle and individuals in gang territory who get to work in some other way. In Column 10, the dummy for gang territory is replaced with similar dummies for individuals in gang territory who own a car and individuals in gang territory who do not. The sample consists only of survey respondents who have a job. Santa Ana is a neighboring municipality, which is approximately 60 kilometers away from San Salvador. The beach is approximately 30 kilometers away from San Salvador. The unit of observation is an individual. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries, as well as a dummy for getting to work by car or motorcycle (in Column 10, a dummy for owning a car). Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 54 TABLE A.VI RESTRICTIONS ON INDIVIDUALS’MOBILITY AND LABOR-MARKET OUTCOMES,BY EDUCATION Works in Freedom of Gang borders prevented Household Works in a Works in a gang territory movement you from finding jobs income firm with firm with where they live in large firms in ↘100 employees ↘200 employees other parts of the city (1) (2) (3) (4) (5) (6) Gang territory →0.323*** -0.072* 0.115** -378.77*** -0.085*** -0.087** →High school degree (0.043) (0.041) (0.042) (113.63) (0.028) (0.036) Gang territory →0.577*** -0.106** 0.083* -158.74 -0.052* -0.057* →No high school degree (0.035) (0.043) (0.047) (106.20) (0.028) (0.030) Mean of dep. var 0.330 0.812 0.407 628.3 0.170 0.133 Observations 1,707 2,275 2,313 2,275 2,033 2,033 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the mobility and labor-market questions by education. The unit of observation is an individual. Omitted controls include a dummy for having a high school degree and a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). TABLE A.VII CONSEQUENCES OF LOW LABOR MOBILITY Household income Works in a firm with Works in a firm with ↘100 employees ↘200 employees (1) (2) (3) (4) (5) (6) (7) (8) (9) Lives in gang territory -352.60 -429.99 -235.09 -0.123 -0.210 -0.105 -0.115 -0.187 -0.102 (112.22)*** (127.82)*** (112.56)** (0.019)*** (0.022)*** (0.023)*** (0.028)*** (0.025)*** (0.030)*** [84.97]*** [98.80]*** [81.33]*** [0.042]*** [0.046]*** [0.041]*** [0.035]*** [0.038]*** [0.035]*** Lives in gang territory, 167.64 85.39 0.182 0.129 0.152 0.110 works in nongang territory (32.69)*** (30.23)*** (0.026)*** (0.025)*** (0.027)*** (0.026)*** [37.08]*** [38.73]** [0.025]*** [0.024]*** [0.024]*** [0.023]*** Has a high school degree 89.11 0.124 0.088 (19.90)*** (0.021)*** (0.018)*** [26.78]*** [0.020]*** [0.019]*** Has a university degree 445.46 0.148 0.132 (76.96)*** (0.029)*** (0.027)*** [62.62]*** [0.032]*** [0.030]*** Mean of dep. var. 625.00 634.70 638.90 0.169 0.169 0.170 0.133 0.132 0.132 Observations 2,314 1,738 1,707 2,071 1,738 1,707 2,071 1,738 1,707 Note: *** p<0.01, ** p<0.05, * p<0.1. The table shows that the discontinuity in income and firm size is significantly smaller or nonexistent for individuals living in gang territory but working outside of gang territory. All the variables come from the 2019 survey. For household income, the unit of observation is a household; for the other variables—an individual. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. In the columns that include a variable for working outside of gang territory, we additionally include a dummy for whether the individual is currently employed (in the survey, unemployed individuals were asked to describe their most recent work experience). Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory, separately for each side of the boundaries. Standard errors in brackets are adjusted to allow for spatial correlation within a 100 meter radius (Conley correction). GANGS, LABOR MOBILITY, AND DEVELOPMENT 55 TABLE A.VIII RESTRICTIONS ON INDIVIDUALS’MOBILITY AND LABOR-MARKET OUTCOMES,BY GENDER Works in Freedom of Gang borders prevented Household Works in a Works in a gang territory movement you from finding jobs income firm with firm with where they live in large firms in ↘100 employees ↘200 employees other parts of the city (1) (2) (3) (4) (5) (6) Gang territory →Male 0.454*** -0.077* 0.119** -370.07*** -0.138*** -0.116*** (0.042) (0.043) (0.050) (114.52) (0.034) (0.037) Gang territory →Female 0.520*** -0.107** 0.084** -332.33*** -0.108*** -0.110*** (0.045) (0.041) (0.041) (107.53) (0.019) (0.030) Mean of dep. var 0.334 0.811 0.407 625 0.169 0.133 Observations 1,738 2,314 2,313 2,314 2,071 2,071 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the mobility and labor-market questions by gender. The unit of observation is an individual. Omitted controls include a dummy for being female and a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). TABLE A.IX FIRMS’LOCATION,PROFITS,REVENUE,AND COSTS Log of firms per km2: Log of the firm’s: Census: Google Maps: Profits Revenue Costs Employees Costs per All firms All firms Cafes & Grocery Pharmacies employee restaurants stores (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gang territory -0.198 -0.027 0.094 -0.019 0.133 -0.303 0.034 -0.197 0.113 -0.056 (0.362) (0.332) (0.330) (0.121) (0.239) (0.406) (0.103) (0.182) (0.144) (0.148) Mean of dep. var. 9.767 10.97 10.44 1.756 8.679 5.062 4.860 2.394 1.882 1.251 Observations 5,631 6,118 6,083 6,120 6,083 156 86 86 86 86 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the number of business establishments and their characteristics. The results in columns 1–6 are based on the supplement to the 2005 economic census. In columns 1–5, the unit of observation is a firm; in columns 6—a sector, the analogue of the census tract in the economic census. The data on the number of business establishments in columns 7–10 come from Google Maps. In these regressions, the unit of observation is a 10 meter bin, denoting distance to the boundaries of gang territory, weighted by the size of the area of the distance bins. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 56 TABLE A.X GANG CONTROL AND DROPOUT RATES Dropout rate Subsample: All obs. Year ↑2007 Year >2007 Male Female All obs. (1) (2) (3) (4) (5) (6) Gang territory 0.019*** 0.021** 0.018*** 0.021*** 0.019*** (0.004) (0.008) (0.004) (0.006) (0.003) Gang territory →Standard program 0.019*** (0.004) Gang territory →Program for adults 0.038** (0.018) Mean of dep. var. 0.020 0.021 0.019 0.023 0.016 0.020 Observations 3,199 684 2,515 3,088 3,186 3,377 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results for estimating Specification (1) for the dropout rates for schools in San Salvador. The data come from the annual census of schools. In columns 1–5, the unit of observation is a school in a year. In these results, omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. In column 6, the unit of observation is the type of program (standard or for adults) in a school in a year. In these results, omitted controls include a dummy for the program being for adults and linear trends in distance to the boundaries of gang territory, separately for each type of program on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). Standard errors in brackets are adjusted to allow for spatial correlation within a 100 meter radius (Conley correction). TABLE A.XI GANG CONTROL AND EXAM SCORES Math Natural sciences Social sciences Languages & literature Subsample: All obs. Year ↑2007 All obs. Year ↑2007 All obs. Year ↑2007 All obs. Year ↑2007 (1) (2) (3) (4) (5) (6) (7) (8) Gang territory -0.835** -0.801** -0.652** -0.603** -0.666*** -0.686** -0.712*** -0.649** (0.337) (0.331) (0.248) (0.250) (0.234) (0.278) (0.240) (0.252) Mean of dep. var. 5.434 5.511 5.776 5.901 6.432 6.382 6.151 5.960 Observations 1,284 436 1,284 436 1,284 436 1,284 436 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results for estimating Specification (1) for the average exam scores in San Salvador schools. The data come from the schools’ administrative records in 1999-2001 and 2005-2017. The unit of observation is a school in a year. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 63 FIGURE A.4.—Household Income After 22 Years of Gang Control Note: The figure illustrates the results for household income from Appendix Table A.I. The left-hand side of the figure presents the results for the full sample (Panel A of Appendix Table A.I), the right-hand side—for the subsample of individuals who have lived in the same location all their life (Panel B of Appendix Table A.I). The results are very similar. The vertical axis represents the average value of household income; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. FIGURE A.5.—Socioeconomic Conditions Before the Gangs’ Arrival: 1st Principal Components of the Dwelling, Household, and Individual Characteristics Note: The figure illustrates the results for the 1st principal components of the dwelling, household, and individual characteristics from Table II. All the variables come from the 1992 census. The unit of observation is a dwelling, a household, and an individual, depending on the specification. All the variables are normalized to vary between zero and one with higher values representing better outcomes. The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. 64 FIGURE A.6.—Stability of the Boundaries of Gang Territory Note: The outcome variable comes from the 2023 individual survey. The vertical axis represents the average share of survey respondents who agree with the statement that the gangs controlled their neighborhood 20 years ago, during the presidency of Francisco Flores Pérez (President of El Salvador from 1999 to 2004); the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. FIGURE A.7.—Excluding Observations Within 100 Meters of the Boundaries of Gang Territory Note: The figure illustrates the regression discontinuity plots for the 1st principal components of the dwelling, household, and individual characteristics from the 2007 census after excluding observations within 100 meters of the boundaries of gang territory. The unit of observation is a dwelling, a household, and an individual, depending on the specification. All the variables are normalized to vary between zero and one with higher values representing better outcomes. The vertical axis represents the average value of the outcomes variable; the horizontal axis— distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. GANGS, LABOR MOBILITY, AND DEVELOPMENT 65 FIGURE A.8.—Firm Characteristics, by Distance to the Boundaries of Gang Territory Note: The figure illustrates the results for firm characteristics from columns 1, 4, and 5 of Table A.IX and column 2 of Table V. The first three variables come from the 2005 economic census; the fourth one—from the 2015 survey of firms conducted by FUSADES. The unit of observation is a firm. The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. 66 FIGURE A.9.—Synthetic Difference-in-Differences Note: The figure illustrates the results of the synthetic difference-in-differences analysis. The plots on the left side of the figure represent the effects of gang presence when the treatment group is defined in the same way as in the baseline difference-in-differences analysis. The plots on the right side of the figure present the same analysis with the treatment group narrowed down to the four largest cities in El Salvador, all of which have had a substantial gang presence since the late 1990s. GANGS, LABOR MOBILITY, AND DEVELOPMENT 67 FIGURE A.10.—Gang-related Homicides, by Distance to the Boundaries of Gang Territory Panel A Panel B Note: The figure illustrates the number of gang-related homicides in 2003-2008 (Panel A) and 2009-2014 (Panel B), by distance to the boundaries of gang territory. In both cases, the largest number of the homicides took place right at the boundaries of gang territory. The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 10 meter bin. 68 S. SUPPLEMENTARY MATERIALS: NOT FOR PUBLICATION S.1. Additional Data Sources Homicides and robberies. The data on gang-related homicides come from the PNC and cover 2003 to 2014. For each observation, we obtained information about the time and day it occurred, whether the perpetrator was a member of a gang, and the address of occurrence. Using these addresses, we manually geocoded the data to obtain the latitude and longitude of the homicides carried out by gang members. The PNC also shared with us the data on gangrelated homicides in 2000, but these data are available only at the municipality level. The data on robberies come from the Metropolitan Planning Office for San Salvador (Oficina de Planficación del Área Metropolitana de San Salvador, OPAMSS) and cover 2014 to 2015. They contain information about the time, date, and location of all robberies, including their latitude and longitude. Incarceration data. The data on incarcerations come from the General Directorate of Prisons (Dirección General de Centros Penales, DGCP) and represent the universe of all individuals incarcerated in El Salvador since the mid 1980s. The records contain information about the crimes the individual committed, the date of incarceration, the municipality of birth, and the last known address. For inmates who entered prison before 1997 and whose last known address is in San Salvador municipality (4,726 individuals), we manually geocoded the residential addresses to obtain the precise geocoordinates used in the analyses. Given that geocoded crime data prior to 2003 are unavailable, the inmates’ residential addresses represent the best measure of criminal activity in the pretreatment period. Household surveys. To analyze household income data beyond San Salvador, we obtained yearly household surveys from DIGESTYC spanning 1992 to 2007. These surveys sample over 10,000 households and include a broad range of questions. In our analysis, we focus on the question related to household income which was asked throughout the years. 2023 survey. In the summer of 2023, we conducted a follow-up survey in gang and nongang neighborhoods in San Salvador, following a similar sampling protocol as our 2019 survey described in Section 3. This survey included in-person interviews with questions about individuals’ mobility and employment. Due to improved security in the country, we were also able to ask about gang activity in their neighborhoods, including questions related to the gangs’ system of borders. Cell phone GPS pings. We purchased cell phone ping data from Quexopa, a data aggregator specializing in Latin America, covering January and February of 2022.66 The dataset includes a unique cell phone identifier, time of capture, operating system, GPS coordinates, and accuracy metrics. 66Although we sought data from 2019, this was the earliest period available. Due to the large volume of data, it is common practice to delete records older than one year for storage reasons. GANGS, LABOR MOBILITY, AND DEVELOPMENT 69 Urban territory. The data on urban density come from New York University’s Atlas of Urban Expansion. The raster map presents the urban areas in the Greater San Salvador region in 1999.67 We transform the data into a binary raster, equal to one when the location is classified as urban. Then, for each of the census tracts from the 2007 census, we calculate the share of census tracts’ territory that is urban. Waterways. The map of the waterways in El Salvador comes from the Humanitarian OpenStreetMap Team.68 Then, for each of the census tracts from the 2007 census, we created a dummy variable for whether the census tract contains a part of the waterway. Road density. The map of the roads in El Salvador comes from the Humanitarian OpenStreetMap Team and reflects the roads in the country in March 2020.69 We then transform the feature-based map into a binary raster file with the resolution of 1 meter→1 meter, where we replace the lines for roads with grid cells equal to one. After that, for each of the census tracts from the 2007 census, we calculate road density, measured in kilometers per square kilometer. Elevation. The data on elevation at the resolution of 3 arc seconds (approximately 90 meters) come from the CGIAR-Consortium for Spatial Information (CGIAR-CSI).70 For each of the census tracts from the 2007 census, we calculate the average elevation inside the census tract. Territory used for coffee production. The map of land use in 1998 (including coffee production) comes from the Ministry of Environment and Natural Resources. We convert the feature-based map into a binary raster, equal to one for areas that are used for coffee production. Then, for each of the census tracts from the 2007 census, we calculate the share of their territory that is used for coffee production. Tree coverage. The data on tree coverage in 2000 come from Global Forest Watch.71 The raster file presents the share of territory covered by trees in each 30 meter→30 meter grid cell. For each of the census tracts from the 2007 census, we calculate the average level of tree coverage inside of the census tract. High school exam scores. The data on the schools’ average high school exit exam scores (Prueba de Aprendizaje y Aptitudes para Egresados de Educación Media, PAES) come from the Ministry of Education. PAES results are reported for math, natural sciences, social sciences, and Spanish language and literature. The data cover the period from 1999 to 2017, but exclude the results for 2002-2004 because in those years the Ministry of Education applied a nondisclosed curve to the test scores, preventing comparison with the other years. 67The data can be accessed here: Atlas of Urban Expansion: San Salvador (accessed on May 4, 2020). 68The map of the waterways in El Salvador can be accessed here: Humanitarian Data Exchange: El Salvador Waterways (accessed on May 4, 2020). 69The map of the roads in El Salvador can be accessed here: Humanitarian Data Exchange: El Salvador Roads (accessed on May 4, 2020). 70The elevation map for El Salvador can be accessed here: CGIAR-CSI (accessed on May 4, 2020). 71The data on tree coverage for El Salvador can be accessed here: Global Forest Watch (accessed on May 4, 2020). 70 2020 survey. In 2020, we conducted a survey of 1,957 individuals in San Salvador to evaluate the extent of gang-related extortion in gang and nongang areas. The survey followed the same procedure as the 2019 survey, except that it was conducted over the telephone. The main reason for conducting the survey over the telephone is that, in in-person interviews, extortionrelated questions could have potentially endangered the enumerators. At the beginning of the survey, the enumerators asked the respondents for their address, and the survey proceeded if the address was in one of the census segments randomly chosen in the sampling procedure. 2005 economic census. The microdata for the 2005 economic census was provided by DIGESTYC.72 After creating a registry of all formal and informal firms in the country, DIGESTYC took a random sample of all the firms to ask a long-form questionnaire on income sources, production and remuneration costs, the year the firm was established, etc. From these questions, DIGESTYC calculated the firms’ revenue and costs. In total, the registry includes 179,817 firms across the country, while the long-form questionnaire covers 46,864 firms (26%). In the analysis, we focus on the long-form questionnaire firms based in San Salvador (6,120 firms). Locations of schools, hospitals, and other establishments. The data on the locations of schools, hospitals, and other establishments in San Salvador come from Google Maps.73 In August 2019, we scraped the data from Google Maps to identify all the establishments in San Salvador. In total, we obtained a dataset with 7,732 establishments. For each observation, Google provides a classification of the type of establishment (e.g., school, hospital, pharmacy). Housing rent. To obtain information on housing rent, in August-September 2018, we scraped the data from the most popular website for rent listings in El Salvador, OLX (now Encuentra24).74 We focused on noncommercial listings in which the entire apartment was being rented out (i.e., not a room in the apartment). The listings included the data on the latitude and longitude of the location, the rent requested by the landlord, as well as information about the apartment such as the number of bedrooms, the number of bathrooms, the number of square meters, and whether the apartment is being rented out by an agency. In total, the dataset contains 1,537 observations. It should be noted that we cannot observe whether a particular apartment was rented out or not. However, after two months, the vast majority of offers were no longer available. It should also be noted that, on average, the properties listed on OLX are larger and more expensive than the overall pool of properties in San Salvador. In particular, many of the cheapest properties may be rented out on the informal market and are not listed online. If there are more 72Although the census was carried out in 2005, the reference year for all the questions was 2004. 73We use the data on the locations of schools and hospitals from Google Maps instead of government records. The primary reason is the accuracy of the data. For instance, in the shapefile the government has provided to us, some of the schools are located outside of El Salvador. However, if we use the data from government records, the results are qualitatively very similar. 74The Salvadoran version of the website can be accessed here: OLX (now Encuentra24). GANGS, LABOR MOBILITY, AND DEVELOPMENT 71 such properties in gang-controlled neighborhoods, our estimates would provide a lower bound on the actual drop in housing rent at the boundaries of gang territory. Nighttime light density. Annual data on nighttime light density (or luminosity) come from the Defense Meteorological Satellite Program-Operational Linescan System (DMSP-OLS) and spans the period from 1992 to 2013.75 In particular, we use the DMSP-OLS data, representing the average stable lights from cities, towns, and other sites with persistent lighting. The data are provided by the National Centers for Environmental Information (NCEI). If for a particular year, the data were available from more than one satellite, we take the average of the two. The resolution of the data on nighttime light density is 30 arc seconds→30 arc seconds (i.e., approximately 1 kilometer→1 kilometer). Therefore, the data are not sufficiently precise to be used in the regression discontinuity design. Gang leaders’ municipalities of birth. The data on the gang leaders’ municipalities of birth come from El Faro, an investigative newspaper. We use the data from their investigative reports, focusing on the gang leaders who were deported from the United States and had been later convicted for committing crimes in El Salvador. Overall, the sample consists of 33 gang leaders both from MS-13 and 18th Street. We then manually match the names of the gang leaders and the crimes they committed to the criminal records from the Ministry of Justice and Public Security of El Salvador, which contain information on the offender’s municipality of birth. S.2. Further Details About the Primary Data Sources 2019 survey. For the 2019 survey, the following sampling procedure was applied. Given the uncertainty about their treatment status, census tracts within 15 meters of the boundaries of gang territory were excluded from the analysis. Then, separately for places inside and outside of gang territory, we split the census tracts into 30 meter bins, denoting the distance to the boundaries (i.e., 15-44 meters to the boundaries, etc.). After that we randomly selected 10 census tracts from each bin and surveyed 8-10 people in each of them.76 If there were fewer that 10 census tracts in that bin, we surveyed individuals in all the census tracts that were available. In total, the survey includes 2,314 respondents. To ensure the safety of the enumerators, if the survey team was denied entry into some of the gang-controlled neighborhoods, those census tracts were replaced by other ones from the same bin. If it was not possible to interview 10 individuals in a census tract (e.g., because after repeated attempts nobody answered the door), additional people were interviewed in other census tracts in the same bin. Gang boundaries. The map of gang-controlled neighborhoods that we use in this study is based on data from 2015. To the best of our knowledge, maps of gang-controlled areas for 75The data and a detailed description of it are available here: DMSP-OLS (accessed on May 4, 2020). 76In areas within 250 meters of the boundaries, we surveyed 10 individuals per census tract. In locations further away from the boundaries, we surveyed 8 individuals per census tract. 72 earlier years are nonexistent. However, according to multiple sources in the police department as well as conversations with the local population, the boundaries of gang territory in San Salvador have remained stable since the late 1990s when the boundaries were formed. If changes to the boundaries do occur, it tends to be a product of turf wars (i.e., MS-13 and 18th Street taking over each other’s territory); not because of the state regaining control over gang territories or the other way round. The data on the gang-controlled neighborhoods in San Salvador come from EDH and are presented in Figure 1. However, to accurately calculate distance to the boundaries of gang territory, we also complement these data with confidential maps from the police on the gangcontrolled neighborhoods outside of San Salvador municipality. Since the regression discontinuity design focuses on the census tracts inside of San Salvador, this never affects the treatment status of the census tract (i.e., whether or not it is located inside of gang territory). However, for the locations outside of gang territory, it does sometimes affect the distance from them to the boundaries of gang territory (i.e., if that location is closer to a gang-controlled location outside of San Salvador). It should be noted that, even with the extended map of gang territory, we are unable to implement the regression discontinuity design outside of San Salvador because the map additionally includes only a small number of locations in the Greater San Salvador area. 1992 and 2007 censal cartography. It should be noted that the boundaries of the census tracts in the 1992 and 2007 censuses were not the same. Therefore, we are not able to perform a difference-in-differences analysis at the level of the census tracts. However, in both cases, the size of the census tracts was quite similar, allowing us to accurately measure the distance from the census tract to the boundaries of gang territory. Thus, the distance between a particular location and the boundaries of gang territory is very similar, regardless of whether we use the 2007 or 1992 census tracts. It should also be noted that, although DIGESTYC digitized a map the 1992 census tracts, it did not fully finish that work. Specifically, the 1992 map does not have the boundaries of 18.9% of the census tracts in northwestern San Salvador. However, the vast majority of those neighborhoods are located more than 420 meters away from gang territory and, therefore, would not be included in the analysis in any case. In particular, nearly all of gang territory (except for a few small “islands”) and the neighborhoods right next to it are included in the 1992 map. Thus, it is highly unlikely that our estimates would change if all the census tracts were included.77 Extortion. Our measures on firm and household extortion payments draw from three sources. First, the data on whether firms have experienced extortion come from a survey of small and medium-sized enterprises conducted by the Salvadoran Foundation for Economic and Social Development (Fundación Salvadoreña para el Desarrollo Económico y Social, FUSADES). The survey also asked whether the firm has experienced gang activity in the location 77DIGESTYC also told us that the work on digitizing the map of the census tracts had to stop because of the lack of funding and that there was no specific reason why some census tracts were digitized and some were not. GANGS, LABOR MOBILITY, AND DEVELOPMENT 79 1 room in apartment, 2018 0.113 0.317 1,537 OLX 2 rooms in apartment, 2018 0.187 0.390 1,537 OLX 3 rooms in apartment, 2018 0.528 0.499 1,537 OLX 4 rooms in apartment, 2018 0.110 0.312 1,537 OLX 5 rooms in apartment, 2018 0.040 0.197 1,537 OLX 6 rooms in apartment, 2018 0.010 0.102 1,537 OLX 7+ rooms in apartment, 2018 0.012 0.108 1,537 OLX 1 bathroom in apartment, 2018 0.157 0.364 1,537 OLX 2 bathrooms in apartment, 2018 0.176 0.381 1,537 OLX 3 bathrooms in apartment, 2018 0.446 0.497 1,537 OLX 4 bathrooms in apartment, 2018 0.141 0.348 1,537 OLX 5 bathrooms in apartment, 2018 0.053 0.224 1,537 OLX 6 bathrooms in apartment, 2018 0.019 0.136 1,537 OLX 7+ bathrooms in apartment, 2018 0.008 0.092 1,537 OLX Square meters, 2018 189.38 264.65 1,537 OLX Rented out by agency, 2018 0.491 0.500 1,537 OLX Panel J: Other RDD variables Urban territory, 1999 0.814 0.297 476 NYU Atlas of Urban Expansion Road density (km per km2), 2020 17.84 8.81 476 Humanitarian OpenStreetMap Has access to waterway 0.326 0.469 476 Humanitarian OpenStreetMap Elevation 720.4 87.91 476 CGIAR SRTM Territory used for coffee production 0.028 0.132 476 Ministry of the Environment and Natural Resources Tree coverage, 2000 0.049 0.116 476 Global Forest Watch Panel K: Difference-in-differences variables Firm stock, 1988-2005 153.61 565.33 1,854 2005 census Luminosity, 1992-2013 10.18 14.07 2,288 DMSP-OLS Household income per capita, 1992-2013 103.42 115.52 171,257 Household surveys 80 TABLE S.II BOUNDARIES OF GANG TERRITORY FROM GEOGRAPHICAL BARRIERS, WITHOUT EXCLUDING THE LARGEST CENSUS TRACTS Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.078** 0.045** -0.054*** -0.073 0.008* -0.072 (0.036) (0.016) (0.015) (0.052) (0.004) (0.055) Mean of dep. var. 0.945 0.021 0.969 0.064 0.003 0.124 Observations 10,047 8,418 8,684 8,684 8,684 8,260 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.011 -0.274** -0.184*** -0.031** -0.220** -0.959** (0.007) (0.106) (0.060) (0.015) (0.091) (0.371) Mean of dep. var. 0.034 0.366 0.697 0.958 0.291 2.978 Observations 8,183 8,296 8,314 8,355 8,293 8,684 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.080** -0.248*** -0.178** -0.063** -0.104*** -0.168*** (0.035) (0.062) (0.068) (0.028) (0.035) (0.042) Mean of dep. var. 0.927 0.436 0.171 0.962 0.354 0.505 Observations 29,268 28,195 28,195 8,418 8,063 28,195 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1), using the locations of major roads and boulevards (geographical barriers) as the predicted boundaries of gang territory. We additionally control for dummies for the four quartiles of the census tract size distribution. All the variables come from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 81 TABLE S.III PLACEBO:EFFECTS OF MAJOR ROADS THAT DID NOT DEFINE THE BORDERS OF GANG TERRITORY, WITHOUT EXCLUDING THE LARGEST CENSUS TRACTS Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Placebo treatment group 0.002 -0.042 0.034 0.124 -0.005 0.082 (0.127) (0.036) (0.030) (0.087) (0.016) (0.091) Mean of dep. var. 0.897 0.030 0.980 0.113 0.010 0.147 Observations 9,441 7,678 7,806 7,806 7,806 7,296 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Placebo treatment group 0.029** 0.022 -0.064 -0.012 0.035 0.191 (0.013) (0.127) (0.096) (0.030) (0.099) (0.498) Mean of dep. var. 0.028 0.321 0.591 0.941 0.249 2.729 Observations 7,175 7,343 7,370 7,410 7,344 7,806 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Placebo treatment group -0.009 0.032 0.026 0.043 0.029 0.019 (0.040) (0.061) (0.051) (0.071) (0.051) (0.045) Mean of dep. var. 0.896 0.355 0.141 0.935 0.355 0.399 Observations 25,509 24,817 24,817 7,678 7,044 24,817 Demographic characteristics Neighborhood characteristics Female Age Urban territory Road density Elevation Tree coverage (19) (20) (21) (22) (23) (24) Placebo treatment group 0.010 0.945 0.016 -9.439 -47.112 -0.030 (0.017) (1.898) (0.051) (7.260) (48.435) (0.020) Mean of dep. var. 0.545 31.25 0.938 19.48 688.7 0.0144 Observations 27,865 27,865 63 63 63 63 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (2), using the locations of major roads that did not contribute to the formation of the boundaries of gang territory as a placebo. We additionally control for dummies for the four quartiles of the census tract size distribution. The unit of observation is a dwelling, household, individual, or census tract, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a dummy for gang territory as well as a linear trend in distance to the placebo boundaries, separately for locations on each side of the placebo boundaries and on each side of the boundaries of gang territory. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the placebo boundaries (separately for each side of the boundaries). 82 TABLE S.IV SOCIOECONOMIC CONDITIONS AFTER EXPOSURE TO GANG CONTROL, SUBSAMPLE OF INDIVIDUALS WHO HAVE ALWAYS LIVED IN SAN SALVADOR Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.047*** 0.026** -0.058** -0.076*** 0.005*** -0.132*** (0.015) (0.010) (0.023) (0.019) (0.002) (0.031) Mean of dep. var. 0.932 0.028 0.934 0.105 0.005 0.178 Observations 72,087 60,675 38,926 38,926 38,926 37,147 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.019** -0.225*** -0.145*** -0.024*** -0.179*** -0.734*** (0.007) (0.044) (0.033) (0.006) (0.037) (0.186) Mean of dep. var. 0.036 0.426 0.683 0.955 0.345 3.048 Observations 36,679 37,328 37,414 37,542 37,292 38,926 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.027*** -0.151*** -0.120*** -0.036*** -0.094*** -0.098*** (0.006) (0.029) (0.028) (0.012) (0.019) (0.020) Mean of dep. var. 0.931 0.445 0.201 0.952 0.374 0.520 Observations 156,627 152,953 152,953 60,675 36,147 152,953 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the subsample of individuals who have always lived in San Salvador. For the dwelling characteristics, none of the observations are excluded because all the dwellings have always been located in San Salvador. For the household characteristics, we limit the sample to those observations for which the head of the household has always lived in San Salvador. All the variables come from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 83 TABLE S.V TWO-DIMENSIONAL REGRESSION DISCONTINUITY IN LATITUDE AND LONGITUDE Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.051*** 0.009* -0.006 -0.076*** 0.004*** -0.141*** (0.007) (0.005) (0.015) (0.008) (0.001) (0.011) Mean of dep. var. 0.932 0.028 0.941 0.108 0.005 0.181 Observations 72,087 60,675 62,169 62,169 62,169 59,776 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.007** -0.256*** -0.175*** -0.024*** -0.199*** -0.806*** (0.002) (0.021) (0.017) (0.003) (0.017) (0.087) Mean of dep. var. 0.033 0.429 0.697 0.952 0.346 3.093 Observations 59,096 60,045 60,168 60,384 60,020 62,169 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.026*** -0.161*** -0.141*** -0.028*** -0.104*** -0.109*** (0.004) (0.012) (0.012) (0.006) (0.009) (0.009) Mean of dep. var. 0.928 0.449 0.208 0.952 0.378 0.522 Observations 208,416 202,935 202,935 60,675 58,293 202,935 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for the variables from the 2007 census, using latitude and longitude as the forcing variables. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in latitude and longitude (demeaned), separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 84 TABLE S.VI EXCLUDING 10% OF THE OBSERVATIONS WITH THE HIGHEST VALUES OF THE 1ST PRINCIPAL COMPONENTS FROM NONGANG AREAS Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.042** 0.023** -0.047** -0.031* 0.005*** -0.064*** (0.016) (0.010) (0.022) (0.017) (0.002) (0.024) Mean of dep. var. 0.929 0.030 0.939 0.081 0.005 0.143 Observations 69,008 57,596 59,569 59,569 59,569 57,176 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.002 -0.165*** -0.116*** -0.018*** -0.124*** -0.500*** (0.006) (0.046) (0.033) (0.006) (0.033) (0.185) Mean of dep. var. 0.028 0.402 0.682 0.950 0.316 2.980 Observations 56,496 57,445 57,568 57,784 57,420 59,569 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.026*** -0.103*** -0.040* -0.032** -0.057*** -0.055*** (0.007) (0.028) (0.022) (0.012) (0.018) (0.019) Mean of dep. var. 0.924 0.421 0.169 0.949 0.359 0.498 Observations 199,162 193,681 193,681 57,596 55,693 193,681 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) after excluding 10% of the observations with the highest levels of the first principal component from nongang areas. For the dwelling characteristics, we use the first principal component of the dwelling characteristics; for the household characteristics—the first principal component of the household characteristics; for the individual characteristics—the first principal component of the individual characteristics. When more than 10% of observations had the first principal component less than or equal to the value of the 10th percentile, we exclude a random subset of observations for which the first principal component is exactly equal to the 10th percentile. The estimates do not depend on which subsample of observations are excluded. In particular, we perform 1,000 iterations of this procedure, and for each variable report the most conservative results, i.e., when they are least significant. All the variables come from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 85 TABLE S.VII HOUSING RENT Log of housing rent Housing rent (1) (2) Gang territory -0.191*** -203.20*** (0.052) (56.33) Number of rooms in the apartment: 2 rooms 0.210*** 19.93 (0.053) (30.79) 3 rooms 0.296*** 87.65** (0.059) (42.09) 4 rooms 0.189** 33.14 (0.070) (73.21) 5 rooms 0.134 2.46 (0.107) (124.27) 6 rooms 0.383*** 330.19** (0.089) (148.86) 7+ rooms 0.365*** 378.31* (0.124) (194.71) Number of bathrooms in the apartment: 2 bathrooms 0.507*** 209.67*** (0.073) (49.22) 3 bathrooms 0.718*** 350.97*** (0.062) (46.61) 4 bathrooms 0.836*** 473.41*** (0.066) (82.91) 5 bathrooms 0.992*** 650.37*** (0.080) (130.00) 6 bathrooms 1.095*** 1,028.51*** (0.113) (213.85) 7+ bathrooms 0.979*** 786.86*** (0.160) (233.44) Square meters 0.140*** 190.59*** (0.018) (22.68) Square meters squared -0.003*** -4.29*** (0.000) (0.61) Rented out by an agency 0.269*** 242.29*** (0.034) (15.55) Mean dep. var 6.731 1,008.81 Observations 1,537 1,537 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) for housing rent requested by landlords, controlling for the characteristics of the apartments that are being rented out. The unit of observation is an apartment. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 86 TABLE S.VIII ESTIMATING THE EFFECTS SEPARATELY FOR MS-13 AND 18TH STREET Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) MS-13 -0.051*** 0.024** -0.058** -0.079*** 0.006*** -0.141*** (0.017) (0.010) (0.025) (0.021) (0.001) (0.031) 18th Street -0.044** 0.027** -0.045** -0.078*** 0.005* -0.126*** (0.017) (0.011) (0.021) (0.022) (0.003) (0.031) Mean of dep. var. 0.932 0.028 0.941 0.108 0.005 0.181 Observations 72,087 60,675 62,169 62,169 62,169 59,776 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) MS-13 -0.015** -0.242*** -0.163*** -0.025*** -0.198*** -0.829*** (0.006) (0.050) (0.034) (0.006) (0.039) (0.194) 18th Street -0.012* -0.187*** -0.119*** -0.019*** -0.159*** -0.615*** (0.006) (0.049) (0.036) (0.006) (0.037) (0.212) Mean of dep. var. 0.033 0.429 0.697 0.952 0.346 3.093 Observations 59,096 60,045 60,168 60,384 60,020 62,169 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) MS-13 -0.036*** -0.179*** -0.145*** -0.036*** -0.102*** -0.119*** (0.007) (0.030) (0.027) (0.012) (0.021) (0.020) 18th Street -0.029*** -0.138*** -0.108*** -0.036** -0.082*** -0.091*** (0.008) (0.031) (0.027) (0.013) (0.021) (0.021) Mean of dep. var. 0.928 0.449 0.208 0.952 0.378 0.522 Observations 208,416 202,935 202,935 60,675 58,293 202,935 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) with the dummy for gang territory replaced with two dummies for areas controlled by MS-13 and areas controlled by 18th Street. All the variables come from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 87 TABLE S.IX EXCLUDING AREAS WITHIN 150 METERS OF THE RIVAL GANG Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Gang territory -0.041*** 0.025** -0.060*** -0.076*** 0.004*** -0.123*** (0.015) (0.010) (0.020) (0.020) (0.001) (0.027) Mean of dep. var. 0.942 0.027 0.939 0.122 0.003 0.206 Observations 60,187 50,742 51,933 51,933 51,933 49,948 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Gang territory -0.012** -0.191*** -0.122*** -0.021*** -0.161*** -0.612*** (0.006) (0.044) (0.031) (0.006) (0.032) (0.192) Mean of dep. var. 0.035 0.475 0.734 0.958 0.383 3.249 Observations 49,271 50,178 50,306 50,480 50,144 51,933 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Gang territory -0.030*** -0.151*** -0.117*** -0.034*** -0.083*** -0.098*** (0.007) (0.028) (0.024) (0.011) (0.018) (0.019) Mean of dep. var. 0.932 0.475 0.231 0.957 0.397 0.540 Observations 174,465 169,910 169,910 50,742 48,619 169,910 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) after excluding gang-controlled neighborhoods that are located within 150 meters of the rival gang. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. All the variable come from the 2007 census. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). 88 TABLE S.X ISLANDS OF GANG TERRITORY Dwelling characteristics Household characteristics Walls made Bare floor Has sewerage Use electricity for No bathroom Has internet of concrete infrastructure lighting and cooking (1) (2) (3) (4) (5) (6) Island of gang territory -0.029** 0.023** -0.084** -0.065*** 0.006*** -0.103*** (0.013) (0.009) (0.038) (0.020) (0.001) (0.030) Rest of gang territory -0.057*** 0.027** -0.028 -0.087*** 0.006* -0.148*** (0.020) (0.010) (0.028) (0.022) (0.003) (0.030) Mean of dep. var. 0.932 0.028 0.941 0.108 0.005 0.181 Observations 72,087 60,675 62,169 62,169 62,169 59,776 Household characteristics Has a motorcycle Has a car Has a phone Has a TV Has a computer Number of rooms (7) (8) (9) (10) (11) (12) Island of gang territory -0.011* -0.216*** -0.130*** -0.018*** -0.167*** -0.709*** (0.006) (0.050) (0.029) (0.005) (0.038) (0.189) Rest of gang territory -0.014** -0.202*** -0.139*** -0.024*** -0.177*** -0.684*** (0.006) (0.048) (0.037) (0.007) (0.037) (0.203) Mean of dep. var. 0.033 0.429 0.697 0.952 0.346 3.093 Observations 59,096 60,045 60,168 60,384 60,020 62,169 Individual characteristics 1st principal component of the: Can read Has a high Has a university Dwelling Household Individual and write school degree degree characteristics characteristics characteristics (13) (14) (15) (16) (17) (18) Island of gang territory -0.040*** -0.194*** -0.148*** -0.026** -0.087*** -0.127*** (0.007) (0.028) (0.025) (0.010) (0.020) (0.019) Rest of gang territory -0.026*** -0.125*** -0.104*** -0.043*** -0.091*** -0.084*** (0.007) (0.033) (0.028) (0.014) (0.020) (0.022) Mean of dep. var. 0.928 0.449 0.208 0.952 0.378 0.522 Observations 208,416 202,935 202,935 60,675 58,293 202,935 Note: *** p<0.01, ** p<0.05, * p<0.1. The table presents the results of estimating Specification (1) with the dummy for gang territory replaced with dummies for the islands of gang territory and for the other gang-controlled locations. All the variables come from the 2007 census. The unit of observation is a dwelling, household, or individual, depending on which characteristics are being considered. In the individual-level regressions, the sample consists of the entire population. Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors in parentheses are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 95 FIGURE S.4.—Socioeconomic Conditions Before the Gangs’ Arrival: Individual Characteristics Note: The figure illustrates the results for the individual characteristics from Table II. All the variables come from the 1992 census. The unit of observation is an individual. All the variables represent the share of individuals that have the outcome variable (can read and write, have a high school degree, etc.). The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. 96 FIGURE S.5.—In-sample Migration Is Not Driving the Results: Education and Firm Size Note: The figure illustrates the results from Appendix Table A.I. The left-hand side of the figure presents the results for the full sample (Panel A of Appendix Table A.I), the right-hand side—for the subsample of individuals who have lived in the same location all their life (Panel B of Appendix Table A.I). The results are very similar. The vertical axis represents the average value of the outcome variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin. GANGS, LABOR MOBILITY, AND DEVELOPMENT 97 FIGURE S.6.—Alternative Bin Size Panel A: 60-Meter Bins Panel B: 20-Meter Bins Note: The figure illustrates the regression discontinuity plots for the 1st principal components of the dwelling, household, and individual characteristics from the 2007 census, using a different bandwidth than in the baseline specification. In Panel A, the dots represent the average value of the outcome variable for 60 meter bins. In Panel B, the dots represent the average value of the outcome variable for 20 meter bins. The unit of observation is a dwelling, a household, and an individual, depending on the specification. All the variables are normalized to vary between zero and one with higher values representing better outcomes. The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. 98 FIGURE S.7.—Housing Rent Note: The figure illustrates the regression discontinuity plots for the residual of housing rent and log housing rent after subtracting the effects of all the controls. The unit of observation is an apartment listing. The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. Omitted controls include dummies for the number of rooms, dummies for the number of bathrooms, a quadratic polynomial in square meters, a dummy for whether the apartment is being rented out by an agency rather than an individual, and a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. FIGURE S.8.—Gang Border Permutations Note: The figure presents the results of a permutation test, illustrating how the regression discontinuity estimates for the three main outcome variables change when the borders of gang territory are shifted in either direction. Each estimate comes from a separate regression specification. The x-axis characterizes the shift in the boundaries of gang territory (in meters). Omitted controls include a linear trend in distance to the boundaries of gang territory, separately for locations on each side of the boundaries. Standard errors are clustered by 30 meter bins, denoting the distance to the boundaries of gang territory (separately for each side of the boundaries). GANGS, LABOR MOBILITY, AND DEVELOPMENT 99 FIGURE S.9.—Availability and Quality of Public Goods Note: The top part of the figure presents the regression discontinuity plots for the number of hospitals and schools per square kilometer; the lower part of the figure presents the regression discontinuity plots for the questions about satisfaction with the availability and quality of public goods from the 2019 survey. In the top part of the figure, the unit of observation is a 10 meter bin, denoting distance to the boundaries of gang territory. In the lower part of the figure, the unit of observation is an individual. For the questions about satisfaction with the availability and quality of public goods, the respondents were asked to rate the availability and quality of public goods on a scale from 1 (extremely unsatisfied) to 7 (extremely satisfied). The vertical axis represents the average value of the outcomes variable; the horizontal axis—distance (in meters) to the boundaries of gang territory. Neighborhoods to the left of the dashed line are located outside of gang territory; areas to the right are controlled by the gangs. The dots represent the average value of the outcome variable in that 30 meter bin.