1. Mapping criminal groups

Mexico's criminal conflict has grown increasingly violent and complex over the past few decades (Trejo and Ley, Reference Trejo and Ley2018, Reference Trejo and Ley2020). I refer to the growing number of criminal groups as “criminal fragmentation,” though it reflects both the territorial expansion of existing organizations and the emergence of new ones. I define criminal groups as at least loosely organized, semi-independent, non-state collectives that engage in illicit activities. Such groups differ considerably. Cartels may be hierarchical or adopt a “wheel network” (Farfán-Méndez, Reference Farfán-Méndez2019). Smaller groups may form a spoke of this wheel or operate independently. Some extend nationwide, while others are local. They may participate in the international drug trade or focus on extortion (Garzón Vergara, Reference Garzón Vergara, Bagley, Rosen and Kassab2015). While groups are sometimes entangled with state structures, here I focus on non-state criminal groups (Snyder and Durán Martínez, Reference Snyder and Durán Martínez2009; Barnes, Reference Barnes2017; Yashar, Reference Yashar2018; Trejo and Ley, Reference Trejo and Ley2020).

I develop original data on criminal organizations by hand-coding a narcoblog, an anonymous citizen journalism website focused on Mexico's drug war. Narcoblogs act as both press aggregators and alternative news sources, offering several benefits. By filtering relevant press, one can be realistically hand-coded. This is useful for identifying criminal actors, especially when names are common in other contexts (e.g., “the Taliban”). Narcoblogs also publish citizen reports, like crime scene photos or narco-banners. Such content is informative, but may go unreported: in 2011 some outlets stopped publishing criminal “propaganda” (Acuerdo para la Cobertura Informativa de la Violencia, 2011). Given anonymity, narcoblogs reduce concerns about underreporting due to violence against journalists (Larreguy et al., Reference Larreguy, Lucas, Marshall and Riaz2020; Dorff et al., Reference Dorff, Henry and Ley2022).

I use the narcoblog Borderland Beat, the longest-running and most consistently active of popular blogs. In addition to coverage, this helps differentiate increased posting from a rise in groups. It also prides itself on verification of information.Footnote ¹ Coverage begins in 2009 and ends in October 2020, when a lawsuit by a convicted criminal leader temporarily shuttered the site. Founded by editor “‘Buggs,”Footnote ² Borderland relies on anonymous volunteer contributors in the United States and Mexico to write, translate, and validate information (Marentes, Reference Marentes2019). Content primarily comes from the press, supplemented with user forums, social media, and submitted emails (Appendix A).

To code criminal groups, I focus on organizations with a unique name. Theoretically, such groups operate at least semi-independently; empirically, this ensures groups can be tracked over time and provides clear guidelines for inclusion. For each post, I identify organizations and their municipalities of operation. I supplement with press information, including whether groups operate under multiple names and their relationships to other organizations (Appendix A).

I identify 473 groups operating between June 2009 and October 2020. Following Sobrino (Reference Sobrino2020), I define nine as major cartels, like the Sinaloa Cartel and BLO. Such groups differ in structure and methods, but represent the most dominant and internationally active organizations in my study period (Durán-Martínez, Reference Durán-Martínez2015, Reference Durán-Martínez2017; Farfán-Méndez, Reference Farfán-Méndez2019; Congressional Research Service, 2022). Twenty-three groups cannot be tied to a municipality, dropping from the data (see Appendix A for details).

Other groups I split into three categories. Cells are affiliates of larger cartels, often local wings or enforcers. Tracking them offers insight into the adoption of more horizontal leadership structures through decentralization (Garzón Vergara, Reference Garzón Vergara, Bagley, Rosen and Kassab2015; Farfán-Méndez, Reference Farfán-Méndez2019). By design cells operate with some independence. For example, La Linea is an enforcer unit for the Juarez Cartel, but grew in power as the larger cartel weakened. Cells also contribute to violence, including by fighting one another: La Gente Nueva and Los Emes warred despite being Sinaloa affiliates (Corcoran, Reference Corcoran2011; Calderón et al., Reference Calderón, Robles, Díaz-Cayeros and Magaloni2015). I identify 177 cells (see Appendix A for more on coding decisions).

Ninety-three groups are splinters that broke with their former compatriots. Unlike cells, they do not operate under a higher authority. They may emerge from cartel breakdown, as with BLO's disintegration. Some form after wars of succession. Cells may become splinters: in Tijuana, the dominant Jalisco New Generation Cartel (CJNG) went to war with its former unit Los Cabos (La guerra interna del CJNG, 2019). I differentiate cells from splinters based on press descriptions, with the caveat that internal group dynamics are difficult to observe. If a group was a cell that split, I code it as a splinter.

Another 171 have no known past or present affiliation to larger organizations. I call these “unaffiliated groups,” though this is a category of exclusion: we may simply lack information. Many are local, like the Mexico City-based Los Tóxicos robbery gang. Some become significant regional players. The Santa Rosa de Lima Cartel, dedicated to fuel theft, in some areas challenged the CJNG. Despite flaws, mapping unaffiliated groups offers insight into organizations that lack large cartels’ resources.

One limitation is that categories are fixed, due to the difficulty of documenting group dynamics. Cells may begin as unaffiliated, for example, but identifying if and when this changes is challenging. Here I combine cells, splinters, and unaffiliated organizations under the category minor groups. Theoretically, they closely resemble what Garzón Vergara (Reference Garzón Vergara, Bagley, Rosen and Kassab2015) call “predatory micro-networks” rather than hierarchical cartels. Empirically, grouping them skirts the challenge of differentiation. Appendix B shows results by subtype and with alternative coding rules.

Among minor cartels, I differentiate between when new organizations emerge and existing organizations expand.Footnote ³ I define a group as emerging when it first appears in the narcoblog dataset (excluding 2009, when all groups are “new”). An organization expands by reaching a new municipality after its first year operating. While imperfect, these proxies provide suggestive evidence for mechanisms underlying fragmentation. Figure 1 shows criminal group presence in 2010 and 2019, the first and last full years of data (to make the descriptive comparison cleaner). Maps demonstrate an increase in criminal groups and the spread of their territory.

Figure 1. Criminal Groups, 2010 and 2019. Criminal groups by year in 2010 and 2019, the first and last full years of data (the full data cover mid-2009 to October 2020).

Data contribute to literature mapping Mexico's criminal organizations (Appendix A). Recent projects use machine learning and news corpuses to track cartels, both large (Coscia and Rios, Reference Coscia and Rios2012; Sobrino, Reference Sobrino2020) and small (Osorio and Beltran, Reference Osorio and Beltran2020). Hand-coding identifies more groups—about three times the number in Osorio and Beltran (Reference Osorio and Beltran2020)—but does not match breadth of coverage.Footnote ⁴ Other datasets use press or government sources, though with a more limited timeframe (Lantia Intelligence, 2022; BACRIM, 2022; Aguilar et al., Reference Aguilar, Guazo, Mezal and Garnica2024). I extend this work by focusing on tracking small groups and drawing on a new source.

Like press sources generally, the data likely bias toward groups that seek territorial control and use violence, and toward municipalities with stronger media presence (Anders, Reference Anders2020). Narcoblogs also do not follow standard reporting practices. Most concerning, some are denounced as mouthpieces for state agents or criminal groups. Using reputable Borderland Beat reduces these concerns (International Crisis Group, 2024). More broadly, however, low-quality information may inflate group counts.Footnote ⁵ Since there is no ground-truth data, Appendix B benchmarks my data against other sources, including existing datasets (Osorio and Beltran, Reference Osorio and Beltran2020; Sobrino, Reference Sobrino2020), press searches, a Spanish-language narcoblog, and leaked military intelligence (Aguilar et al., Reference Aguilar, Guazo, Mezal and Garnica2024). My results increase confidence that the data are not significantly biased or low quality.Footnote ⁶

2. Understanding fragmentation

What drives fragmentation? I provide evidence that the kingpin strategy contributed to the devolution of large cartels and that, in turn, profit opportunities compelled groups to seek resources in new areas (Castillo and Kronick, Reference Castillo and Kronick2020; Trejo and Ley, Reference Trejo and Ley2020; Alcocer, Reference Alcocer2022). While in line with past literature, the narcoblog dataset allows testing how such mechanisms influence fragmentation.

2.1 The kingpin strategy

To explore leadership decapitation and criminal fragmentation, I combine US sanctions lists with the narcoblog data, aggregated to the municipality level. While sanctions lists capture groups with international presence—meaning they are far from complete—they include the highest-profile kingpins. My main specifications use two-way fixed effects to estimate change in criminal groups following kingpin removals, controlling for state-specific time trends; whether a municipality has a conservative mayor (Magar, Reference Magar2018); and log hectares of poppy and marijuana seized, a proxy for drug presence (Ch, Reference Ch2023). A municipality is “treated” if a group operated there in the year of or year prior to a kingpin removal. Municipalities remain “treated” for three years (see Appendix A for details).

The number of criminal groups increases in affected territories following kingpin removals (Table 1; Figure 2, left panel). For minor groups this is due to both emergence and expansion (Figure 2, right panel). Appendix A shows these results are primarily due to increased splinters and cells, suggesting decentralization of large cartels. The rise in major cartels points to intergroup competition after a power vacuum.Footnote ⁷ Appendix B presents robustness checks, including with alternate independent variables and estimators. Main results largely hold.

Table 1. Kingpin removals and criminal groups

Robust SEs clustered by municipality. Controls for mayoral party, poppy eradication, and marijuana eradication, plus state-year linear time trends.

***p < 0.001, **p < 0.01, *p < 0.05, $^ + {\rm p}< 0.01$.

Figure 2. Kingpin removals and criminal groups. Event study plots showing major and minor groups (left) and the emergence or expansion of minor groups (right). The independent variable is the number of years to or from a kingpin capture.

While following a difference-in-differences design, these results are not causal: successful kingpin removals may reflect weakened cartels. Indeed, major groups increase before attacks. However, qualitative evidence suggests some relationship between the kingpin strategy and fragmentation. Take the state of Sonora, once Sinaloa Cartel territory. Leader El Chapo's arrest and extradition led to a succession conflict between his lieutenant and his sons, backed by cells Los Cazadores and Los Salazar. In the power vacuum, the CJNG entered, as did the Caborca Cartel, via its armed wing La Barredora 24/7 (Dittmar, Reference Dittmar2022).

2.2 Fuel theft and fragmentation

Increased competition and drug interdiction encourage criminal groups to diversify (Herrera and Martinez-Alvarez, Reference Herrera and Martinez-Alvarez2022). One profitable alternative to drugs is fuel theft (huachicoleo), the stealing of gas to sell at below-market rates (often by tapping pipelines). Huachicoleo increased after Mexico began removing fuel subsidies in 2010, following losses during the 2008 financial crisis. Full deregulation occurred in 2017, with prices jumping as much as 20 percent overnight (Appendix A) (Planta and Jordan, Reference Planta and Jordan2013; Un 20 de aumento a cualquiera le pega, 2017). Cartels thus had “even more incentive to undercut the prices of the legal market” (Jones and Sullivan, Reference Jones and Sullivan2019, p. 7). Exploiting plausibly exogenous pipeline placement, I use a two-way fixed effects design to measure the relationship between the value of territory and armed group presence.Footnote ⁸ As a proxy for value, my main independent variable interacts domestic gas prices (in inflation-adjusted pesos, divided by ten for visualization) and logged pipeline length by municipality (in km).

Table 2 suggests that areas with more pipeline attracted more criminal groups as gas prices rose. This is driven mostly by expansion—only unaffiliated groups appear somewhat more likely to emerge (Appendix A). Findings point to profit opportunities encouraging groups to expand their territories, in line with work using other data sources (Alcocer, Reference Alcocer2022; Franco-Vivanco et al., Reference Franco-Vivanco, Martinez-Alvarez and Martínez2023). Robustness checks, including alternative independent variables, are given in Appendix B.

Table 2. Gas pipelines and criminal groups

Robust SEs clustered by municipality. Controls for mayoral party, poppy eradication, and marijuana eradication, plus state-year linear time trends.

***p < 0.001, **p < 0.01, *p < 0.05, $^ + {\rm p}< 0.01$.

Qualitative evidence illuminates how the decentralization and splintering of large cartels contributed to territorial expansion. The dominant Gulf Cartel and Zetas moved into fuel theft in 2010, as subsidies decreased. Following leadership removals, their splinters—like Sangre Nueva Zeta and Grupo Sombra—increasingly engaged in huachicoleo.