How Does Leadership Decapitation Affect Violence? The Case of Drug Trafficking Organizations in Mexico Brian J. Phillips, Center for Research and Teaching in Economics (CIDE)

Many governments target leaders of violent groups, but consequences of this strategy are unclear. Additionally, most studies examine political groups such as terrorists, ignoring criminal organizations—even though they can represent serious threats to security. This article presents a theoretical framework for how political and criminal groups differ and uses the framework to explain how group type should condition leadership removal’s effects. Decapitation should weaken criminal organizations, temporarily reducing violence. However, as groups fragment and newer groups emerge to address market demands, violence increases in the longer term. Empirical analysis using original data on Mexican criminal organizations generally supports the argument. Interestingly, the short-term violence reduction is only associated with leaders arrested (not killed) and when the target is a midlevel leader as opposed to the top leader. These results differ markedly from those found in studies of political groups.1

I

n 1992, Peruvian counterterrorism agents captured Abimael Guzmán, leader of Sendero Luminoso, largely halting its attacks—and Peruvian political violence generally. The following year, Colombian forces killed Pablo Escobar, shutting down his Medellín Cartel and reducing violence. However, organized crime violence in the area continues, carried out by newer groups. Meanwhile in Mexico, the government has used a similar strategy against drug trafficking organizations (DTOs) for years, but violence connected to the groups continues to kill thousands of people annually. Does leadership removal reduce violence?2 How does the context in which it is used condition its effects? Many governments kill or capture high-level members of violent organizations to disrupt the groups’ activity, but it is unclear if this is generally successful. A puzzle exists in studies of leadership removal. Analyses of relatively political groups, such as terrorists, increasingly find that it disrupts targeted groups and reduces violence (Johnston 2012; Price 2012).

However, the few studies of criminal groups suggest that the policy might be counterproductive (Dickenson 2014; Guerrero Gutiérrez 2011a). Why the difference? This article presents a theoretical contribution by suggesting distinctions between political and criminal groups and uses this framework to explain how leadership removal should affect criminal violence in particular. The focus on organized crime is pertinent for political scientists because research in Comparative Politics and International Relations finds important connections between crime and politics. Criminal organizations can threaten international security (Andreas 2004; Williams 1994), the economy (Daniele and Marani 2011), and governance (Naím 2012; Schedler 2014), and criminal violence affects political participation and attitudes (Bateson 2012; Carreras 2013).3 Given the connections between crime and politics, to what extent can research on politically violent groups help us understand criminal organizations? Perhaps surprisingly, the

Brian J. Phillips is an assistant professor at the Center for Research and Teaching in Economics (CIDE), Mexico City, Mexico. 1. Supplementary material for this article is available at the “Supplements” link in the online edition and at http://sites.google.com/site/brianjphillips. Replication data will be available at http://sites.google.com/site/brianjphillips. 2. This article uses the terms “leadership removal” and “decapitation” interchangeably. 3. Beyond crime affecting politics, interesting new research shows, for example, how US policies affect violent crime in Mexico (Dube, Dube, and García-Ponce 2013). The Journal of Politics, volume 77, number 2. Published online February 17, 2015. http://dx.doi.org/10.1086/680209 q 2015 by the Southern Political Science Association. All rights reserved. 0022-3816/2015/7702-0002 $10.00

324

Volume 77

literature has barely engaged this question. Scholars of political violence increasingly use the organizational level of analysis (Asal and Rethemeyer 2008; Pearlman and Cunningham 2012; Shapiro 2013), but it is unclear how research on terrorist groups, for example, can be applied to organized crime. The next section discusses the role of violence in organized crime. The literature on leadership removal is discussed— most of which analyzes political groups. I then present an argument for how criminal groups differ from more political groups and how leadership removal should affect criminal groups in unique ways. The argument uses the logic of incentives and the notion that criminal and political groups serve different “markets.” (In the online appendix, case studies of the Sinaloa Cartel and the FARC show how leadership removal affected these different types of groups in distinct ways.) The empirical section presents original data on Mexican DTO leadership removal. As the argument suggests, decapitation is associated with decreased violence in the short term. In the longer term, it is associated with increased violence. Additional tests suggest the short-term drop in violence only occurs after an arrest, instead of a kill. Furthermore, the short-term reduction in violence only occurs when a midlevel leader, as opposed to the top leader, is targeted. These results are strikingly different from those found in studies of leadership removal on political groups such as terrorists.

SHOULD LEADERSHIP REMOVAL AFFECT ORGANIZED CRIME VIOLENCE? Why might the arrest or killing of a leader of a criminal group affect violence? Before assessing the specific mechanisms of how leadership loss affects groups, it is helpful to think about why criminal organizations use violence in the first place. Violence is such a fundamental part of the identity of criminal groups that some scholars refer to these organizations as “violent entrepreneurs” (Gambetta 1993; Volkov 2002). One of the “indispensable” qualities of a member of a criminal organization is the “proven ability to use violence” (Varese 2011, 18). Empirically, areas with an organized crime presence have higher levels of violence than areas without such a presence, even when taking other factors into consideration (Marselli and Vannini 1997). One reason criminal groups use violence is related to market competition (Gambetta 1993, 40). Interorganizational violence emerges when groups compete for territory, and each attempts to signal that is in possession of the territory and/ or willing to fight for it. When relationships among criminal groups are stable, there is less violence. However, equilibria are often upset, and “elimination contests” occur (Volkov

Number 2

2015 / 325

2002). Some examples are the United States in the 1930s and Italy in the 1960s and 1980s. Another reason for organized crime violence is related to the environment of illegality in which these groups operate. Actors in the illegal market cannot turn to the state for protection, so a black market for protection emerges, and groups want to show that they are capable guarantors (Gambetta 1993, 42–43). Related to this, criminal groups, without normal legal recourses, use violence to enforce contracts. While violence is essential for criminal groups, excessive violence can be costly. Violent enforcement or signaling activities are literally costly, as they require resources and can result in the death or imprisonment of perpetrators. Criminal groups try to regulate violence for these reasons (Gambetta 1993, 114). Organized crime violence happens for a variety of reasons, and groups try to use it efficiently, but in general violence is a necessary and regular occurrence in the presence of criminal organizations. The causes of violence by criminal groups—such as territorial competition and contract enforcement—are different from the motivations of violence used by political groups. Kydd and Walter (2006) identify five reasons why terrorism is used: attrition, intimidation, provoking a harsh government response, spoiling negotiations, and outbidding rival groups for public support. Criminal groups use intimidation, and possibly attrition, but they generally do not use violence for the other reasons, which are inherently political. There are additional reasons that political actors use violence, such as advertising their cause (e.g., Hoffman 2006, 198). Religious terrorist groups can use violence as an end in itself, killing nonadherents to please a supernatural audience as part of a “cosmic war” (Juergensmeyer 1991). These types of violence are not consistent with the goals of criminal groups, which generally want to be left alone by the state. These contrasting motivations for using violence are important for understanding the distinct consequences decapitation has for criminal and political groups.

Leadership Removal Often Reduces Violence Given the important role that violence plays for criminal groups, and the obvious policy significance of this behavior, can leadership removal curtail the bloodshed? Research increasingly shows that this is an effective tactic for reducing the violence of terrorist and guerrilla groups. Because there has been so little systematic research on consequences of leadership targeting regarding criminal groups, this section considers studies of terrorist and guerrilla groups, as well as criminals, and the following section presents an argument for why group type should condition the effects of this policy.

326 / How Does Leadership Decapitation Affect Violence? Brian J. Phillips

Governments target dissident or criminal group leaders because they expect the policy to bring about desired consequences. They hope it will either cause the group to entirely fail or at least diminish its ability to carry out operations. A fundamental assumption behind the notion that leadership removal can work is the importance of leadership to an organization. Leaders can inspire members, coordinate action, and distribute the selective incentives crucial to recruitment and membership retention (Olson 1965; Oots 1989, 141). In some cases, leadership removal can cause immediate collapse of the group. However, any forced leadership change is likely to lead to uncertainty—halting momentum, and disrupting business. Removing a leader can slow groups down because of a number of issues related to leadership transition, as Gambetta notes regarding criminal organizations (Gambetta 1993, 63, 115). Operations might need to be put on hold while a new leader is selected. In some cases, the leader is so important that the group disintegrates after an arrest or death. In other cases, a faction is unhappy with the new leadership and breaks off—making the rump group less powerful than it had been. Additionally, there is likely a learning curve as the new leader learns the position. Leadership removal’s violence-reducing effects find substantial support in the literature—in both case studies and quantitative analysis—of political groups. Organizational decapitation is said to have ended or crippled many groups, including, arguably, Turkey’s Kurdistan Workers’ Party, the Real Irish Republican Army, Japan’s Aum Shinrikyo, Italy’s Red Brigades, and France’s Action Directe (Cronin 2009, 18, 31; Johnston 2012). In-depth analyses find that leadership decapitation produced sustained decreases in attacks when used against an animal rights group in the United Kingdom (Donovan and Coupe 2013) and the FARC in Colombia (Morehouse 2014). Quantitative studies, examining hundreds of cases of leadership removal, have also found that it generally reduces violence. Price’s (2012) analysis suggests that the policy increases the chance of terrorist groups ending. Johnston (2012) finds that insurgencies are more likely to end, and insurgent attacks are likely to decrease, after a successful leadership hit. His analysis is innovative because he compares attempts with successful removals, and only finds violencereducing effects after the successful removals. These studies suggest that leadership removal generally diminishes violence over the long term—at least for political groups.

When Does Leadership Removal Provoke Increased Violence? In spite of the above, there are theoretical reasons that leadership removal might have negative unintended conse-

quences. There are two primary arguments for why it might be counterproductive. The first involves so-called martyrdom effects, while the second is related to group fragmentation. Martyrdom effects occur because after leadership removal, members might desire vengeance and be spurred to act (Jordan 2009, 755). State actions against a group can also inspire public sympathy, especially for political groups, which could facilitate martyrdom operations. The fact that state actions can inspire public sympathy is why political groups often use violence to provoke a harsh government response—again, principally an action of politically motivated groups. Group fragmentation is also a potential consequence of leadership removal and can lead to new dynamics of violence. When a leader is removed, would-be successors might try to demonstrate their power, including by eliminating rivals. New groups can splinter off, leading to intergroup violence. This appears in some cases of criminal group leadership removal in Mexico and is part of the reason some analysts suggest this “kingpin” strategy has increased violence (e.g., Grillo 2011, 239; Guerrero Gutiérrez 2011a). Even if the group does not fully fragment into new groups, intragroup fighting related to potential fragmentation can cause violence as well. While fragmentation is generally theorized to lead to violence, it could have time-dependent effects. There could be a decrease in violence as fragmentation begins, as this is consistent with a disrupted group. With time, however, the newly created groups resulting from fragmentation could begin to fight, suggesting a longer-term increase in violence. In spite of theoretical reasons why leadership removal might lead to more violence, evidence suggests this only happens with some particular types of groups (e.g., Rowlands and Kilberg 2011). Jordan (2014) argues terrorist groups that are bureaucratically organized and have strong communal support are likely to survive leadership removal and retaliate. There is some evidence that religious terrorist groups increase violence after leaders are targeted (Mannes 2008). Finally, two studies find removal of criminal group leaders is associated with upticks in violence (Dickenson 2014; Guerrero Gutiérrez 2011a).4 Given that leadership removal seems to have divergent effects depending on aspects of the targeted organization, this likely explains some of the mixed results in the litera-

4. The current article differs from these papers in important ways. First, the theoretical approach of this article is new, comparing political and criminal groups, and determining how differences should condition effects of policies used against them. Second, empirically, the present article controls for more alternative explanations and covers a longer temporal sample.

Volume 77

ture. Studies have generally pooled together various group types (e.g., ethnic and religious), only to find that results differ across subsamples. Effects also likely depend on whether the group is primarily motivated by political or criminal goals, but this has not yet been discussed in the literature.

HOW CRIMINAL GROUPS DIFFER FROM OTHER LEADERSHIP REMOVAL TARGETS The majority of the research on leadership removal has focused on insurgent or terrorist organizations—groups with political goals. Can these studies be applied to criminal organizations such as those in Mexico? Illicit groups such as DTOs have a great deal in common with other violent nonstate actors, so research findings are often applicable, but differences must be taken into consideration.5 A number of studies illustrate overlaps and contrasts between political and criminal groups (Flanigan 2012; Makarenko 2004; Williams 2009, 2012). Criminal organizations are fundamentally motivated by a desire to make money illegally. Insurgents and terrorists differ in that they have political goals—national liberation or policy change, for example. The primary relationship criminals have with government is a desire to be left alone by it. It is motives, not methods, that separate criminal groups from terrorists and insurgents, to paraphrase Shelley and Picarelli (2002). There are many interesting examples of criminal groups with some political dimensions and political groups using criminal means to raise funds. However, the general consensus in the literature is that groups can generally be identified as either criminal or political (insurgent, terrorist, etc.) based on primary organizational goals.6 Building on research examining criminal and political organizations, I argue that DTOs differ from more political organizations in two fundamental ways. First, the incentives that criminal groups offer their members are more financial than other types of incentives. Second, criminal group activity is explained largely by market forces, as opposed to ideological or ideational conditions. These differences can help explain why leadership removal should affect criminal groups differently from other groups. Incentives explain shorter-term effects, while market forces can explain longer-term consequences. While the current

5. This is similar to the assertion that leadership removal could have unique consequences for terrorist groups as opposed to guerrillas (e.g., Frankel 2011, 26). 6. There are instances of “convergence” (Makarenko 2004) where criminal and political groups change categories or blur boundaries. However, Flanigan (2012) and Williams (2012) discuss Mexican DTOs directly and argue that they are neither terrorist nor insurgent organizations.

Number 2

2015 / 327

study only empirically examines criminal groups to show that leadership removal has divergent effects on these groups compared to the data we already have on political groups, the online appendix includes an analysis of how decapitation affected the Sinaloa Cartel and the FARC in distinct ways. Table 1 summarizes how political and criminal groups differ according to incentives and markets. Note that these are ideal types, and in reality there are overlaps. For example, many political groups use material incentives as well (Shapiro 2013). However, criminal groups are more likely to depend on such incentives for their survival as well as their day-to-day operations.

Incentives: Differentiating Group Types—and Why Leaders Matter Incentives are important because organizations use them to keep members involved and recruit new members, as is suggested by foundational research on group behavior (Olson 1965; Wilson 1973). Two types of incentives are particularly relevant: material incentives and purposive incentives (Wilson 1973, 33–34). Material incentives are basically financial and other tangible benefits. Purposive incentives are “intangible rewards that derive from the sense of satisfaction of having contributed to the attainment of a worthwhile cause” (Wilson 1973, 34). Criminal organizations primarily depend on material incentives; they are businesses.7 Wilson argues that businesses are the ideal-typical group relying on material incentives (Wilson 1973, 36). Relatively political groups such as terrorists, however, can draw many recruits solely based on their interest in supporting a particular cause.8 Incentives are important for understanding leadership removal consequences, in part because removal can disrupt the provision of incentives to members. Leaders are crucial for most groups, as discussed above, because of their role in allocating incentives to members. The impact on political groups, motivated by purposive incentives, is unclear: On the one hand, leaders can be helpful to political group mobilization, such as by reiterating the urgency of the struggle (Cronin 2009, 15). If a political group leader is especially

7. Hoffman argues that criminals are motivated by personal gain, but a “terrorist is fundamentally an altruist,” attempting to achieve a greater good. (Hoffman 2006, 37; italics in original). 8. Many guerrilla and terrorist groups pay their members. While this is helpful for mobilization of these political groups, it is more necessary for the mobilization of criminal groups. Furthermore, political groups depending on material incentives might draw lower-quality members (Weinstein 2007). Other arguments suggest that the terrorist groups able to retain nonfree-riding members through material incentives are highly organized and rare (Berman 2009).

328 / How Does Leadership Decapitation Affect Violence? Brian J. Phillips

Table 1. Differences between Criminal and Political Groups Political Organizations Incentives for members “Market”

Purposive incentives Ideas and public opinion

Criminal Organizations Material incentives Illicit goods market

charismatic (as is often the case), leadership removal could be fatal for the group. On the other hand, members of political groups should in many cases still be inspired by the “cause,” regardless of a leadership loss or a rocky management transition. Overall, the use of the purposive incentive framework suggests that political groups can be affected in divergent ways by leadership removal, depending on factors such as whether the group is religious. This is consistent with research on terrorist groups (Jordan 2009, 2014). In the short term, at least, effects for political groups are unclear. Groups primarily motivated by material incentives, criminal organizations, seem more likely to be consistently affected by the loss of a leader in the short term. Political groups are often voluntary, but business leaders have authority to hire, fire, promote, resolve intragroup disputes, and make other impactful decisions (Wilson 1973, 215). Criminal organizations have memberships dependent on material incentives, which are coordinated by leadership. Unexpected changes in leaders are likely to affect the flow of paychecks, and this should disrupt business and possibly encourage members to seek employment elsewhere. A disrupted group should produce less violence. This challenges arguments that leadership removal can strengthen targeted groups (Jordan 2009) or has no effect on them (e.g., Mannes 2008), and research on criminal groups suggesting the policy is only associated with more violence (Dickenson 2014). Note that the disruption of criminal-group activity could only occur in the short term. Leadership removal is likely to affect business, reducing activity, including the violence deemed necessary for contract enforcement, reputation, and other goals. However, given the lucrative material incentives associated with organized crime, and a general lack of substitutable sources of incentives, participants are unlikely to leave the business altogether. Longer-term effects for criminal organizations are better understood by different dynamics, discussed below. Disruptions to the distribution of material incentives likely explain shorter-term decreases in violence of criminal groups.

Hypothesis 1: Leadership removal should disrupt criminal group activity, leading to less violence in the short term.

The “Market” Determines Broader Effects Regarding the “market” that drives organizational behavior, as indicated in Table 1, criminal and political groups differ substantially in this regard as well. This should affect longer-term consequences of leadership removal. Relatively political groups depend on ideas and public opinion, whether they are guerrillas working among the people or terrorists targeting the public to pressure the government. They compete with other groups within a marketplace of ideas, trying to offer new visions for the world (Della Porta 2013, 205–08). This is related to the notion that political groups vie for “market share” of members and support from a wider community and as a result are responsive to public opinion. DTOs, on the other hand, exist because there is a demand for illicit goods, and these organizations are able to make a handsome profit by manufacturing product or delivering it to customers. DTOs compete for market share of these illegal products. These dissimilar raisons d’etre suggest possible differences in longer-term outcomes after a government hit on group leadership. Organizational decapitation of a political group could have serious long-term consequences if the act affects group members’ and the public’s view of the group. One goal of decapitation is to “break the back” of an insurgency or terrorism campaign. The purpose is to discredit the group to the public, taking away its momentum. This should reduce violence because, as discussed earlier, terrorists and guerrillas use violence mostly for political reasons. If such a group loses hope in achieving its political goals, or loses the ability to mobilize for related reasons, it makes sense that its violence is likely to diminish. When the Peruvian government arrested Sendero Luminoso’s Guzmán, it showed video of him in a cage, in a striped jail uniform, asking his followers to lay down their arms. This was apparently devastating to the organization—demoralizing its armed members and its supporters in the public. The “betrayal, in the eyes of his followers, devastated the movement that had been built entirely around his philosophy and charisma” (Langdon, Sarapu, and Wells 2004, 70). The way the arrest was presented affected the group’s popularity and perhaps the wider idea of the feasibility of the Sendero movement. Within the marketplace of ideas, violent leftism no longer seemed tenable. No other group replaced Sendero Luminoso. Guzmán’s arrest provides an interesting example of how public opinion can condition the effect of leadership re-

Volume 77

moval on political groups. Similarly, Cronin’s analysis of case studies finds that “killing the leader of a group that has widespread popular support either has no measurable effect or is counterproductive” (2009, 14; italics added). Jordan (2014) makes a similar argument. Apparently, most terrorist groups are not popular enough to be resistant to leadership removal; this explains why it usually reduces political violence (e.g., Johnston 2012). However, the role of popular support suggests a key difference between political and criminal groups. Criminal groups serve a literal market. As long as there is a demand for a product and that product is illegal, there will be a function for organizations willing to produce and distribute it (e.g., Varese 2011). In the process of their business, these groups use violence. Unfortunately for governments seeking to thwart these groups, the demand for illegal drugs does not seem to be declining, and this is likely to condition the effects of leadership removal of criminal groups. When a government removes a high-level member of a criminal group, it is likely that the group stumbles somewhat (Hypothesis 1) and could possibly collapse. However, because of the broader market, the lifeblood that feeds such groups, even if the targeted group is destroyed, it is likely that another group will take over the newly opened territory (e.g., Grillo 2011, 239). It often takes time before fractures in the group lead to violence or before new groups form, but the result is usually more violence. As mentioned, leadership removal ended the Medellín Cartel, but other groups flourish in the same area today. The Mexican government has destroyed some DTOs, but it acknowledges its strategy has created many smaller DTOs.9 An ever-present illegal product market has two specific implications for security relating to DTO decapitation. First, the unintended consequence of violence related to group fragmentation seems likely. Fragmentation should in many instances lead to violence because of the high stakes of market control. There is a great deal of money to be made through the black market, so if members of a group see a chance to split off for more profit, they are likely to do it. But the rump group, along with other competitors, is unlikely to cede market share peacefully.10 Second, leadership removal on its own is unlikely to be an effective strategy against DTOs, unless the government

9. Fausset, Richard, “Peña Nieto Team Decries Past Drug Cartel Strategy—And Keeps It,” Los Angeles Times, December 21, 2012. 10. Fragmentation occurs with political groups, but it is not as likely to be as violent. First, political groups worry about damaging the wider “cause” and often try to minimize intergroup violence (Morrison 2013,

Number 2

2015 / 329

wants to continuously decapitate each new organization as it arises. Leadership removal can break the momentum of political groups, and their related public support, but defeated criminal groups will probably just be replaced by other criminal groups willing to fight for the market. Hypothesis 2: Regardless of short-term effects on violence after a criminal group leadership removal, the longer-term effect will be a return to status quo ante or an increase in violence.

EMPIRICAL ANALYSIS To empirically evaluate the consequences of leadership removal on the violence of criminal groups, I gathered data on violence associated with Mexican criminal organizations between 2006 and 2012 and information on the 32 federal entities in which they may operate. The unit of analysis is Mexican state-month. The estimation technique is discussed in detail below, but it is a time-series cross-sectional negative binomial regression with state-level fixed effects, spatial lags, and other controls. The dependent variable is a count of the number of homicides linked to organized crime for the particular statemonth being analyzed. The primary data source is the Mexican government, which released data for each state from December 2006 until September 2011.11 These data count what the government refers to as “deaths from alleged criminal rivalry” (“fallecimientos por presunta rivalided delincuencial”). The variable ranges from 0 to 452, with a mean of 26. The state-month with the most homicides is Chihuahua, August 2010. A secondary measure of drug-related deaths, used for robustness checks but not reported in the article, comes from the Mexico City newspaper Reforma. The data are organized and stored by the Trans-Border Institute (TBI) at the University of San Diego (Rios and Shirk 2011).12 TBI data cover from November 2007 to November 2012. Those data and the government measure are highly correlated,

160). Second, decapitation can “break the back” or reduce momentum of a political movement, reducing likelihood that midlevel leaders will fight to lead a decapitated terrorist group or that other groups will fight for a decapitated group’s niche. 11. Data from December 2006 to December 2010 were hosted on the Mexican President’s website, but access was taken down after the change of presidency at the end of 2012. Data from January to September 2011 are online at http://www.pgr.gob.mx/temas%20relevantes/estadistica/estadisticas .asp as of December 6, 2013. 12. The data are online at https://justiceinmexico.org/data-portal /homicides/.

330 / How Does Leadership Decapitation Affect Violence? Brian J. Phillips

but the TBI data are consistently lower than the government data. The TBI data could be lower because of incomplete media reporting related to threats from DTOs. Regardless, because of the apparently more complete coverage of government data, I use government data as the primary measure. When models are run with TBI data as a robustness check, results are generally similar. The models use a number of measures of leadership removal, to gauge different effects over time. To see if a relationship exists between decapitation and violence in the area in which the targeted group operates, information was gathered on the geographic territory of each major DTO in Mexico, each year. Sources used included Mexican newspapers, STRATFOR maps, and reports (e.g., Guerrero Gutiérrez 2011b). For example, the Tijuana Cartel is coded as operating primarily in Baja California. To determine effects of the arrest of the group’s leader in October 2008, Baja California is coded “1” for decapitation for that month. Other groups have influence over larger areas. The Zetas’ territory is coded as including nine states in 2012, so all nine states are coded “1” for decapitation in September of 2012, when a high-level leader was arrested.13 The primary sources for decapitation information are the Mexican government and news sources. The Calderón administration made a media spectacle of every arrest or killing of a high-level DTO member, so evidence of decapitations is not difficult to find. My analysis finds 63 leadership kills or arrests, from the first organizational decapitation in September 2007 through November 2012.14 The majority of these, 54, were arrests. There was only one leadership removal in 2007 and five in 2008, but at least 12 in each other year of the Calderón presidency (December 2006–December 2012). The primary decapitation measure, decapitation in the past three months, is a dichotomous variable coded “1” for the state in which a group is operating when the Mexican government kills or arrests a high-ranking leader of the group in the past three months. The period of three months is used to allow time for an effect to occur. For example, if

13. The average group is substantially operating in an average of three states (median p 2) over the course of 2006–12. The Sinaloa Cartel and the Zetas are in more states than other groups, with averages of six and eight states respectively over the seven years. However, if control variables are included to take into consideration the presence of these unusual groups in a state, results are substantively the same. 14. This rate, about one decapitation per month, is slightly higher than that of Guerrero Gutiérrez’s (2011a) study, which found about .7 decapitations per month between September 2007 through February 2011. This is indicative of an increase in use of the tactic as the Calderón presidency progressed.

a leader is arrested in March, the state or states in which his group operates are coded for this variable in April, May, and June. This permits us to see if a decapitation is associated with drug-related homicides in subsequent months. Additionally, six- and nine-month periods after a decapitation are tested. Models also include a measure of the cumulative number of past decapitations against groups operating in the state. This variable measures the aggregate effect of organizational decapitation over time—the longer-term effects of leadership removal referred to in Hypothesis 2. The highest value of the variable is 10, which both Michoacán and Tamaulipas had in 2011.15 The models include a number of control variables to take into consideration other factors that are likely to affect the number of drug-related deaths in a state-month. All models include a lag of the dependent variable because it seems likely that the previous month’s drug-related homicides are related to the number in the month being analyzed. Models include disputed territory, coded “1” if sources indicate substantial competition over territory in the state at that time. News and policy reports are used to code the variable. This variable is coded for each year because of difficulty coding DTO territory each month. Neighbor state deaths is the mean of drug-related homicides in states bordering the state being analyzed. Violence in one state is often related to violence in neighboring states because of spillover or other neighborhood or spatial effects. This variable addresses issues of spatial autocorrelation or the notion that states are not independent from each other (Franzese and Hays 2008).16 Independent variables are lagged one month, to avoid reverse causality issues, and because it takes time for violence to be affected. Variables coded for the entire year are lagged one year. Models include year dummy variables to take into consideration temporal issues. These year controls are important because, for example, Veracruz’s dynamics of violence in 2011 were substantially different from those in other years. State-level fixed effects are used to further take into consideration the lack of independence across state-month observations, including nonobservable or difficult-to-measure attributes. A Hausman tests suggests fixed effects are generally preferable to random effects.

15. Results hold if these states, or other states with many decapitations or homicides, are excluded from the analysis. 16. The spatial lag, like other independent variables, is also temporally lagged. This is important to reduce the risk of simultaneity bias. See Franzese and Hays (2008, 758).

Volume 77

The estimation method is a negative binomial regression because OLS regression is not appropriate due to the count nature of the dependent variable and its nonnormal dispersion (Long 1997). This is a standard estimator for counts of incidents, and the dispersion and proportion of zeroes fits this model better than a Poisson or zero-inflated model. Results show Bayesian Information Criterion (BIC) scores, where a lower number suggests better model fit.

Results Table 2 shows results with government data on drug-related homicides used for the dependent variable. Model 1 includes the measure of decapitation within the past three months and the measure of cumulative decapitations. Decapitation

Number 2

2015 / 331

in the past three months is statistically significant and negatively signed, suggesting that killing or arresting a leader in recent months is associated with a reduction in drug-related homicides. This suggests support for Hypothesis 1. Cumulative decapitations is statistically significant and positive, suggesting that a higher number of past decapitations affecting the state is associated with a higher number of drugrelated homicides—while controlling for other important factors. This suggests a longer-term positive relationship between decapitation and violence, consistent with Hypothesis 2. Interestingly, a positive relationship between cumulative decapitations and violence is the opposite of what studies of terrorist groups have found (Morehouse 2014). In the case of the FARC, cumulative decapitations are associated

Table 2. Negative Binomial Regressions of Drug-Related Homicides in Mexico

Decapitation in past three months

Model 1

Model 2

Model 3

Decapitation in Past Three Months

Decapitation in Past Six Months

Decapitation in Past Nine Months

2.126 (.039)* 2.119

Decapitation in past six months

(.043)* 2.039

Decapitation in past nine months

(.048) Cumulative decapitations Drug-related deaths previous month Disputed territory Neighbor state deaths Constant

.054

.054

.042

(.014)*

(.014)*

(.015)*

.004

.004

.004

(.000)*

(.000)*

(.000)*

.110

.115

.117

(.052)*

(.052)*

(.052)*

.003

.003

.003

(.001)*

(.001)*

(.000)*

.828

.865

.840

(.100)*

(.100)*

(.100)*

State fixed effects

Yes

Yes

Yes

Year fixed effects

Yes

Yes

Yes

11,119

11,121

11,128

n

1824

1824

1824

(months)

(57)

(57)

(57)

BIC

Note—Dependent variable is Mexican government data on drug-related homicides in Mexican state-months. * p ! .05.

Table 3. Arresting vs. Killing DTO Leaders and Targeting Top vs. Other DTO Leaders

Arrest, past three months Kill, past three months

Model 4

Model 5

Model 6

Model 7

Model 8

Model 9

Removal in Past Three Months

Removal in Past Six Months

Removal in Past Nine Months

Removal in Past Three Months

Removal in Past Six Months

Removal in Past Nine Months

2.076 (.040) 2.010 (.059) 2.121 (.045)* .064 (.051)

Arrest, past six months Kill, past six months

2.055 (.049) .110 (.051)*

Arrest, past nine months Kill, past nine months

2.097 (.053) 2.119 (.045)*

Top leader, past three months Other leader, past three months

2.063 (.043) 2.127 (.050)*

Top leader, past six months Other leader, past six months Top leader, past nine months Other leader, past nine months Cumulative decapitations Drug-related deaths previous month Disputed territory Neighbor state deaths Constant State fixed effects Year fixed effects BIC n (months)

.047 (.014)* .004 (.000)* .117 (.051)* .003 (.000)* .828 (.099)* Yes Yes 11,132 1824 (57)

.050 (.015)* .004 (.000)* .122 (.052)* .003 (.000)* .865 (.100)* Yes Yes 11,129 1824 (57)

.034 (.015)* .004 (.000)* .132 (.052)* .003 (.000)* .833 (.100)* Yes Yes 11,131 1824 (57)

.055 (.014)* .004 (.000)* .108 (.052)* .003 (.000)* .827 (.099)* Yes Yes 11,126 1824 (57)

Note—Dependent variable is Mexican government data on drug-related homicides in Mexican state-months. * p ! .05.

.060 (.015)* .004 (.000)* .104 (.052)* .003 (.000)* .865 (.100)* Yes Yes 11,127 1824 (57)

2.038 (.039) 2.026 (.050) .043 (.015)* .004 (.000)* .113 (.052)* .003 (.000)* .840 (.100)* Yes Yes 11,135 1824 (57)

Volume 77

with decreased violence, which can be seen quantitatively in Morehouse’s study or in a case study in the online appendix of the current article. Models 2 and 3 show different tests of the hypotheses and suggest that the negative relationship between decapitation and drug-related homicides only lasts until about six months. This short-term effect is markedly different from results of decapitation studies involving terrorist or guerrilla groups, which find more sustained decreases of violence (e.g., Johnston 2012). The coefficient for decapitation in the previous six months is statistically significant and negative, which suggests support for Hypothesis 1. The coefficient for decapitation in the previous nine months is not statistically significant, consistent with Hypothesis 2. Control variables return expected results. Regarding substantive impact, incident rate ratios (not shown) suggest that a decapitation in the previous three months is associated with a 12% reduction in drug-related homicides, other factors held constant. The estimated impact is an 11% reduction for a decapitation in the previous six months. Regarding cumulative decapitations, each additional arrest or kill is associated with a 5% increase in drugrelated homicides, other factors held constant. These results are robust to different model specifications. Excluding the lagged dependent variable, using TBI data instead of government data, or using random effects (to include time-invariant variables) returns similar results. Results for hypothesized relationships are not changed. Table 3 shows a final evaluation of the consequences of leadership decapitation. The models test to see if there is a difference in results when a leader is arrested as opposed to killed or when the top leader is targeted instead of a midlevel leader. Models 4–6 show an interesting result: only arrests have a negative statistically significant relationship with homicides, and this occurs up to six months after the arrest. (Arrests in the previous three months are negatively associated with homicides, Model 4, but only at the p ! .10 level of statistical significance.) Kills are positively related to drug-related homicides but only if a capo has been killed in the past nine months (Model 6). The kill result should be taken with caution, as there were only nine leadership killings in the sample, but the difference between arrests and kills is remarkable. The short-term effectiveness of arresting leaders as opposed to killing them is interesting because studies of political groups have shown the opposite result: killing leaders of those types of groups is more effective than arrests at reducing violence (Johnston 2012; Jordan 2009). A possible explanation for this divergence of results between political and criminal groups is that killing a political group’s leader

Number 2

2015 / 333

can sap essential motivation, causing group members and the wider public to lose confidence in the organization. Killing a DTO leader, however, is more likely to quickly lead to fragmentation and fights over market territory for these types of groups. This would suggest that the short-term drop in violence generally seen after leadership removal of criminal groups might not occur if the leader is killed. (Arrested DTO leaders often continue to lead from prison, which might slow down operations but is less likely to cause fragmentation.) This implies an interesting conditioning effect based upon whether a leader is killed or arrested, but overall it is consistent with the argument about incentives and markets of the respective groups. In models 7–9, the decapitation measures are disaggregated by whether the targeted leader was (1) the top leader of the organization or (2) a midlevel leader such as the boss of a geographic region or a particular division of workers such as hit men. Targeting a top leader is not significantly related to drug-related homicides in the following three, six, or nine months. Targeting a midlevel leader, however, is negatively associated with drug-related homicides up to three or six months after the leadership removal. The difference in results between targeting top leaders and targeting other leaders is interesting because, as with killing vs. arrests, the results here are distinct from what some studies of political groups have shown. Jordan (2009) tests for differences in the likelihood of terrorist group collapse depending on leader type but finds no difference. Similarly, Rowlands and Kilberg (2011) mostly find no difference between top and other leaders. Most other studies have not disaggregated decapitation effects by type of leader because of challenges with data collection. Regarding the results reported presently, it is possible that midlevel managers are more consequential to criminal groups in the short term in particular because they handle day-to-day operations—the fundamental business tasks including collecting revenue and paying members (Hope 2013). This would explain why targeting midlevel managers seems to reduce criminal violence in particular and is consistent with the notion that disruptions to the flow of material incentives are likely to interrupt operations, including violence.

CONCLUSION Scholars and governments seek to understand the consequences of leadership removal, but most research on this tactic has looked at political organizations such as terrorist groups. At the same time, criminal organizations increasingly threaten governments. This study contributes to discussions of these topics by explaining demarcations between political and criminal groups and presenting evidence for

334 / How Does Leadership Decapitation Affect Violence? Brian J. Phillips

how leadership removal affects the latter. Groups differ in terms of the incentives they offer members and the “markets” in which they operate. Criminal groups depend on material incentives, and leaders coordinate such incentives, so leadership decapitation can disrupt business—in the short term. However, another aspect of criminal organization behavior is that longer-term dynamics are related to illicit markets. As long as markets demand the services of criminal groups, any debilitated organizations will probably be replaced. Criminal violence is likely to continue in spite of leadership removal and could increase because of it. Empirical tests using original data suggest that this argument finds support in the case of the Mexican drug war. This is the most comprehensive quantitative study of decapitation of criminal organizations. Results showed that decapitation in the past three, or possibly six, months is associated with a reduction in drug-related homicides in the states in which the targeted group operates. This goes against what is basically conventional wisdom that decapitation has only increased violence in Mexico. In the longer term, however, decapitation is associated with no change in violence or an increase. This contrasts with recent research on terrorist groups, which finds leadership removal associated with decreased violence (Johnston 2012; Morehouse 2014; Price 2012). The case studies in the online appendix further illustrate how leadership removal can have divergent effects for criminal groups and political groups. Some readers might be surprised that DTO leadership removal in Mexico is associated with any decrease in violence, even short-term, because of the dramatic national increase in bloodshed in recent years. The government’s decision to directly and militarily confront the DTOs starting in late 2006 has clearly had negative consequences (Rios 2013). However, interestingly, a short-term drop in violence followed by a longer-term increase is what happened at the national level in Mexico. The homicide rate decreased between 2006 and 2007 and then began its several-year increase.17 The results have interesting implications for research on organizational decapitation and on subnational organized violence generally. Researchers should think more about categories of organizations and how differences can condition relationships.18 The framework of incentives and markets can be used to understand both criminal and political

17. According to UNODC data, the Mexican homicide rates per 100,000 inhabitants for 2006, 2007, and 2008 were 9.7, 8.1, and 12.7, respectively. 18. For examples of how different samples of violent groups lead to divergent results, see Phillips (2015).

groups, including why government policies should have divergent effects on different types of groups. Studies of political groups might find value in splitting samples based on each group’s level of criminal activity.19 The findings suggest that government policies deserve further scrutiny. As long as supply and demand exist for illegal goods such as drugs, new criminal organizations are likely to replace defeated groups. Leadership decapitation does not seem to be a fruitful solution for long-term reductions in violence. Alternative solutions include addressing supply or demand issues, increasing opportunities for people likely to join criminal groups, and improving judicial institutions. A note should be made regarding to what extent leadership decapitation is effective against criminal groups. Beyond wanting to reduce violence, the Mexican government has stated other goals, such as reducing the control of larger DTOs (Chabat 2010). It is possible that through fragmentation, this goal is in the process of being achieved.20 Is the breakup of some large DTOs worth a likely increase in violence? Overall, scholars and policy makers should think seriously about differences across group types. The effects of government policies against violent groups are likely to be conditioned by each group’s type—as this indicates its members’ sources of incentives and the wider “market” context in which the group operates. Regarding leadership removal in particular, extant research suggests it often reduces the violence of political groups. Evidence presented here, however, suggests markedly different results when leadership removal is used against criminal groups. The policy apparently affects criminal and political groups differently because of their dissimilar motivations.

ACKNOWLEDGMENTS Previous versions of the article were presented at CIDE, the Universidad Católica de Córdoba, and at the 2013 ISA and APSA annual conferences. I thank participants for their helpful feedback. I also thank Jorge Chabat, Jean Daudelin, Ana Carolina Garriga, Justin Hastings, Jenna Jordan, Gerardo Maldonado, James Ron, Carlos Vilalta, and Joe Young for valuable comments on drafts and Elsy González Cubría for excellent research assistance.

19. Additional single-country studies could be productive. Groups in Iraq, for example, vary from criminals to jihadists (Williams 2009) and could be compared. 20. Fragmentation could eventually lead to weaker groups and less violence. Future research could consider organizational density and the inverse-u relationship that scholars have found between the number of organizations in an area and their survival (Hannan and Freeman 1989).

Volume 77

REFERENCES Andreas, Peter. 2004. “The Clandestine Political Economy of War and Peace in Bosnia.” International Studies Quarterly 48 (1): 29–52. Asal, Victor, and Karl Rethemeyer. “The Nature of the Beast: Organizational Structures and the Lethality of Terrorist Attacks.” Journal of Politics 70 (2): 437–449. Bateson, Regina. 2012. “Crime Victimization and Political Participation.” American Political Science Review 106 (3): 570–87. Berman, Eli. 2009. Radical, Religious, and Violent: The New Economics of Terorrism. Cambridge, MA: MIT Press. Carreras, Miguel. 2013. “The Impact of Criminal Violence on System Support in Latin America.” Latin American Research Review 48 (3): 85– 107. Chabat, Jorge. 2010. Combatting Drugs in Mexico Under Calderon: The Inevitable War: Centro de Investigacion y Docencia Economicas (CIDE) Working Paper DTEI 205. Cronin, Audrey Kurth. 2009. How Terrorism Ends: Understanding the Decline and Demise of Terrorist Campaigns. Princeton, NJ: Princeton University Press. 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 (1): 132–42. Della Porta, Donatella. 2013. Clandestine Political Violence. Cambridge: Cambridge University Press. Dickenson, Matt. 2014. “The Impact of Leadership Removal on Mexican Drug Trafficking Organizations.” Journal of Quantitative Criminology 30 (4): 651–76. Donovan, John, and Richard Timothy Coupe. 2013. “Animal Rights Extremism: Victimization, Investigation and Detection of a Campaign of Criminal Intimidation.” European Journal of Criminology 10 (1): 113–32. Dube, Arindrajit, Oeindrila Dube, and Omar García-Ponce. 2013. “CrossBorder Spillover: U.S. Gun Laws and Violence in Mexico.” American Political Science Review 107 (3): 397–417. Flanigan, Shawn Teresa. 2012. “Terrorists Next Door? A Comparison of Mexican Drug Cartels and Middle Eastern Terrorist Organizations.” Terrorism and Political Violence 24 (2): 279–94. Frankel, Matt. 2011. “The ABCs of HVT: Key Lessons from High Value Targeting Campaigns Against Insurgents and Terrorists.” Studies in Conflict & Terrorism 34 (1): 17–30. Franzese, Robert J., and Jude C. Hays. 2008. “Interdependence in Comparative Politics: Substance, Theory, Empirics, Substance.” Comparative Political Studies 41 (4/5): 742–80. Gambetta, Diego. 1993. The Sicilian Mafia: The Business of Private Protection. Cambridge, MA: Harvard University Press. Grillo, Ioan. 2011. El Narco: Inside Mexico’s Criminal Insurgency. New York: Bloomsbury. Guerrero Gutiérrez, Eduardo. 2011a. La raíz de la violencia. Nexos, June 2011. Guerrero Gutiérrez, Eduardo. 2011b. Security, Drugs, and Violence in Mexico: A Survey. Washington, DC: North American Forum. Hoffman, Bruce. 2006. Inside Terrorism. New York: Columbia University Press. Hope, Alejandro. 2013. Violencia 2007–2013. La tormenta perfecta. Nexos, November, 2013. Johnston, Patrick B. 2012. “Does Decapitation Work? Assessing the Effectiveness of Leadership Targeting in Counterinsurgency Campaigns.” International Security 36 (4): 47–79. Jordan, Jenna. 2009. “When Heads Roll: Assessing the Effectiveness of Leadership Decapitation.” Security Studies 18 (4): 719–55. Jordan, Jenna. 2014. “Attacking the Leader, Missing the Mark: Why Terrorist Groups Survive Decapitation Strikes.” International Security 38 (4): 7–38.

Number 2

2015 / 335

Juergensmeyer, Mark. 1991. “Sacrifice and Cosmic War.” Terrorism and Political Violence 3 (3): 101–17. Kydd, Andrew H., and Barbara F. Walter. 2006. “The Strategies of Terrorism.” International Security 31 (1): 49–80. Langdon, Lisa, Alexander J. Sarapu, and Matthew Wells. 2004. “Targeting the Leadership of Terrorist and Insurgent Movements: Historical Lessons for Contemporary Policy Makers.” Journal of Public and International Affairs 15 (Spring): 60–78. Long, Scott J. 1997. Regression Models for Categorical and Limited Dependent Variables. Vol. 7, Advances Quantitative Techniques in the Social Science Series. Thousand Oaks, CA: Sage. Makarenko, Tamara. 2004. “The Crime-Terror Continuum: Tracing the Interplay Between Organised Crime and Terrorism.” Global Crime 6 (1): 129–45. Mannes, Aaron. 2008. “Testing the Snake Head Strategy: Does Killing or Capturing its Leaders Reduce a Terrorist Group’s Activity?” Journal of International Police Solutions 9: 40–49. Marselli, Riccardo, and Marco Vannini. 1997. “Estimating a Crime Equation in the Presence of Organized Crime: Evidence From Italy.” International Review of Law and Economics 17 (1): 89–113. Morehouse, Matthew. 2014. “It’s Easier to Decapitate a Snake than It Is a Hydra: An Analysis of Colombia’s Targeted Killing Program.” Studies in Conflict & Terrorism 37 (7): 541–66. Morrison, John F. 2013. The Origins and Rise of Dissident Irish Republicanism: The Role and Impact of Organizational Splits. New York: Bloomsbury. Naím, Moisés. 2012. “Mafia States: Organized Crime Takes Office.” Foreign Affairs May/June: 100–12. Olson, Mancur. 1965. The Logic of Collective Action: Public Goods and the Theory of Groups. Cambridge, MA: Harvard University Press. Oots, Kent Layne. 1989. “Organizational Perspectives on the Formation and Disintegration of Terrorism Groups.” Studies in Conflict and Terrorism 12 (3): 139–52. Pearlman, Wendy, and Kathleen Gallagher Cunningham. 2012. “Nonstate Actors, Fragmentation, and Conflict Processes.” Journal of Conflict Resolution 56 (1): 3–15. Phillips, Brian J. 2015. “What is a Terrorist Group? Conceptual Issues and Empirical Implications.” Terrorism and Political Violence. In press. doi:10.1080/09546553.2013.800048. Price, Bryan C. 2012. “Targeting Top Terrorists: How Leadership Decapitation Contributes to Counterterrorism.” International Security 36 (4): 9–46. Rios, Viridiana. 2013. “Why Did Mexico Become So Violent? A SelfReinforcing Violent Equilibrium Caused by Competition and Enforcement.” Trends in Organized Crime 16 (2): 138–55. Rios, Viridiana, and David A. Shirk. 2011. Drug Violence in Mexico: Data and Analysis Through 2010. San Diego: Trans-Border Institute, Joan B. Kroc School of Peace Studies, University of San Diego. Rowlands, Dane, and Joshua Kilberg. 2011. Organizational Structure and the Effects of Targeting Terrorist Leadership. Centre for Security and Defence Studies Working Papers, Norman Paterson School of International Affairs. Carleton University Schedler, Andreas. 2014. “The Criminal Subversion of Mexican Democracy.” Journal of Democracy 25 (1): 5–18. Shapiro, Jacob N. 2013. The Terrorist’s Dilemma: Managing Covert Organizations. Princeton, NJ: Princeton University Press. Shelley, Louise I., and John T. Picarelli. 2002. “Methods not Motives: Implications of the Convergence of International Organized Crime and Terrorism.” Police Practice and Research 3 (4): 305–18. Varese, Federico. 2011. Mafias on the Move. Princeton, NJ: Princeton University Press.

336 / How Does Leadership Decapitation Affect Violence? Brian J. Phillips Volkov, Vadim. 2002. Violent Entrepreneurs: The Use of Force in the Making of Russian Capitalism. Ithaca, NY: Cornell University Press. Weinstein, Jeremy M. 2007. Inside Rebellion: The Politics of Insurgent Violence. Cambridge: Cambridge University Press. Williams, Phil. 1994. “Transnational Criminal Organizations and International Security.” Survival 36 (1): 96–113.

Williams, Phil. 2009. “Illicit Markets, Weak States, and Violence: Iraq and Mexico.” Crime Law and Social Change 52 (3): 323–36. Williams, Phil. 2012. “The Terrorism Debate Over Mexican Drug Trafficking Violence.” Terrorism and Political Violence 24 (2): 259– 78. Wilson, James Q. 1973. Political Organizations. New York: Basic Books.