Living in Fear: The Dynamics of Extortion in Mexico’s Drug War

Territorial Control

A DTO’s territorial control, as well as its expected control over time, are critical for understanding its behavior toward civilians. Drawing on Olson (1993), Metelits (2009), and Arjona (2017), we formulate the following hypothesis: If a DTO is in control of a region and expects to remain in control, it will develop self-restraint and even provide some share of the benefits from their trade to the population in order to maintain loyalty, enhance its legitimacy, and increase its capacity through continued control (Berman & Laitin, 2008; Berman, Shapiro, & Felter, 2011; Felbab-Brown et al., 2017; Iannaccone, 1992). We are building on ideas that coercion is costly and fails to produce full cooperation in many cases. For example, a community may keep silent because its members fear torture or even death if they report on the DTO. However, as is the case with rebel groups and even governments, this strategy may not produce active community collaboration (Arjona, 2017; Kalyvas, 2006; Mampilly, 2011; Wood, 2003). Obtaining high-quality intelligence about their territory may require self-restraint and even reciprocation of benefits, rather than coercion (Wood, 2003), which, at the extreme, can also backfire and inspire a strong campaign by a community against the DTO (Kalyvas, 2015). We therefore expect a DTO with this type of monopoly to cooperate, or at a minimum, restrain itself in its interactions with the communities in which it operates.

In contrast, we expect competition between DTOs for territorial control to be accompanied by increases in civilian extortion for several reasons. First, territorial contestation shortens time horizons, incentivizing armed groups to behave more as “roving” bandits rather than as “stationary bandits” (Olson, 1993). DTOs do not anticipate that they will be in control for long in these cases, so they reap as much as possible while they are. In addition, when armed groups compete for territorial control, sustaining cooperative relationships with the community becomes more difficult because competition creates a common-pool problem: Why would a criminal group restrain from preying on the community through extortion (or other negative behavior toward the population) when they expect other criminal groups not to restrain themselves (for which, potentially, all of the DTOs operating in the community may also be blamed) (Humphreys & Weinstein, 2006)? In the extreme, competing DTOs can commit so much abuse that citizens will refuse to pay extortion fees—or will migrate—because no criminal organization can credibly offer protection. Third, because fighting an active challenger is costly, DTOs facing contested territorial control may also adjust to declining profits from their “core” business by diversifying into a host of other criminal activities, including extortion, kidnapping for ransom, human trafficking, and theft. With the exception of extreme scenarios where profits from some of these activities dry up, these activities provide greater profit than drug trafficking alone in the short term.

Based on this discussion, we expect not only that DTOs should be more prone to extortion when they are vying for territorial control, but also that extortion should be most pronounced where DTOs use violence against each other. This violence not only typically shortens time horizons; it also imposes financial costs, which encourages DTOs to turn against the population to extract resources, as just described. In addition, turf wars tend to empower the most violent contingents of DTOs, who may then either shape group behavior or even start operating on their own, with little restraint in their interactions with the community.

At the same time, when DTOs increase extortion, they also tend to stop providing benefits to communities: DTOs reap fewer rewards from information when a rival has clearly invaded an entire area, rather than when these rivals only have secret ties or subtle incursions into mainly monopolistic territories. DTOs may also be less likely to receive credit for providing assistance to these communities. Moreover, once violence spreads, DTOs typically also have fewer resources with which to provide assistance. Finally, as less violent contingents of DTOs are disenfranchised, their preferred policies of providing benefits from their trade to the community may also be overruled.

Overall, then, we theorize that DTO monopolistic control should be associated with lower rates of extortion and higher rates of public service provision. Conversely, competition among rival criminal gangs should be associated with higher rates of extortion and lower rates of service provision, especially when violent conflict breaks out between DTOs.

DTO Relationships With the State

The second factor we consider is the relationship between DTOs and the state. In many cases, criminal groups “must find ways to work with the state or receive protection from the state to function” (Arias, 2017, p. 10). Some forms of DTO–state collaboration generate low levels of violence—for example, the symbiotic relationship between the PRI and DTOs during the era of one-party rule allowed the state to control the division of territory among the limited number of large DTOs. But, in contrast, when DTOs compete against each other and establish collusive pacts with different state actors to buy impunity, high rates of violence follow.

In terms of community collaboration, we theorize that complicity between DTOs and law enforcement—prosecutors, judges, and police—can produce impunity that encourages extortion and other criminal activity. When DTOs extort, rob, kidnap, or even make individuals disappear, the community cannot access legal punishment if the police, prosecutors, and judges are colluding with the criminals. Drawing from the economics of crime literature (Becker, 1968; Stigler, 1970), we hypothesize that higher levels of collaboration with law enforcement institutions are associated with a higher probability that DTOs will engage in extortion and other criminal activities affecting the community, given that this lower DTOs’ costs of engaging in these behaviors.

Mexico’s federal system also exhibits variation in the relationships between DTOs and the state. At the extreme, local authorities are at times too weak to resist or even demand concessions from DTOs, which become de facto rulers. In other states, local (or even regional and national) authorities cooperate with, and in some instances, submit to DTOs. Finally, some states resist DTO corruption and capture. We lack reliable measures about degree of local state collaboration, but the party controlling the governorship and the municipality serves as reasonable proxy. As Trejo and Ley (2017) argues, PAN governments have adopted policies to undo the informal protection deals that DTOs had established during PRI control. Moreover, during the Calderón administration, PAN-controlled states had more federal resources, including the armed forces and the federal police to fight DTOs (Durán-Martínez, 2015). This meant that states governed by the PAN under President Calderón were not only more inclined to undo corrupt deals with DTOs, but were also more able to resist DTOs because they received more resources to do so. We therefore expect less extortion from DTOs in states with PAN incumbents during the Calderón administration, the period we focus on in this article.

In contrast, we expect PRI incumbents to be more willing to sustain informal protection deals created under the one-party rule (Blume, 2017; Ríos, 2015; Shirk & Wallman, 2015; Trejo & Ley, 2017). In these places, we expect that collusive agreements between DTOs and governors, local prosecutors, and police are more likely to persist, translate into higher levels of impunity, and, in turn, produce higher levels of DTO extortion.

With respect to other political parties like the left-wing PRD, we do not have clear expectations. On one hand, local PRD governments might behave much like the PAN in that, when they take office, they disrupt informal protection deals between the PRI and DTOs. On the other hand, the PRD lacked the federal resources provided to PAN states, which made them significantly weaker. Existing literature has shown that PRD politicians have been disproportionately targeted by DTOs for assassination (Blume, 2017; Trejo & Ley, 2016). Politicians get killed for a variety of reasons, including (a) accepting money from a DTO and then getting caught in fighting between DTOs, (b) refusing to cooperate with criminal organizations, or (c) having a political opponent with DTO connections (Blume, 2017). If these arguments are correct, PRD politicians’ vulnerability to assassinations may also suggest greater involvement in DTO capture. We might therefore expect higher levels of extortion among PRD governments.

Leadership Style and Organizational Capacity

Beyond territorial competition between DTOs and state collaboration with DTOs, we theorize that leadership style and organizational characteristics also change the chances that DTOs will engage in extortion. Mexican drug cartels are large criminal networks centered on one or several capos who control a number of regional bosses. Regional bosses, in turn, control cities, towns, rural regions, and the routes for drug distribution. Regional bosses hire local commanders who then control the sicarios (soldiers). Sicarios are assisted by halcones (falcons), who are lower ranked cartel members who patrol local territories (Grillo, 2011).

Beyond these general characteristics, however, DTOs vary in their use of organizational hierarchy, which also changes over time. Some cartels such as the Sinaloa Cartel and the Gulf Cartel are organized hierarchically. Other cartels such as the Zetas and the Cartel Jalisco Nueva Generación are decentralized. These are loose associations of various criminal groups, rather than a strictly structured top-down group.

DTOs also vary in their size, from criminal cells that are mostly local (e.g., Guerreros Unidos or Los Ardillos), to regional networks (e.g., La Familia Michoacana, the Beltrán Leyva Organization, and the Tijuana Cartel), to national and transnational criminal enterprises (e.g., the Sinaloa Cartel, the Zetas, and the more recent Cartel Jalisco Nueva Generación).

Building on insights that hierarchical structures can better overcome principal–agent and other disciplinary problems (e.g., Shapiro, 2013), we argue that, in principle, less hierarchical DTOs should have more problems disciplining their criminal cells and soldiers and less capacity to engage in collaborative relationships with the community; this should then translate to more extortion and less assistance.

Moreover, in principle, DTOs with stable leadership should be better able to control their forces than DTOs that suffer leadership turnover. Leadership stability creates longer time horizons, which are necessary to develop ongoing cooperative relationships with the community (Olson, 1993) and to engage in governance provision (Felbab-Brown et al., 2017), because a stable leader knows that he will still be in control in the future to reap the rewards of investment in a community.

As argued above, DTOs in Mexico have suffered increasing rates of leadership turnover in the last decade, which, in turn, has shifted their focus from a strategy of trafficking drugs while maximizing collaboration with the community to a strategy of diversifying their portfolio to other crimes to increase their immediate profit. This metamorphosis was in part the unanticipated result of President Calderón’s (2006-2012) “beheading” strategy. When violent gangs are either separated from capos who know how to traffic drugs or forced out of drug trafficking, they turn to other mechanisms to generate profits, including extortion. We therefore expect lower levels of assistance and higher levels of extortion among DTOs that suffer leadership turnover.

Characteristics of Turf

Beyond the characteristics of the criminal organization, some characteristics of their turf might influence DTO behavior toward the population. Territories vary in their value and in the logistics of the local criminal activity. These might influence how DTOs engage with the community. In our statistical models, we will control for three types of turfs. First, drug-producing areas, which in the Mexican context mainly include regions that allow for marijuana and poppy cultivation. Growing these crops might require continued cooperation from the local community and might therefore be associated with more assistance.

Second, transit points or turf that allows DTOs to transport drugs within the country. In the Mexican context, these include dirt roads, small highways, and even high-speed freeways or airstrips. Because of the high degree of mobility, these operations are less embedded in the community and rely instead on “halcones” and some deals with states. Transit points might therefore be associated with less community assistance.

Third, border-crossings allow for drug smuggling across countries. Border-crossings are incredibly valuable turf in Mexico because of the value of the U.S. drug market, which is estimated to be around 100 billion U.S. dollars annually (Kilmer et al., 2014). Control of this type of turf might require more active community collaboration, not only to keep trafficking points secret—for example, the location of hidden tunnels used to smuggle drugs—but also for intelligence. Monopolistic control of these locations is critical to DTOs, and so they want information about any other operations occurring in these crossings, including their use without permission or “tax” payments, or any designs on them by rival DTOs. We therefore expect that DTOs will offer significant community assistance with this type of turf.

Based on our theoretical discussion, we empirically test the following hypotheses. Our main independent variable, territorial control, has different influence on extortion and assistance in opposite directions, so that more contestation and more inter-cartel violence increase extortion and reduce assistance. For our other independent variables, our theory has strong implications for either extortion or assistance, and sometimes both. We articulate each hypothesis as follows:

List Experiments and Survey Design

There is a significant literature on the problems related to measuring citizens’ engagement in and attitudes toward crime and civil conflict (Bruck, Justino, Verwimp, & Tedesco, 2016; Bullock, Imai, & Shapiro, 2011; Kalyvas & Kocher, 2009; Lyall, Blair, & Imai, 2013; Matanock & Garcia-Sanchez, 2014; Mosher, Meithe, & Hart, 2011; Stylianou, 2003; Warr, 2000). The highly sensitive nature of these topics motivates all actors involved to hide information.

The literature suggests maximizing the incidence of truthful responses to sensitive issues with unobtrusive measures, including list experiments (Blair & Imai, 2012; Corstange, 2009; Glynn, 2013; Gonzalez-Ocantos, Kiewit de Jonge, Melendez, Osorio, & Nickerson, 2012; Holbrook & Krosnick, 2010; Imai, 2011; Kuklinski, Cobb, & Gilens, 1997; Kuklinski, Sniderman, Knight, Piazza, Tetlock, Lawrence, & Mellers, 1997; Sniderman, 2011; Sniderman & Grob, 1996). ⁷

We conducted a series of list experiments in the nationwide survey carried out in Mexico in July 2011. ⁸ Table 1 below describes each of the experiments in the survey. It is important to note that the extortion questions were framed as issues of avoiding harm. Respondents understood we were not simply asking about transactions involving the purchase of drugs, the payment of private security guards, or the ubiquitous practice of bribing traffic policemen.

OPEN IN VIEWER

Table 1.

Description of List Experiments.

	Card 1	Card 2
Introduction to all groups	Please tell me how many of these things you have done in the past 6 months. We just want to know how many you have done, do not tell me which ones.	Please tell me how many of these things you have done in the past 6 months. We just want to know how many you have done, do not tell me which ones.
Control group	1. I got drunk at a party I went to.	1. I have received benefits from the Oportunidades program.
	2. I did some exercise outdoors.	2. I have participated in a tanda. a
	3. I attended church almost every Sunday.	3. I gave charity (limosna).
Treatment Group 1	4. I have seen cars or trucks with armed men who are not policemen in broad daylight.	4. I have given money to drug or criminal organizations so that they protect me.
Treatment Group 2	4. I asked for help from someone working for organized crime.	4. I have given money to the police so that they protect me.

a.

Rotating Savings and Credit Association (ROSCA).

We randomly selected 300 sampling points from the 64,937 electoral precincts in Mexico in 2011. The sample was stratified by type of electoral precinct (urban and rural) as defined by the Federal Electoral Institute (IFE), and by levels of drug-related violence at the municipal level (low, medium, and high) for a total of six strata.

Our strategy was to oversample areas of high violence to guarantee sufficient spread across areas of differing degrees of DTO dominance. We used the “Deaths Presumably Related to Criminal Rivalry” (SEGOB, 2011) dataset published by the Mexican government to stratify our sample. The data set classifies deaths into three categories: (a) homicides resulting from violence between DTO members (Executions), (b) killings from clashes between the Mexican authorities and DTOs (Confrontations), and (c) deaths related to unexpected attacks by DTOs on Mexican authorities (Aggressions). ⁹ We classified the level of inter-cartel violence using the total number of drug-related deaths in municipalities from January to May of 2011.

Drug-related deaths are highly concentrated in a few places: Half of these deaths occurred in 19 municipalities during the study period, whereas about two thirds of the municipalities reported no executions between DTO members. Given this distribution, we classified the high-violence group as the five most violent municipalities, which have a third of the total deaths related to criminal rivalry. ¹⁰ The medium-violence group includes the next 14 municipalities with most drug-related deaths, so that both groups together account for half of the homicides. The low-violence group comprises the rest of the municipalities (n = 2,437). ¹¹

Within each strata, we sampled electoral precincts with a selection probability proportional to its population size using the quartiles of the distribution of the number of registered voters, for a total of 22 groups. ¹² We collected nine questionnaires in each sampling point. Respondents were randomly assigned to one control group and two treatment groups of similar size for a total of 900 observations in each group and 2,700 observations for the whole sample. Summary statistics by treatment group and balance tests are shown in Appendix Table A1. Given the randomization strategy, regional characteristics are identical across groups. There were small differences in education, age, and gender across groups that are taken into account in the analysis.

Testing the List Experiment

List experiments rely on two assumptions: (a) no liars and (b) no-design effects. Floor and ceiling effects can generate “liars” if an individual has performed none or all of the actions listed; he or she may lie so as to not reveal that he or she has performed (or not performed) the sensitive item. Thus, all of the lists are designed to include items rarely expected from the same individual, so that most individuals will have performed at least one of the control items but not all of them (as noted above).

The other assumption inherent in the list experiment is that there is no “design effect.” That is, adding an item to the list will not have an effect upon the responses for the other items on the list. A “design effect” would entail that the items are not independent from each other.

We follow Blair and Imai (2012) to formally test that the identification assumptions for list experiments hold. Suppose that in a list experiment there are J control items and a single sensitive item J + 1. Let Z i j be an indicator variable of whether individual i has a preference for item j or, as in our experiment, she has been exposed to such situation. Let Y i ( 0 ) = j = 1 J Z i j and Y i ( 1 ) = ∑ j = 1 J + 1 Z i j represent the potential answers respondent i would give to the list experiment under the control and treatment conditions, respectively. Finally, let Z i , J + 1 * be the truthful answer to the sensitive item.

The “no-design effect” assumption implies that Y i ( 1 ) = Y i ( 0 ) + Z i , J + 1 ( 1 ) , while the “no-liars assumption” entails that Z i , J + 1 ( 1 ) = Z i , J + 1 * . Under both assumptions, the mean difference estimator is an unbiased estimate of the proportion in the population with an affirmative preference for (or exposure to) the sensitive item. Moreover, under these assumptions, it is possible to estimate the joint distribution of ( Y i ( 0 ) , Z i , J + 1 * ) , where each type of respondent type is represented by π y z = Pr ( Y i ( 0 ) = y , Z i , J + 1 * = z ) . In particular, if T i is the treatment status of individual i , then π_y1, which is the estimated proportion of respondents in the sample that have been exposed to y control items and to the sensitive item, can be estimated as follows:

π y 1 = Pr ( Y i ≤ y | T i = 0 ) − Pr ( Y i ≤ y | T i = 1 ) .

(1)

Consequently, the proportion π_y0 of respondents that have been exposed to y control items but not to the sensitive item can be inferred from the difference between the estimated mass probability Pr ( Y i = y | T i = 0 ) and π_y1.

All of the experiments were piloted in advance to avoid potential identification threats. The distribution of answers for each experiment for the control and two treatment groups are shown in Table 2. The table also shows the estimated π_y1 for each experiment. According to these estimates, approximately 9.7% of the (unweighted) sample has been extorted by DTOs, whereas 11.9% have received help from them. The prevalence of DTO activities is evident, with a remarkable 38.2% of respondents estimated to have seen armed convoys during daylight hours. Also interesting among these summary statistics, and quite worrisome, is that police extortion is estimated to be even higher than criminal extortion.

OPEN IN VIEWER

Table 2.

Answer Distribution of List Experiments.

Card 1
Y i	Control group		Treatment groups
	Frequency	Proportion	Convoys			Help from crime
			Frequency	Proportion	πy,1	Frequency	Proportion	πy,1
0	148	16.6%	108	12.1%	4.5%	127	14.2%	2.4%
1	424	47.5%	293	32.7%	19.3%	391	43.7%	6.2%
2	265	29.7%	346	38.6%	10.4%	300	33.5%	2.3%
3	56	6.3%	113	12.6%	4.0%	68	7.6%	1.0%
4			36	4.0%		9	1.0%
Total	893	100%	896	100%	38.2%	895	100%	11.9%
Card 2
Y i	Control group		Treatment groups
	Frequency	Proportion	Narco extortion			Police extortion
			Frequency	Proportion	πy,1	Frequency	Proportion	πy,1
0	181	20.2%	171	19.1%	1.1%	161	17.9%	2.3%
1	505	56.4%	465	52.0%	5.6%	470	52.4%	6.3%
2	167	18.7%	202	22.6%	1.7%	207	23.1%	1.9%
3	42	4.7%	45	5.0%	1.3%	55	6.1%	0.4%
4			12	1.3%		4	0.4%
Total	895	100%	895	100%	9.7%	897	100%	10.9%

The table shows the number respondents and the distribution of answers for each outcome and treatment group. The table also shows the estimated proportion of each respondent type π y z = Pr ( Y i ( 0 ) = y , Z i , J + 1 * = z ) under the no-design effect and no-liars assumptions (see Imai, 2011), where Y i ( 0 ) represents the total number of affirmative answers for control items and Z i , J + 1 * denotes the truthful occurrence of the sensitive item. Bold faced values are simple percentages, not estimated coefficients. They are in bold only to emphasize them.

Under the assumption of no-design effects, we should expect the cumulative distribution of y in the control group to be greater than or equal to the corresponding distribution under the treatment (Blair & Imai, 2012). Nevertheless, this distribution should be greater by at most one item at each level y . Algebraically, these two hypotheses are equivalent to the null hypothesis that π y z ≥ 0 for all y and z . As this is the case for each list experiment, we conclude that there is no evidence of design effects in our survey. ¹³

The low proportion of respondents in the two control groups for which Y i ( 0 ) = 3 also confirms that the experiments have no ceiling effects. The absence of design effects (i.e., the inclusion of the sensitive item does not seem to reduce the number of reported items) and the positive estimated proportions π₁₁ are clear indicators of the absence of substantial floor effects.

Average Effects

Our baseline linear model to measure the prevalence of DTO extortion and assistance in our sample is as follows:

y i j = α + β 1 T i + ∑ k δ k X i k + ∑ l γ l Z j l + ε i j ,

(2)

where y i is the number of items reported by the respondent i in municipality j and T i is an indicator variable for treatment. The model also includes k individual covariates and l socioeconomic characteristics at the municipality level.

The results of the list experiments show a significant presence of DTOs, including their use of both extortion and assistance strategies toward the communities in which they operate. Table 3 shows our estimated coefficients for each of the four list experiments described above. The estimated coefficients of the models in the first column are equivalent to those of a simple difference in means test. The models in column 2 adjust for the sampling design using individual sampling weights, and the models in column 3 also include individual and regional socioeconomic characteristics.

OPEN IN VIEWER

Table 3.

Average Effect of List Treatment.

	Average effect
Treatment	(1)	(2)	(3)
Card 1
Convoy	0.38***	0.46***	0.45***
	(0.04)	(0.07)	(0.07)
Help from criminals	0.12***	0.23***	0.22***
	(0.04)	(0.06)	(0.06)
Card 2
Criminal extortion	0.10**	0.13**	0.12**
	(0.05)	(0.06)	(0.05)
Police extortion	0.11***	0.14**	0.17***
	(0.04)	(0.07)	(0.06)
Sampling weights	No	Yes	Yes
Individual covariates	No	No	Yes
Regional covariates	No	No	Yes

The rows show the estimated coefficients of four different list experiments included in the Survey on Public Safety and Governance in Mexico in 2011. There were two questions (Card 1 and Card 2) for which respondents were randomly assigned into one control and two treatment groups (see Table 1). The coefficients are estimates of the proportion of the survey sample exposed to the sensitive item. The models in column 2 adjust for sampling weights, and the models in column 3 include individual and regional socioeconomic characteristics. Robust standard errors in parentheses.

*

p < .1. **p < .05. ***p < .01.

Our results show that, on average, one in ten Mexicans were extorted by criminal organizations in the 6 months prior to the survey. Police extortion seems to be as prevalent as criminal extortion. The prevalence rate of help from criminals is twice as high once we adjust for our oversampling of violent places. This suggests that narco help, as we demonstrate below, is more prevalent in places with low levels of violence. Finally, our results indicate that almost half of the Mexican population had seen a non-state armed convoy during daylight in the months prior to the survey. The security situation certainly does not represent stable state control. Column 3 shows that our results are not substantially affected when adjusting for slightly imbalanced individual characteristics.

Explanatory Variables

To test the specific hypotheses derived from our theory, we use a multivariate regression model of survey data for list experiments as proposed by Imai (2011) and Blair and Imai (2012). ¹⁴ We specified a linear model with identical covariates for the two treatments analyzed in this article, inquiring into the dynamics of extortion by DTOs and the assistance also provided by these criminal groups. The baseline models are similar to the ones described in Equation 2 but also include interaction variables of the treatment indicator with each of the individual and municipal characteristics. We adjusted our baseline specification as follows:

y i j = α + β 1 T i + ∑ k ( δ k + τ k T i ) X i k + ∑ l ( γ l + η l T i ) Z j l + ε i j ,

(3)

where τ k and η l are the coefficients for the interaction terms between the treatment indicator and the k individual and l regional characteristics, respectively.

Our theory highlights DTO territorial contestation as key independent variable shaping DTO behavior. To measure DTO contestation, we use Coscia and Rios’s (2012) ¹⁵ data on areas of operation of Mexican DTOs. As noted above, those authors used a Web crawler to extract information on the activity of criminal groups in Mexico from Google News. This consists of panel data where each column is a dummy variable indicating whether a given DTO was present in a municipality during a certain year. The data are available at the municipal level from 1990 to 2010. We define a criminal group as having a “dominant presence” in a municipality if it has reported operations (i.e. it appears in the database) each year between 2008 and 2010. We chose this period of three years to reduce potential classification errors in a single year.

For our analysis, a municipality is “contested” if two or more groups have a dominant presence; a municipality is a monopoly if only a unique, dominant group has a presence. We estimate that, between 2008 and 2010, DTOs had a monopolistic presence in 10.9% of the municipalities (n = 268) and contested in an additional 4.9% (n = 120). We estimate that more than half of the population lived in a municipality with a dominant presence of at least one criminal group (59% of the population), from which less than half lived in contested territories.

We stratified municipalities by levels of violence (“low,” “medium,” and “high”) as described in the “Implementation” section. Table 4 shows that all high-violence municipalities in our sample were contested and all low-violence municipalities had no DTO activity, which makes us more confident about our measure of territorial control. Maps in Appendix Figure A1 show areas of dominant presence for selected DTOs. Figure 3 displays our sampling points and the estimated number of cartels in each municipality.

OPEN IN VIEWER

Table 4.

Municipal Characteristics by Level of DTO’s Contestation.

	Units	Period	Contested(mean)(1)	Monopoly(difference)(2)	No cartel(difference)(3)
Violence and extortion a
Executions	total	2008-2011	153.06	−119.95***	−149.88***
Executions	rate	2008-2011	20.53	−4.44	−14.84***
Homicides	rate	2008-2011	33.63	−3.89	−16.96***
Homicides by firearms of males between 19 and 39 years of age	rate	2008-2011	93.72	−20.79	−56.32***
Extortion (SNSP)	rate	2008-2011	3.65	−1.28**	−3.11***
Socioeconomic characteristics
Urban municipality	percent	2011	0.71	−0.15***	−0.40***
Population	thousands	2011	253.83	−118.60***	−231.76***
PAN mayor	percent	2011	0.30	−0.04	−0.07*
PRI mayor	percent	2011	0.53	−0.02	−0.16***
Other mayors	percent	2011	0.17	0.06	0.24***
Marginalization index	unit	2010	−1.01	0.38***	1.15***
Low marginalization	percent	2010	0.72	−0.14***	−0.48***
Medium marginalization	percent	2010	0.20	0.02	0.03
High marginalization	percent	2010	0.08	0.12***	0.44***
Human development index	unit	2010	0.84	−0.03***	−0.08***
Illiteracy	percent	2010	0.08	0.02***	0.10***
Schooling	years	2010	7.71	−0.68***	−1.87***
Drug trafficking indicators a
Producer	percent	2008-2011	0.18	0.03	−0.03
Transit point	percent	2011	0.57	−0.15***	−0.44***
Border-crossing	percent	2011	0.07	−0.04*	−0.06***
Road distance to the United States	km	2011	795.73	87.51*	272.45***
Joint operation	percent	2011	0.54	−0.15***	−0.32***
Leader capture	percent	2008-2011	0.06	−0.01	−0.05***
Opium poppy eradication	plots	2008-2011	14.96	68.42	−0.54
Marijuana eradication	plots	2008-2011	65.23	37.17	−52.99***
Cocaine seizures	kg	2008-2011	96.91	50.74	−72.14**
Heroine seizures	kg	2008-2011	1.78	1.71	−1.48***
Marijuana seizures	tons	2008-2011	10.18	2.85	−9.13***
Municipalities	total		120	268	2,068

The table shows socioeconomic characteristics, drug trafficking indicators, and violence and extortion rates at the municipal level according to the number of DTOs with reported activities. Contestation was estimated using the Coscia and Rios (2012) database, which maps the yearly territorial presence of criminal groups using Google News queries. We define a cartel as having had a dominant presence in a municipality if it has had operations (as it appears in the database) in every year between 2008 and 2010. A territory is contested if two or more cartels have had a dominant presence. DTO = drug trafficking organizations; SNSP = Secretariado Nacional de Seguridad Pública; PAN = National Action Party; PRI = Institutional Revolutionary Party.

a.

Data from 2008 to 2011 correspond to the period between January 2008 and May 2011. Rates from 2011 were annualized to estimate average yearly rates across municipalities.

*

p < .1. **p < .05. ***p < .01.

OPEN IN VIEWER

Figure 3.

DTO dominance and sampling points.

Our theory argues, furthermore, about the importance of arrests of DTO’s leadership. This may matter either because it increases DTO violence or because it changes the structure of the organization. We include a dummy variable to indicate whether a drug capo or lieutenant was captured in a municipality within the last 3 years (January 2008 to May 2011). We use Calderón et al.’s (2015) dataset and update it up to 2011 with data from Guerrero (2011a).

Our theory also argues that state capture by DTOs is potentially important. We therefore control for indicators of which party controlled a municipality by the time of the survey, an important proxy as described above. We estimate the partisan effect of a PRI and a PAN municipal government, with the PRD as the base category. ¹⁶ We also present the second model where we interact territorial control by DTOs with party incumbency. Moreover, following studies that stress the importance of coordinated efforts to fight—or protect—DTOs (Durán-Martínez, 2015), we examine cases where the same party controls the state and the municipality.

We also control for whether there was a military joint operation in place at the time of the survey. These operations were coordinated efforts between the military and police forces (federal, state, and local) to regain state control and reestablish the rule of law in places overridden by violence and criminal activity. This also indicates federal resources directly used against DTOs in these states. At the time of the survey, there were six joint operations in nine states.

To test for the effects of DTO organization, we classify these groups according to their hierarchical organization (“Standard,” “Regional,” or “Clustered”), drawing from the United Nations Office on Drugs and Crime’s (UNODC; 2002) typology of crime organizations. Standard hierarchy groups are the most common form of organized crime organizations. They are characterized by a single leader with a direct line of command to regional bosses and a clearly defined hierarchy where systems of internal discipline are strict. The Sinaloa Cartel, prior to the arrest of “El Chapo” Guzmán in 2016, and the Juárez Cartel typify this classification. Regional hierarchy groups are characterized by a single leadership structure and a clear line of command from the center—but they also have an important degree of autonomy at the regional level. The Gulf Cartel typifies this classification. Clustered hierarchy groups are formed by component criminal organizations or gangs that have a governing arrangement between these affiliates. Each component has a higher degree of autonomy than those in the other two types of organizations. Los Zetas Cartel typifies this classification.

Classifying groups into these categories is done using information produced through surveys by the United Nations Office on Drugs and Crime’s typology of crime organizations among security experts and organizations from 16 member countries, in combination with existing literature and journalistic accounts.

DTO organization is subject to frequent changes when members are captured, killed, or otherwise removed. Our classification corresponds to how DTO organization looked in 2011, when the survey was collected. To further test these characteristics, especially should the reader disagree with our classification, we ran our models using our categorization and a single indicator for each individual DTO.

Our theory also argues that the characteristics of turf matter, including suitability for drug production, transportation infrastructure, and distance to the U.S. border. We use information on the Mexican transportation network and available information on government seizures to measure these. ¹⁷ Our theory distinguishes between the following territories: drug-producers, transit points, and border-crossings. We define a municipality as a “drug producer” if at least five hectares of crops of marijuana or opium poppy were eradicated by the government between January 2010 and May 2011. A municipality is a potential “transit point” if it has strategic transportation infrastructure, in particular airports, ports, and/or highways. Finally, a municipality is a potential “crossing point” if it has one of the more than 30 known border-crossings to the United States.

We also include a dummy variable indicating whether the polling point (electoral section) is considered urban according to Mexico’s IFE and a variable indicating the municipal level of development by using the marginalization index constructed with 2010 census data (Consejo Nacional de Población, 2010).

The models below also include individual variables that allow us to highlight groups of respondents that show distinctive patterns, and to control for any imbalance in randomization. We include a full set of demographic variables—sex, age, occupation (including unemployment), education, and recipient of social transfers through the Oportunidades program as a proxy for poverty—that control for individual characteristics which may affect citizens’ likelihood of being victimized. ¹⁸

Table 4 presents summary statistics of municipal characteristics by the degree of territorial contestation between DTOs for all municipalities in Mexico. Column 1 shows summary statistics of contested municipalities. Columns 2 and 3 show the difference for places with monopolistic control and no dominant DTO, respectively.

We also include in the table extortion-related complaint rates at local public prosecutor offices as reported by the Secretariado Nacional de Seguridad Publica (SNSP). We use these data as an alternative measure for extortion in subsequent sections, comparing it with our reported rates through the list experiment in our survey.

It can be seen that violent conflict between DTOs and territorial contestation are highly correlated. Contested territories are more violent than those with monopolistic control and even more with respect to places with no DTO activity, both in total number of deaths and rates per hundred thousand inhabitants (using alternative measures for violence). Similarly, extortion-related complaint rates at local public prosecutor offices are also higher in contested places.

With respect to socioeconomic characteristics, the differences in Table 4 also suggest that territorial contestation tends to happen in urban and more developed places, with strategic transportation infrastructure, and closer to the U.S. border. In contrast, monopolies are more prevalent in rural and poorer areas. Nevertheless, they seem to grow more crops of marijuana and opium poppy and transit almost twice the drugs than contested places, although none of these differences were statistically significant.

Interestingly, kingpin captures tend to happen equally between contested municipalities and those under monopolistic control, but state action in the form of joint operations tends to focus on contested places. It is worth noting that there are no differences in incumbent parties between these two types of territories. Nevertheless, municipalities with no DTO dominant presence tend to have fewer PRI governments.

Territorial Control

Table 5 shows the estimated coefficients for the experiments on DTO extortion and community assistance. To untangle the relationship between contestation and violent conflict between DTOs, we estimated three models for each experiment: one using our proxy for territorial contestation, the second using our categories for levels of DTO violence, and the third one interacting both variables. For interpretation purposes, we use as the baseline category places of high violence, all of them contested in our sample. We also group places of medium and low violence into a single category to make comparisons on interacted variables more intuitive.

OPEN IN VIEWER

Table 5.

Estimates on Narco Extortion and Help.

Variable	Narco extortion			Narco help
	(1)	(2)	(3)	(4)	(5)	(6)
Sensitive item
Violence and contestation
Monopoly	−0.13**			0.18***
	(0.05)			(0.07)
No cartel	0.13			0.18
	(0.10)			(0.12)
Low Violence		−0.14**			0.08
		(0.06)			(0.07)
× Contested			−0.10			−0.03
			(0.08)			(0.09)
× Monopoly			−0.19***			0.16**
			(0.05)			(0.08)
× No Cartel			0.05			0.16
			(0.11)			(0.14)
Government interventions
Joint operation	0.19**	0.13*	0.15**	0.13	0.14	0.11
	(0.08)	(0.08)	(0.07)	(0.08)	(0.09)	(0.09)
Leader capture	−0.14***	−0.06	−0.10	−0.21***	−0.25***	−0.20***
	(0.05)	(0.05)	(0.06)	(0.06)	(0.07)	(0.07)
Characteristics of the turf
Drug producer	−0.12**	−0.12**	−0.11*	−0.02	−0.04	−0.02
	(0.06)	(0.05)	(0.06)	(0.06)	(0.07)	(0.06)
Transit point	0.07	0.00	0.06	−0.16*	−0.18**	−0.17*
	(0.11)	(0.11)	(0.10)	(0.09)	(0.09)	(0.09)
Border-crossing	−0.03	−0.12**	−0.07	0.17**	0.23***	0.16*
	(0.06)	(0.06)	(0.06)	(0.08)	(0.08)	(0.09)
Other controls
PRI mayor	−0.11	−0.09	−0.10	0.01	−0.07	0.02
	(0.10)	(0.09)	(0.09)	(0.11)	(0.11)	(0.11)
PAN mayor	−0.09	−0.03	−0.05	−0.05	−0.11	−0.04
	(0.10)	(0.09)	(0.10)	(0.11)	(0.12)	(0.12)
Urban	0.08	0.08	0.09	0.16**	0.16**	0.16**
	(0.09)	(0.09)	(0.09)	(0.07)	(0.08)	(0.07)
Marginalization—Medium	0.03	0.06	0.05	−0.20	−0.19	−0.19
	(0.14)	(0.14)	(0.14)	(0.13)	(0.13)	(0.13)
Marginalization—High	−0.00	0.11	0.01	−0.28**	−0.22*	−0.28**
	(0.16)	(0.15)	(0.16)	(0.13)	(0.11)	(0.13)
Observations	1,778	1,778	1,778	1,782	1,782	1,782
R 2	.10	.10	.10	.08	.07	.08

The rows show estimated coefficients for the sensitive item of two list experiments included in the Survey on Public Safety and Governance in Mexico in 2011. There were two questions (Card 1 and Card 2) for which respondents were randomly assigned into one control and two treatment groups (see Table 1). A full coefficient table is available upon request. All models include individual-level characteristics. Robust standard errors by municipality in parentheses. PRI = Institutional Revolutionary Party; PAN = National Action Party.

*

p < .1. **p < .05. ***p < .01.

The coefficients for monopolistic control, lower violence, and their interaction are negative and significant in Columns 1 through 3, indicating higher levels of extortion by criminal groups in contested and high-violence areas as compared with less violent areas under monopolistic control. These results support our theory about the effects of territorial contestation increasing and monopoly control decreasing extortion.

In contrast, the coefficients in the last three columns of Table 5 show that, as predicted by our theory, areas of monopolistic control are associated with more assistance from criminal groups. Moreover, the model in column 6 predicts more help in places under monopolistic control and low violence than in those with no dominant presence of DTOs and low violence.

In terms of the other covariates in the model, leadership arrests have a significant and substantial negative coefficient for assistance. These results support the idea that DTOs with stable leadership can better sustain collaborative relationships with the community. The capture of a national or regional leader does not seem to influence extortion, contrary to our expectations. The effect of drug capo arrests on extortion appears to be mediated through how these cause significant increases of inter-cartel violence, as demonstrated by Calderón et al. (2015) and Phillips (2015). More inter-cartel violence, in turn, is associated with more DTO extortion.

The models in Table 5 control for joint operations, where we find significantly higher levels of extortion. There is substantial debate in the literature about whether joint operations directly caused violence or if, instead, they were assigned to places where violence was already escalating (Escalante Gonzalbo, 2011; Merino, 2011; Rosas, 2011). Our results cannot determine causality, although they do demonstrate that places where joint operations took place are also associated with significantly higher levels of extortion.

Finally, we find significantly more community assistance in border-crossings and less assistance in transit points, in support of our theory. Drug-cultivating areas exhibit less extortion and, contrary to our expectations, we do not find more assistance here.

The party controlling the municipality has no effect on extortion nor help. In subsequent sections, we find strong effects when accounting for the party controlling the state, which suggests that governors more than mayors are a key determining factor. In terms of individual-level variables, we found higher rates of extortion affecting poor, middle-aged males with medium levels of education. In contrast, we found that the poor and elderly population tend to receive the most help from narcos. We did not find significant differences by sex or between other age groups.

Figure 4 shows the average predictions from the full interacted model. For ease of interpretation, we present predicted values and their 90% and 95% confidence intervals. As expected, the model predicts a 22% incidence of extortion in contested municipalities with high levels of violence and the lowest prevalence in places under monopolistic control. The model also predicts a 17% prevalence of narco help in the later municipalities, the prediction is significantly different from zero.

OPEN IN VIEWER

Figure 4.

Predicted rates of DTO extortion and help.

DTO Relationships With the State

Our theoretical approach stresses that, conditional on levels of contestation, DTOs will be able to extort more if they collaborate with local law enforcement. In Mexico’s federal system, gaining collaboration from governors is critical because they appoint local prosecutors and police, which in turn shapes which crimes get investigated and prosecuted (Trejo & Ley, 2017). Moreover, following a strand in the literature that stresses the importance of coordinated efforts to fight or protect DTOs (Durán-Martínez, 2015), this section presents models on partisan control of governors and mayors.

Table 6 shows the estimated coefficients of our models of the effects of partisan control on DTO extortion and community assistance, and Figure 5 displays the predicted rates of Models 1 and 3. As described, the baseline scenario is a contested territory with a PRD governor. Consistent with our theory, results show that areas with a PAN governor have significantly lower extortion rates, especially those with a co-partisan mayor. Municipalities under this PAN juxtaposition also have the lowest predicted rates of extortion for both contested and monopolized territories (not shown here). We do not find significant differences for PAN-controlled municipalities with respect to help.

OPEN IN VIEWER

Table 6.

Estimates on Narco Extortion and Help by Incumbent Party.

	Narco extortion		Narco help
Variable	(1)	(2)	(3)	(4)
Sensitive item
PRI governor, other mayor	−0.01		0.03
	(0.07)		(0.10)
PRI governor and mayor	−0.15*		0.13*
	(0.08)		(0.08)
PAN governor, other mayor	−0.23***		0.07
	(0.08)		(0.10)
PAN governor and mayor	−0.24***		−0.03
	(0.09)		(0.08)
Contested
× PRI governor, other mayor		0.04		0.19
		(0.11)		(0.16)
× PRI governor and mayor		−0.14		0.22**
		(0.11)		(0.10)
× PAN governor, other mayor		−0.26**		0.16
		(0.11)		(0.11)
× PAN governor and mayor		−0.23*		0.11
		(0.12)		(0.10)
Monopoly	−0.09*		0.20***
	(0.05)		(0.06)
× PRI governor, other mayor		−0.16		0.24
		(0.11)		(0.18)
× PRI governor and mayor		−0.28***		0.42***
		(0.09)		(0.09)
× PAN governor, other mayor		−0.33***		0.48**
		(0.12)		(0.20)
× PAN governor and mayor		−0.23**		0.14*
		(0.09)		(0.08)
× Other incumbents		−0.08		0.22
		(0.14)		(0.19)
No cartel	0.15		0.17
	(0.09)		(0.12)
Observations	1,778	1,778	1,782	1,782
R 2	.11	.11	.08	.09

The rows show estimated coefficients for the sensitive item of two list experiments included in the Survey on Public Safety and Governance in Mexico in 2011. There were two questions (Card 1 and Card 2) for which respondents were randomly assigned into one control and two treatment groups (see Table 1). A full coefficient table is available upon request. All models include individual-level characteristics. Robust standard errors by municipality in parentheses. PRI = Institutional Revolutionary Party; PAN = National Action Party.

*

p < .1. **p < .05. ***p < .01.

OPEN IN VIEWER

Figure 5.

Predicted rates of narco extortion and help by incumbent party.

Moreover, in line with our theoretical expectations, PRI-governed states show significantly higher extortion rates than PAN-controlled areas. The difference in the predicted rate of extortion in places with PRI juxtaposition, relative to places with PAN juxtaposition, is more than 20 percentage points. Noticeably, we also find the highest predicted rates of DTO assistance when the PRI holds both the municipality and the governorship.

Surprisingly, the highest levels of extortion are found in the baseline scenario, that is, PRD-controlled governorships. This seems to be consistent with the argument that PRD governments were easy prey for powerful DTOs without federal government backing during Calderón’s presidency, either through killing a disproportionate number of their mayors or, apparently, coercing or capturing them to impose a predatory criminal order without much state restraint.

Leadership Style and Organizational Capacity

This section explores the role of criminal organization on DTO strategies toward the population. Appendix Figure A1 shows the areas of operation of selected DTOs between 2008 and 2010 using the Coscia and Rios (2012) database. The figure also displays if the areas are under monopolistic control or under contestation according to our definitions above.

Conditional on levels of contestation, we expect higher levels of extortion and lower assistance among clustered hierarchy groups than among standard and regional hierarchy groups. To test these hypotheses, we run an individual model for each type of organization (Standard, Regional, or Clustered). The model includes a dummy variable indicating the presence of that particular type of organization interacted with our variable for territorial contestation (Monopoly, Contested, No Cartel).

The coefficients in Table 7 show the estimated differences in DTO extortion and assistance between contested and monopolistic areas of operation of the corresponding organization. Each row of coefficients correspond to a different model. The table also reports coefficients for models where each DTO is treated separately.

OPEN IN VIEWER

Table 7.

Estimates on Narco Extortion and Help by Type of DTO Hierarchy.

	Narco extortion	Narco help
Sensitive item
Standard × Contested	0.09	−0.14
	(0.06)	(0.16)
Sinaloa × Contested	0.10	−0.11
	(0.06)	(0.15)
Juarez × Contested a	−0.17	−0.22
	(0.11)	(0.16)
Regional × Contested
Golfo × Contested	0.05	0.05
	(0.19)	(0.32)
Clustered × Contested	0.14**	−0.20***
	(0.06)	(0.07)
Zetas × Contested	0.22*	−0.38***
	(0.11)	(0.10)
Familia × Contested	0.35*	0.01
	(0.18)	(0.16)
Tijuana × Contested	0.15***	−0.11
	(0.06)	(0.11)
Beltran Leyva × Contested	0.03	−0.24***
	(0.08)	(0.07)

The rows show estimated coefficients for the sensitive item of two list experiments included in the Survey on Public Safety and Governance in Mexico in 2011. There were two questions (Card 1 and Card 2) for which respondents were randomly assigned into one control and two treatment groups (see Table 1). Groups are classified according to their hierarchical organization (Standard, Regional, or Clustered) drawing from the UN typology of crime organizations (UNODC, 2002). Each row of coefficients correspond to a different model. A full coefficient table is available upon request. All models include individual-level characteristics. Robust standard errors by municipality in parentheses. DTO = drug trafficking organizations.

a.

All municipalities with dominant presence of the Juárez Cartel were contested during the period of study.

*

p < .1. **p < .05. *** p < .01.

The results are broadly consistent with our theoretical expectations. Clustered criminal groups exhibit significantly higher levels of extortion under conditions of territorial contestation, as reflected in the negative and statistically significant coefficient for the interaction of these variables. In terms of assistance, we also find significantly lower levels provided by clustered DTOs. We do not find evidence that standard and regional hierarchy groups engage in significantly more extortion or less assistance under conditions of high territorial contestation. We should emphasize that our findings probably underestimate the levels of extortion for Los Zetas and the Gulf DTOs: As mentioned above, we were not able to collect surveys in Tamaulipas, where these two groups have fought a bloody turf war.

In combination with the evidence above that capo removals matter, this evidence suggests that DTO organization has important effects on interactions with the community.

Robustness Tests and External Validity

Our empirical approach uses a particular set of variables and definitions for violence and territorial contestation. In this section, we use alternative measures for the dependent and some of the independent variables as well as other data sources as robustness checks. Tables 8 and 9 show the estimated coefficients of alternative models for narco extortion and community assistance.

OPEN IN VIEWER

Table 8.

Robustness Checks and Out-of-Sample Validity: Narco Extortion.

Source	Survey					SNSP	ENVIPE
Dependent variable	List experiment					Extortion rate	Extortion prevalence	Extortion in neighborhood
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
High violence
× Monopoly	−0.13**	−0.12**		—	—	−8.22***	—	—
	(0.05)	(0.05)				(2.31)
× No Cartel	0.13	0.24*		—	0.06	—	—	—
	(0.15)	(0.15)			(0.07)
Medium violence			−0.12
			(0.08)
Low violence			−0.21**
			(0.10)
× Contested	−0.09	−0.14		−0.14**	−0.22***	−4.21*	−14.90	−65.23*
	(0.24)	(0.19)		(0.06)	(0.04)	(2.38)	(13.06)	(39.49)
× Monopoly	−0.19	−0.25*		−0.26*	0.10	−5.59**	−24.16*	−76.97**
	(0.13)	(0.13)		(0.14)	(0.11)	(2.31)	(12.82)	(38.86)
× No Cartel	0.10	−0.02		−0.13	−0.02	−6.77***	−26.44**	−91.14**
	(0.12)	(0.12)		(0.11)	(0.09)	(2.30)	(13.07)	(39.02)
Observations	1,778	1,778	1,778	1,778	1,778	2,456	896	896
R 2	.10	.10	.10	.10	.11	.16	.06	.11
Variable for violence	Executions rate, 2008-2011	Homicides by firearms, 2008-2011	Violence in locality, 2011	Executions total, Jan-May 2011
Variable for contestation	Coscia & Rios, 2008-2010		None	Coscia & Rios, 2010	Osorio, 2008-2010	Coscia & Rios, 2008-2010
Unit of analysis	Individual					Municipality

Columns 1 to 5 show estimated coefficients for the sensitive item of a list experiment included in the Survey on Public Safety and Governance in Mexico in 2011. Respondents were randomly assigned into one control and two treatment groups (see Table 1). Data on extortion complaints in column 6 come from the Secretariado Ejecutivo del Sistema Nacional de Seguridad Pública (SNSP, 2011). Data on victimization in columns 7 and 8 come from the Encuesta Nacional de Victimización y Percepción sobre Seguridad Pública (ENVIPE; Instituto Nacional de Estadística y Geografía, 2011). Data on executions come from the “Deaths Presumably Related to Criminal Rivalry” data set published by the Mexican government (SEGOB, 2011). Data on homicides by firearms are based on death certificates (SINAIS, 2011). A full coefficient table is available upon request. All models include regional socioeconomic characteristics. Robust standard errors in parentheses.

*

p < .1. p < .05. ***p < .01.

OPEN IN VIEWER

Table 9.

Robustness Checks: Help From DTOs.

Source	Survey
Dependent variable	List experiment
	(1)	(2)	(3)	(4)	(5)
High Violence
× Monopoly	0.21***	0.19***		—	—
	(0.08)	(0.07)
× No Cartel	0.18	0.25		—	−0.25**
	(0.16)	(0.17)			(0.12)
Medium violence			0.14**
			(0.07)
Low violence			0.13
			(0.20)
× Contested	0.57***	0.36*		0.09	0.02
	(0.21)	(0.21)		(0.07)	(0.06)
× Monopoly	0.24*	0.30*		0.23	0.03
	(0.13)	(0.15)		(0.15)	(0.08)
× No Cartel	0.29**	0.26*		0.03	0.10
	(0.14)	(0.15)		(0.14)	(0.13)
Observations	1,782	1,782	1,782	1,782	1,782
R 2	.08	.08	.08	.07	.08
Variable for violence	Executions rate, 2008-2011	Homicides by firearms, 2008-2011	Violence in locality, 2011
Variable for contestation	Coscia & Rios, 2008-2010		None	Coscia & Rios, 2010	Osorio, 2008-2010

Columns 1 to 5 show estimated coefficients for the sensitive item of a list experiment included in the Survey on Public Safety and Governance in Mexico in 2011. Respondents were randomly assigned into one control and two treatment groups (see Table 1). Data on executions come from the “Deaths Presumably Related to Criminal Rivalry” data set published by the Mexican government (SEGOB, 2011). Data on homicides by firearms are based on death certificates (SINAIS, 2011). A full coefficient table is available upon request. All models include regional socioeconomic characteristics. Robust standard errors in parentheses. DTO = drug trafficking organizations.

*

p < .1. **p < .05. ***p < .01.

First, we use alternative measures for violence. Columns 1 in both tables replicate our analysis using the tertiles of the average annual rate of drug-related executions from 2008 to May 2011 as our measure for violence. Although we believe that the total number of executions is a more direct measure of the intensity of the inter-conflict, we present this model to account for population size. As before, we use contested municipalities with high levels of violence (last tertile) as the baseline scenario. We observe once again that DTO extortion is lower and help is more prevalent in places with monopolistic control.

Some critics of government data on drug-related deaths argue that it suffers bias from underestimation or misclassification of cases. The database only includes cases with an open investigation from the federal or local public attorney offices and known by the federal authorities, and their inclusion in the database might change if they no longer meet the established criteria (SEGOB, 2011). Therefore, in Model 2 we use homicide rates by firearms of males between 15 and 39 years old (SINAIS, 2011). This series has some advantages: (a) homicide data come directly from death certificates, reducing potential underestimation biases; (b) homicides in this group of age and gender best reflect the geo-temporal variance of reported executions between criminal groups (Calderón et al., 2015); and (c) homicides by firearms are directly related to drug-related violence, since most of the executions occur in this modality. Using this series as our measure of violence, we found very similar results to those in our baseline model.

The models in column 3 do not rely on external data for our measures of violence and contestation at the municipal level. Instead, we use a measure of local violence directly taken from our survey. We asked our respondents, “From a scale from 0 to 10, where 0 means no violence and 10 means the worst level of violence, how would you grade violence in your locality?” We estimated the average grade at the precinct level and defined our categories of violence (low [0-3], medium [3-8], and high [8-10]). This measure could also be conceived as a proxy of territorial contestation, given that it is to be expected that respondents would answer that they observe more violence where DTOs contest a territory. We observe again that extortion is less prevalent and help from narcos more frequent in localities with medium and low violence.

The second strategy in our robustness checks is to use alternative measures of territorial contestation. Our baseline measure defines that a cartel has a dominant presence in a municipality if it has had activity in that place for at least three consecutive years (2008-2010) using the Coscia and Rios (2012) database. We chose the window of 3 years to avoid overestimating the number of cartels due to temporary or single events within a year that do not necessarily reflect DTOs’ control or influence over a municipality. ¹⁹ As a robustness check, we present in column 4 of Tables 8 and 9 our baseline model using the number of DTOs in 2010 (the last year available) as reported in Coscia and Rios (2012). We observe once again that extortion is significantly lower in places of low violence, with the highest reduction in places where a single cartel operates. The coefficient for help is also positive and substantive (but not significant) for this scenario.

An alternative data set on DTO’s area of operation comes from Osorio (2015). The author bases his analysis on online articles from local and national newspapers and government reports and uses a supervised coding protocol for text analysis and automated identification of event data. ²⁰ The database identifies drug-related events by type (executions, confrontations, government interventions, etc.) and actors involved, including criminal groups. Column 5 in both tables presents models using our baseline definition for territorial contestation but using Osorio’s data set on the number of DTOs. Results show that extortion is significantly lower in contested places with low violence versus places with high violence. In this model, however, we did not find significant effects for extortion and help for the monopoly scenario.

It is worth noting that the Coscia and Rios (2012) and Osorio (2015) datasets differ considerably after 2004. The number of municipalities with at least one DTO in the first data set more than doubles the number reported in the second data set. Moreover, there is a group of municipalities in the Osorio (2015) data set that reports no DTO presence but have high levels of executions and homicides, as measured with official statistics. ²¹

Finally, to demonstrate the external validity of our findings, the last three columns in Table 8 present models with alternative measures of extortion at the municipal level using an extended sample of municipalities in Mexico. The model in column 6 uses official data on extortion rates in 2011 as reported by the Mexican Government (SNSP, 2011). The database contains the number of cases of extortion with a pretrial or full investigation as reported by local public prosecutor offices from all municipalities in Mexico. The estimated coefficients confirm our results that extortion rates are considerably higher in contested areas than those under the monopoly and no cartel scenarios. Similarly, coefficients show that violence intensifies the prevalence of this criminal activity in the later scenarios.

Data from reported cases severely underestimates crime incidence because most of the crimes go unreported due to fear, insecurity, and lack of trust in the government. An alternative strategy is to rely on self-reported data from victimization surveys. The last two columns in Table 8 show similar models using questions about extortion included in the 2011 National Victimization Survey as the dependent variable (ENVIPE, 2011). The survey asks respondents if they have been victims of extortion in the last 6 months or if there has been extortion in their neighborhood. The variables are expressed as the number of people reporting extortion per thousand inhabitants using population weights from the victimization survey. Once again, we observe that the worst scenario for civil society is that of two or more criminal groups engaging in violent confrontation for control of the drug market. In fact, the predicted rate of individual extortion for the contested high-violence scenario (50.3) is at least 20% greater than those for the monopoly (41.1) and no-cartel (38.8) low-violence scenarios. We observe similar results when the prevalence of extortion in the neighborhood is used as the dependent variable.

The Need for List Experiments

Our results underline an important and broader point beyond this particular paper: Crime appears to be systematically underreported in Mexico. Comparing crime incidents reported to local public attorney offices (SNSP) and victimization surveys (ENVIPE) in 2011, we estimate that underreporting bias is as high as 87.2% for common crimes and 93.2% for extortion. Even victimization surveys can suffer from underreporting if citizens feel unsafe revealing predatory or abusive behavior by criminal groups or the police.

Our estimates for criminal extortion in 2011 reveal that at least 12% of the population living in contested territories have been extorted by criminal groups. This compares to the 6% of respondents in contested areas giving the same answer when asked directly in the national victimization survey. Our estimate is only closer to the 17% of the population reporting extortion in their neighborhoods throughout DTO-contested places. This comparison of extortion rates reveals the advantages of using experimental methodologies like the item count technique to increase truthful responses when measuring criminal incidents.