Understanding the Escalation of Drug Violence in Mexico

Javier Osorio

Understanding the Escalation of Drug Violence in Mexico

Draft: May, 2011

UNDERSTANDING THE ESCALATION OF

DRUG VIOLENCE IN MEXICO

Javier Osorio

Department of Political Science

University of Notre Dame

Paper prepared for the

FifthAnnualConferenceof the

International Society for the Study of Drug Policy

May 23-24, 2011

Utrecht, Netherlands

ABSTRACT

This research disentangles the dynamics of drug related violence. The central argument holds that the escalation of drug violence in Mexico is characterized by a cumulative and accelerating process driven by strong internal inertias, as well as reciprocal interactions between different dimensions of drug violence. As conflict heats up, internal inertias become more intense and feedback effects accelerate in an increasing pace. The empirical assessment is based on the first large N database of drug related violence in Mexico from 2000 to 2010. This dataset was built using a customized automated coding protocol to identify violent events from government press releases. Statistical analysis of multivariate time series reveals that the dynamics of drug related violence evolve in a rapid and non-linear way, thus generating the intensification and acceleration of inertias and feedback effects between violent events, seizures and arrests.

Draft version: May 2011

Please address your comments to

INTRODUCTION

Despite of lacking protection against fraud and violence offered by the legal systems and the state, illegal markets rarely present high levels of violence (Reuter 2009). However, as shown by the Mexican war on drugs, when violence occurs it can reach unprecedented levels of intensity. In December 2006, the Mexican President Felipe Calderón launched a full-fledged military campaign against drug-trafficking organizations (DTOs). The punitive strategy triggered an escalation of violence that has generated an estimated death toll of more than 39,000 people in four years and a half. Despite the fact that large scale drug violence is rare, the lethality of the Mexican war on drugs surpasses that of other forms of intense political violence. Drug violence in Mexico is approximately 40 times deadlier than the standard threshold of 1,000 deaths used to define civil wars (Sambanis, 2004) and has already killed about four times more people than the median civil war death toll of 10,500 casualties (Lacina, 2006). What explains such a dramatic escalation of drug related violence in Mexico?

The objective of this paper is to disentangle the dynamics of large scale drug violence in Mexico. The main argument advanced in this research is that the escalation of drug violence is characterized by a cumulative and accelerating process driven by strong inertias, as well as reciprocal interactions between different dimensions of drug related violence. From this perspective, drug violence is not considered a homogenous and static process. It is rather a highly dynamic sequence of events that evolve in a non-linear way. As conflict heats up, inertial trends become more intense and feedback effects accelerate in an increasing pace.

It has already been argued that violence begets violence. However, the objective of this research aims to disentangle the different mechanisms operating in the ongoing wave of large-scale violence of the Mexican war on drugs. To do so, the general concept of drug related violence is disaggregated into three different–yet mutually related–components: events of violence, seizures and arrests. In particular, this research holds that events of violence, confiscation of material goods and apprehensions of members of drug trafficking organizations are highly interconnected processes.

The statistical analysis reveals that as conflict evolves, previous events of violence generate further violence; increasing seizures augment the levels of violence; and increasing arrests reduce violence in the short run, but generate more violence a few days later. The empirical assessment also finds that previous seizures increase the number of material confiscations; and increasing violence decreases the number of seizures in the short run but augments confiscations in the mid-term. In addition, regression results indicate that increasing number of arrests in the past raise the number of arrests in the present; recent seizures also increase apprehensions but as time goes by seizures decrease the number of arrests; finally, increasing levels of violence generate more arrests. In general, the empirical assessment reveals that the dynamics of drug related violence evolve in a rapid way generating the intensification and acceleration of inertias and feedback effects between violent events, seizures and arrests.

In 2009, Peter Reuter stated that “given the prominence of the Mexican drug market homicides as a national problem, it is striking that there is no evidence of systematic data collection about who is killed by whom for what reasons”(Reuter 2009, 283). Mexican authorities systematically deny access to their records on drug violence arguing nationally security concerns and, when they release official counts, their estimates are largely divergent. For example, on August 2010, the General Attorney reported an aggregated death toll of 24,800 people and, just a few days later, the Mexican Intelligence Agency reported a larger count of 28,000 (López, 2010). In addition, three newspapers (Milenio, Reforma and El Universal)independently track drug-related killings but neither their data nor their methodologies are public and, when compared, their counts diverge up to 25%(Shirk, 2010). The lack of systematic data on drug violence prevents a basic diagnostic and the evaluation required in any policy assessment. It also precludes civil society, media, scholars, policy experts and government authorities from engaging in an evidence-based debate about the best way to reduce drug violence.

To overcome this problem, this research presents the first database of drug violence in Mexico providing detailed information on different aspects of drug related violence between 2000 and 2010. The information comes from press releases from four Mexican government agencies and was built using a customized computerized protocol for event coding. The automated coding successfully identified 14,948 events of drug related violence in Mexico between January 1st, 2000 and December 31st, 2010.

This paper is structured in the following way. The first section outlines the literature on the determinants of violent crimes. The second section addresses the empirical and methodological innovations of this research. The third part analyzes the endogenous determinants of different aspects of drug related violence. Then, the fourth section compares the dynamics of drug violence before and after Calderón triggered the war on drugs in Mexico. Finally, the fifth section presents the conclusions.

DETERMINANTS OF VIOLENT CRIME

The Nobel Prize in Economic Sciences, Gary Becker (1968), advanced the thesis that criminal acts result from a rational decisions evaluating the costs and benefits of engaging in criminal activities. Baker’s pioneer research started the economic branch of the literature on criminality which has mainly focused on analyzing the costs of crime imposed by penalties such as apprehension, seizures and capital punishment, and the benefits and opportunity of looting.

Several authors have focused on analyzing the “costs” side of the equation. Early research conducted by Ehrlich (1973; 1975) finds that increasing efficacy of capital punishment is a key determinant for preventing crime. More recent research by Taylor (1978) and Levitt (1996; 1997) find a significant effect of policing and punishment on crime reduction. Although some researchers find and opposite relationship between punishment and crime (see Archer & Gartner, 1984), there is broad agreement that increasing effectiveness of deterrent policies contribute to reduce crime rates.

Economists have also focused on analyzing the “benefits” side of the equation and the structure of incentives motivating an individual to engage in criminal activities. This branch of the literature consistently finds that economic conditions serve as key structural determinants of criminal behavior. In particular, increasing unemployment rates and raising economic disparities have a strong crime-inducing impact (Ehrlich, 1973; Fajnzylber, Lederman, & Loayza, 2000). Youth in poverty are also more likely to be arrested than adults (Freeman, 1992). Among these variables, economic inequality–often measured as the GINI coefficient–is one of the strongest predictors of crime. An astringent robustness check of the link between increasing economic inequality and higher rates of crime conducted by Fajnzylber, Lederman and Loayza (2002) confirms the strength of this relationship between and within countries.

The sociological approach has contributed to the economic literature on crime through the theory of relative deprivation (Runciman 1966; for a critical review of the relative deprivation theory see Stack, 1984). This theory indicates that the feeling of disadvantage and unfairness leads the poor to seek compensation and satisfaction by all means, including committing crimes against both poor and rich. Other authors have also found that the lack of social capital increases the incidence of crime (DiIulio, 1996; Freeman, 1986) and the prevalence of social interaction with criminals increases an individual’s propensity to engage in criminal activities (Case & Katz, 1991; Glaeser & Sacerdote, 1996).

Recent research has found a mismatch between the rapid variations of crime trends and the slow changes of structural economic variables often referred as the “root causes” of crime. This branch of research claims that trends of criminality are determined by endogenous dynamics. Fajnzylber, Lederman and Loayza (2000) use cross-national data to analyze crime trends and argue that in addition to economic variables, criminal inertia–understood as the persistence of crime over time–is a key determinant to understand variation of crime trends. Caulkins, Feichtinger, & Veliov (2007) develop a formal model to argue that cycles of violence are driven by internal dynamics of violence. According to this perspective, initial events of violence generate further violence until the conflict decimates the population of violent offenders, thus generating a decline in violent crime. In addition, Peter Reuter (2009) argues that the escalation of drug violence is caused by internal micro-mechanisms of enforcement, competition, internal mobility and discipline in high-level drug markets.

This paper contributes to this vein of research emphasizing the endogenous trends of violence. In doing so, this research overcomes key limitations of previous studies. In contrast to aggregated cross-national data used by Fajnzylber, Lederman and Loayza (2000), this research provides highly detailed data on a daily basis for a single country for a long period of time. It also provides empirical evidence to inform the intuitions of the formal model developed by Caulkins et al. (2007). In addition, it contributes to provide a large N rigorous analysis to complement the insightful intuitions suggested by Reuter (2009). The central argument of this research holds that the escalation of drug related violence is characterized by a cumulative and accelerating process driven by strong inertias and reciprocal interactions between different dimensions of drug related violence.

EMPIRICAL CONTRIBUTION

As mentioned before, one of the key challenges of studying the dynamics of drug related violence is the lack of systematic and reliable data. The core evidence for this research comes from a database recording daily events of drug-related violence in Mexico between January 1st, 2000 and December 31st, 2010. The time frame allows variation of the dynamics of conflict as it considers violence prior to the onset of the war on drugs in December 2006, and afterwards. The database comprises the frequency of three different dimensions of drug related violence: violent events, seizures and arrests.

The database draws information from 7,913 press releases issued between January 1st, 2000 and December 31st, 2010 by four key Mexican government agencies including the Mexican Army (Secretaría de la Defensa Naional, SEDENA), the Navy (Secretaría de Marina-Armada de México, SEMAR), the Police (Secretaría de Seguridad Pública, SSP) and the Office of the General Attorney (Procuraduría General de la República, PGR) at both federal and state levels. The vast quantity of information considered in this research represents a methodological challenge. Using human coders to manually code all this information would be cost prohibitive. To overcome this challenge, this research relies on computerized textual annotation to create the database on drug violence using automated event coding.

Scholars studying political conflict have relied on machine coding to analyze violent events based on newspaper reports. This research strategy has been successfully implemented by scholars to study international conflict (e.g. King & Lowe, 2003; Schrodt et al., 2010). This research improves on previous efforts by focusing on sub-national actors and, for the first time, adapting the protocol to codify text in Spanish. I rely on Text Analysis by Augmented Replacement Instructions (TABARI) (Schrodt, 2009), a software package that allows the computer to identify, categorize and codify event data from text documents. TABARI codes a data event by identifying three key pieces of information: the actor undertaking the action, the target receiving the action and the action itself. Automated coding has several advantages over manual coding as it reduces costs and time, eliminates coder's fatigue and bias, coding rules are transparent, and databases are easily reproducible (Shellman, 2008). For this project, I developed specific dictionaries for actors and actions in Spanish to identify key aspects of drug violence. Based on these dictionaries, TABARI “reads” specific words in documents and transform textual information into numeric codes, which are then stored in a database format to be analyzed using statistical methods.

In order to illustrate how TABARI works, consider the header and abstract of a real news report (Cedillo, 2010) in a document hypothetically named “000001”:

Monterrey, Nuevo Leon, Friday November 12, 2010. Soldiers kill a gunman in a confrontation in NL. A group of armed men ran across a convoy of the Immediate Reaction Group, which unleashed a chase and a shootout in the streets of Monterrey.

Using hypothetical dictionaries assigning numeric codes for the place, date, actors and actions, TABARI would “read” the news report and codify the underlined text in the following way:

Monterrey19122 Nuevo Leon19 Friday November11 1212 2010.2010 Soldiers3001 kill562 a gunman2068 in a confrontation715 in NL.19 A group of armed men2073 ran across437 a convoy of the Immediate Reaction Group3175 which unleashed a chase553 and a shootout793 in the streets of Monterrey19122

Then, this information would be transformed into numeric codes in a database format as presented in Table 1. In this way, TABARI extracts detailed information disentangling the complexities of violence. For the particular purpose of this paper, data of violent events are aggregated to consider only three general kinds of actions: violent events, seizures and arrests.