The impact of media coverage on crime rates andcrime style

Viridiana Rıos∗

November 13, 2017

Abstract

Contributing to a more nuanced and detailed understanding of the effects thatmedia contents may have on crime, this paper presents compelling empirical ev-idence that media portrayals of crime influences the probability of using similarcrime styles, not of engaging in more criminal actions. We tracked 31,676 homi-cides carried out by drug cartels operating at the US-Mexico border, its stylisticcharacteristics, and the detail with which the media covered each of them. Withthis uniquely detailed and large data, we estimated reaction functions to determinethe Granger causality between the quality of media coverage, crime rates and crimestyle. Our estimates show that when the media covers crimes in larger detail, crim-inals react by increasing the certain criminal styles such as leaving credit-takingbanners or exposing the tortured bodies of their victims. Up to 53% of the variancein criminal styles can be traced back to changes in the quality of media coverage.These results indicate that media coverage does not encourage criminal acts, butmay strongly influence the modality under which crimes are conducted.

∗Assistant Professor, Purdue University; viririos@purdue.edu

1

There is a widespread notion that detailed media coverage of criminal acts creates models

of crime that function as direct causes of criminality, or as crime “triggers”, for at-risk

populations (Surette, 2013). For example, evidence suggests that at least 30 separate

occurrences of gun violence were inspired by the Russian Roulette scene from the movie

The Deer Hunter (Gunter, 2008). Additionally, the shooter in The University of Virginia

Tech massacre was initially thought to be copying events from an action movie (Nizza,

2013). Lastly, the rape of a nine-year-old with a bottle was connected to a scene from the

movie “Born Innocent”. (Gunter, 2008).

As a result, it is not surprising to see the implementation of public policies that aim

to reduce crime by controlling media contents, particularly in developing countries. As

a result, it is not surprising to see the implementation of public policies that aim to

reduce crime by controlling media contents, particularly in developing countries. In 2010,

for example, Zimbabwe’s government attempted to regulate international news outlets

and prevent negative media portrayals that might encourage violent uprisings against it

(Surette, 2013). Likewise, during the nineties, Colombia convinced 35 top communication

outlets to sign the “Discretion Agreement” and reduce the coverage of violent paramilitary

groups by appealing to national security and violence reduction. Mexico followed a similar

path in 2011, when the Federal government proposed an agreement (“Agreement on Media

Coverage of Violence”) with 700 media outlets to reduce coverage of drug-related crime

(Philip and Berruecos, 2012).

Despite all this, much research is still needed to prove whether the relationship between

media coverage and crime rates is empirically robust and unidirectional (Surette, 2013).

The unsettled debate about whether media coverage promotes criminal behavior or crimi-

2

nal ideation is one of the most lively areas of criminality research (Surette, 2013; Ferguson

et al., 2008), and it is also becoming important to the study of political violence.

While some argue that exposure to media coverage of crime is itself criminogenic, mean-

ing likely to cause crime, others see the media as an instructional guide for individuals

that were going to commit crimes anyway. The first group has developed the General

Aggression Model, which argues that exposure increases crime by “triggering” aggres-

sive behavior and by assisting in the development of callous dispositions (Bushman and

Anderson, 2002; Anderson et al., 2007). The second group has a more favorable view,

arguing that exposure merely provides criminals with inspiration, it influences the style

of the crime but does not cause it. In other words, media coverage of crime here acts as

a “rudder”; although it can change the direction the boat is going, it is not the source of

propulsion (Ferguson et al., 2008; Surette, 2013).

This paper examines whether media coverage of crime functions as a crime “trigger”, or

as a crime “rudder”, by testing both models. To accomplish this, we rely on a unique

database that records the violent criminal activities of drug cartels operating at the US-

Mexico border, and the stylistic characteristics of their crimes. Specifically, we identify

the rates of drug-related homicides per week in more than 2,500 counties/municipalities.

We also identify whether these murders were accompanied by public banners, known as

“narco-messages”, a stylistic feature that is sometimes used by drug cartels (Campbell

and Hansen, 2014; Martin, 2012; Atuesta, 2017). Criminal activities, especially homicides,

are typically understood as clandestine in nature, however, we find counterintuitive evi-

dence that the use of narco-messages occurs the most when the press has covered similar

modalities in the weeks before.

3

Our results show that “criminogenic styles”, such as using credit-taking banners and

exposing bodies/dismemberment in crime scenes, react to media coverage. These “crim-

inogenic styles” are such in that they do cause crimes to take on certain features, but

do not cause the crime itself. Specifically, we provide evidence of Granger causality from

media coverage to crime style, but not to crime rates. Furthermore, we provide evidence

that crime style reacts to media coverage unidirectionally, rather than as a cycle. Our

results show that as much as 53% of the error in the forecast of credit-taking crime styles

is explained by changes in the quality of media coverage. Moreover, we find that when the

media covers crimes in larger detail, criminals increase the use of credit-taking and body

exposure styles in their crimes. These findings provide valuable insight into one of the

most controversial areas of criminality research: the relationship between crime coverage

in the media and criminal behavior.

The rest of the paper proceeds as follows. The first section discusses the existing literature

about media coverage and crime. The second section presents original data and describes

empirical tests. The third and fourth sections present results and robustness tests. Both

sections show that media coverage of crime functions as a crime “rudder”, but not as

crime “trigger”, particularly when media is detailed in covering the style of crimes. We

conclude by discussing how the paper contributes to the literature, and by suggesting

possible avenues for future research.

Media coverage and crime

A large body of criminology literature has tried to determine whether exposure to crim-

inogenic media content “triggers” delinquency and aggression, or just acts as a “rudder”,

4

by providing instructional models of style for individuals that already had criminal inten-

tions. The debate in the literature as to which understanding best describes the role of

the media has, overall, lacked conclusive results.

In support of the “trigger” theory, many studies have shown some significant effects

between media consumption and crime (Bushman and Anderson, 2002; Anderson and

Carnagey, 2004; Anderson and Bushman, 2001; Doley et al., 2013; Chadee et al., 2017).

Moreover, the relation between media and criminal conduct has also been attributed to

the constant portray of violence that might result in ideas for potential criminal minds

(Surette, 2013; Surette and Gardiner-Bess, 2013).

Other scholars have shown that violent media content can drive individuals that already

have violent thoughts and behavior (Surette and Gardiner-Bess, 2013; Ivory and Kaestle,

2013; Ferguson, 2015).1 (Gunter, 2008) provides evidence that shows how criminals have

modeled their crimes from media content. Moreover, this relation is stronger for people

that have similarities with criminal character(Huesmann, 1986; Surette, 2013). Other

scholars have found evidence of increased hostility and engagement in physical fights

among individuals exposed to violent video games (Anderson et al., 2007, 2010), and

violent song lyrics (Anderson et al., 2003).

However, there is also supporting evidence for the “rudder” theory, as many studies have

shown a lack of correlation, or lack of causality, between media content and crime (Fergu-

son et al., 2012, 2009, 2008; Ferguson and Kilburn, 2009; Freedman, 1984; Savage, 2004;

Freedman, 2008; Ferguson, 2002; Huesmann and Taylor, 2006; Gunter, 2008; Murray,

2008). Even if a significant and causal effect could be found, some academics claim that

1For example, violent media tends to be watched the most by those with a relatively higher tastefor violence (Brown and Hamilton-Giachritsis, 2005; Huesmann and Taylor, 2006; Coyne, 2007; Savage,2008; Savage and Yancey, 2008; Bushman and Huesmann, 2014; Helfgott, 2015).

5

it could not be large enough to change aggregated criminal rates in notable ways (Surette

and Gardiner-Bess, 2013; Elson and Ferguson, 2014; DeCamp and Ferguson, 2017). This

literature refuses to assume viewers are “passive receptacles of learning”, and instead

conceives individuals as “active shapers and processors of media culture” (Elson and Fer-

guson, 2014). As a result, they argue that the interaction between media and behavior is

shaped by media users, not media content (Elson and Ferguson, 2014).

Furthermore, several studies have shown a stronger and more solid association between

media coverage of crime and the stylistic decisions of criminals. To these scholars, the

media is more a source of crime techniques than a source of crime motivation. In general,

this literature argues that the media’s criminogenic influence concentrates in populations

with pre-existing susceptibility to criminality. More specifically, it shows that environ-

mental factors and psychological predispositions are behind disruptive behavior disorders,

rather than just exposure to media or video games. Based on this research, courts have

systematically ruled against attributing any significant impact to the media (Ferguson

and Dyck, 2012; Ferguson et al., 2012; Grimes et al., 2008; Ferguson and Savage, 2012).

The work of Surette (2014), for example, analyzed anonymous surveys of 574 male and

female correctional inmates. Surette concluded that the media influenced the style of

crimes by providing instructional models to predisposed individuals, yet it did not prompt

crime per se (Surette, 2014). He coined the term “ruddering” to illustrate how media

portrayals of criminality shape the stylistic form the crime takes on, instead of acting as

a direct catalyst for the crime itself (Surette, 2013).

This debate stands unresolved, mostly because of the difficulties involved in finding ap-

propriate ways to measure and correlate the role of the media as a “trigger” or “rud-

6

der” for crime. The first difficulty is gathering data on “copycat” criminals (Surette,

2013). Most studies rely on evidence gathered from post-crime interviews with offenders

(Surette, 2013). This means that all these studies are subject to perception biases and

self-justification. Offenders may rationalize their criminal behavior by attributing it to

the impact of the media, or to other third parties, because they find it comforting to

avoid the cost of decision making. To mitigate this problem, studies have tried to identify

copycat criminals based on the facts of the case (Surette, 2013). Despite this, the lack

of a centralized and comparable source of information that describes crimes occurring at

different times and in different police jurisdictions is a serious challenge. Furthermore,

even if two crimes appear to be similar, scholars cannot be certain that they were inspired

by the same source (Stack, 2000; Coleman, 2004; Surette, 2015).

A second difficulty is that media coverage and criminal behavior may enforce one another.

Front-pages prioritize content with higher audience ratings and editors approval (Reiner

and Newburn, 2007; Fink and Schudson, 2014; Coddington, 2014). Criminal groups may

perpetuate noticeable violent actions to gain more visibility to intimidate their enemies.

Indeed, there is much evidence that media coverage is proactively sought by some criminals

who wish to advance their criminal objectives by receiving attention and inducing fear.

Terrorists are the most common type of such criminal (Schmid and De Graaf, 1982;

Hoffman et al., 2010; Kearns et al., 2014; Abrahms and Conrad, 2017; Brown, 2017; Min,

2013), followed by other criminal organizations (Gambetta, 2009; Duran-Martınez, 2015;

Mendoza Rockwell, 2016; Atuesta, 2017; Martin, 2012; Campbell and Hansen, 2014).

Furthermore, some research has found that criminals tend to value media coverage the

most when there is intergroup competition (Hoffman, 2010; Hoffman et al., 2010; Martin,

2012), when criminal groups identify with ethnic separatist groups instead of groups with

7

broad goals (Min, 2013), when criminal groups operate in democratic governments (Min,

2013), when criminal groups are non-religious (Wright, 2009), when attacks are less lethal

for civilians (Abrahms and Conrad, 2017), when criminals are trying to perform particu-

larly dramatic attacks (Wright, 2009), when criminal groups are “leaderless resistance” or

not hierarchical (Brown and Hamilton-Giachritsis, 2005), and when terrorist groups want

to claim credit (Wright, 2009).

In sum, testing for the criminogenic effects of media exposure is problematic because of

the dynamic process between the media, which may be driven by sensationalism, and

criminals, who may be benefited by the direct/indirect reputational gains of media cover-

age. Identifying which of these two reactions happens first is a methodological challenge,

particularly when –as has been the case– academics lack data on the strategic choices of

criminals that has been observed over long-enough time periods to provide material for

statistical inference.

By directly tackling this issue, this paper contributes to studying how media immersion

and narrative influences social behavior (Slater, 2007; Zhang et al., 2010; Valenzuela et al.,

2012), to identifying the effects of violent media contents on criminal behavior (Surette,

2013), and to discerning how media, a crucial element of democracy, can tackle governance

(McCarthy-Latimer and Kendrick Jr, 2016).

Data

Our study uses drug cartels operating at the US-Mexico border as our subjects because

of their notably different stylistic patterns (Martin, 2012), and because their crime rates

8

vary over time and geographically (Coscia and Rios, 2012). Homicides perpetrated by

drug cartels vary in style, stylistic differences can be found in the use of specific types of

weapons, victims’ characteristics, types of torture, the display of dismembered or decapi-

tated bodies at the crime scene, and in the use of other intentionally public displays, such

as banners, that explain their motivations or intents (Shirk and Wallman, 2015; Duran-

Martınez, 2015). Drug cartel-related crimes can be observed across 1,068 municipalities,

from border cities like Rıo Bravo and Tamaulipas that had only 18 homicides from 2008

to 2010, to places like Juarez that endured 6,300 homicides during the same period. A

burgeoning literature seeks to understand the levels of violence in Mexico (Dell, 2015;

Rıos, 2015; Shirk and Wallman, 2015; Osorio, 2015; Calderon et al., 2015; Trejo and Ley,

2017), but research analyzing the dynamics within the conflict, such as communication

strategies, has not kept pace.

Mexican drug cartels are highly relevant as case studies for the interaction of crime and

media. It has been strongly questioned if media outlets should exhibit the violent activi-

ties of drug cartels. While some argue that media’s priority is to show to the public the

truth, other arguments rely on the fact that violent media coverage inspire new forms

of violence and serves as a channel for criminal organizations to gain visibility and dis-

seminate their messages (Reyes, 2011). Despite this controversy, the Mexican National

Institute of Information Access has decided to reserve the right to publish the records

of drug cartel-related crime for up to 12 years, arguing that allowing publication sooner

would also “allow criminal groups to generate instability and insecurity among the Mex-

ican population” (Martin, 2012).

Our research was made possible due to the existence of two unique datasets that we

collected to describe crime rates and crime styles among drug cartels operating at the

9

US-Mexico border. First, to measure “Crime Rates”, we obtained official registries of

homicides attributed to drug cartels by Mexican intelligence and security institutions

(SNSP, 2011). During a short sui generis period (2007-2011), Mexican authorities allowed

public access to a database that recorded whether a homicide was drug cartel-related. This

gave us the unique opportunity to identify drug cartel-related violence in a georeferenced

panel. To be considered drug cartel-related, a homicide needed to meet six criteria. These

are (i) use of high-caliber firearms, (ii) signs of torture or severe lesions in victims, (iii)

bodies found at the crime scene or in a vehicle, (iv) victims were taped, wrapped, or

gagged, (v) murders happened in a prison and involved criminal organizations, and (vi)

if one of several “special circumstance” occurred. Among these “special circumstances”

are whether the victim was abducted prior to assassination (known as a “levanton” in

Spanish), ambushed or chased, whether the victim was an alleged member of a criminal

organization, and whether a criminal organization publicly claimed responsibility for the

murder (Molzahn et al., 2012). In the period from 2008 to 2010, we identified 31,676 drug

cartel-related homicides across 1,112counties/municipalities. Overall, these homicides

represent 95% of all homicides perpetrated during those years and in those areas.

Second, to measure “Criminogenic Style”, we constructed a database that identifies two

discretionary stylistic features that sometimes accompany drug cartel-related homicides:

credit-taking banners and bodies/dismemberment put on display. We matched official

registries of homicides with our original data set of about 1,800 credit-taking banners

displayed publicly by drug cartel organizations during 2007-2010. These banners, also

known as “narco-messages” (Campbell and Hansen, 2014; Martin, 2012; Atuesta, 2017;

Mendoza Rockwell, 2016; Duran-Martınez, 2015), are used by drug cartels to take-credit

for their criminal actions, or to clarify their degree of responsibility for them. Narco-

10

messages may also be displayed to intimidate potential victims, to communicate with

local citizens, or to give instructions to investigators, policemen, or journalists (Martin,

2012). The content of the banners may be a threatening maxim like “you cannot be on

good terms with both God and the Devil”, or an appeal to the public, like the claim that

“this [battle for turf] is for the good of all”, or like the banner dedicated to “the brave,

noble, and loyal people” wishing them “Merry Christmas, ho, ho, ho”. Narco-messages

are a creepy stylistic innovation that stretches the boundaries of traditional graffiti with

its bizarre mimicry of the formality of Mexican political campaign propaganda (Salopek,

2011).

Our database allowed us to identify the share of drug cartel murders that were accom-

panied by a credit-taking banner, we classified this as a dummy called “Credit-taking

Style”.

The use of credit-taking messages in Mexico is fascinating in its own right, but this is not

the only example of such phenomena. We have evidence of similar criminal messaging

occurring around the world, in places such as Colombia, Southern Italy, and in numerous

U.S. cities (Cowell, 1992; Carbray, 2002; Martin, 2012; Ortiz, 2013). Criminal messaging is

also a major part of terrorism, which has been described as violent propaganda or violent

communication (e.g., (Schmid and De Graaf, 1982)). A line of research has examined why

terrorist groups sometimes claim their attacks (Wright, 2009), among reasons for doing

so are intergroup competition (Hoffman et al., 2010), communication within the group

(Brown and Hamilton-Giachritsis, 2005), and operation in democratic countries (Min,

2013). However, this literature also established that there are reasons militant groups

do not seek credit. Such reasons could be that the attack was especially heinous, or the

desire to avoid a bad reputation (Hoffman, 2010; Abrahms and Conrad, 2017).

11

Interestingly, and specific to the case of US-Mexico border drug cartels, we were also

able to identify the share of credit-taking banners that were combined with another crim-

inogenic style: the display of bodies or dismemberment (typically beheadings) (Duran-

Martınez, 2015). The stylistic display of human heads or bodies is worth analysis be-

cause it helps us understand why the appearance of the crime scene is important to

these kinds of criminals and how they use violence as a narrative. Narco-messages and

bodies/dismemberment are typically displayed openly in high-traffic public areas where

authorities, citizens, and journalists can easily find them. They are often found on roads,

hanging from bridges, or dispersed around various parts of a city (Lantz, 2016). Out of

the total of 1,800 narco-messages, 51% were accompanied by a body, or part of one. We

created a dummy for homicides with this characteristic and called it “Body Exposure

Style”.

We also developed two ways to measure the extent and the quality of the media cover-

age of criminogenic contents. To create the variable “Media Coverage”, we identified the

amount of drug cartel-related homicides that were covered by the press. Beginning in

2007, news-media organizations closely followed organized crime-related homicides, keep-

ing count independent of government sources. The most comprehensive coverage was done

by Reforma, a major national newspaper with a large pool of correspondents, but other

national newspapers like Milenio and El Universal also had their own independent counts

(Molzahn et al., 2012). The “Justice in Mexico Program” www.justiceinmexico.org has

also gathered statistics regarding media coverage of drug-related homicides from 2006 to

2017 (Reforma, 2017; Milenio, 2017). Out of the 32,199 homicides officially classified as

perpetrated by drug cartels, 23,737 were covered by the press. Coverage is very different

across time, and geography. For example, although the press covered 76% of the homi-

12

cides in 2010, in 2007 it only covered 63%. Also, in Oaxaca, 26% of drug cartel-related

homicides were not covered, meanwhile, in Queretaro were all were covered by the press.

It is important to note that Mexican news outlets and the Mexican government used a

similar methodology to classify a homicide as drug cartel-related (Milenio, 2017).

Finally, to identify the quality of media coverage, we measured the degree of detail with

which each of the 1,800 narco-messages was reported. We classified coverage as “detailed”

when the specific content of the narco-message was reported. This means that messages

like “Esto les va a pasar a los que no respeten a los grandes jefes de La Familia” (this

will happen to those that don’t respect La Familia’s great chiefs) could be read in their

entirety in the newspaper (El Sol de Toluca 2009). Coverage classified as “non-detailed”

was when the reporter only hinted at the existence of a narco-message without explicitly

quoting its content. Out of our total sample, 74% of the messages were detailed. The share

of detailed messages over total messages is represented by the variable “Media Coverage

Quality”.

Our final dataset is thus a weekly-frequency panel with 5,408 observations over a period

of 508 weeks between 2007 and 2010. 2. We exploit state-level variations in measures of

the extent and quality of media coverage, and in measures of the relative frequency of the

two criminogenic styles: narco-messages and body/dismemberment displays. Descriptive

statistics of our measures of media coverage, quality of media coverage, and criminogenic

styles are shown in Table 1.

(Table 1 about here)

2We use weeks as our frequency measure because media coverage measures were only available perweek (Reforma, 2017; Milenio, 2017)

13

On average the media fails to cover 1.28 homicides per week, per state. The lowest

recorded coverage was during the 13th week of 2010 when, in the state of Chihuahua,

the media failed to report on 77 drug related homicides. There are instances in which

media covers more homicides than the official count like during the 43th week of 2010

for the state of Guerrero (61 more homicides). This happens when mass murder graves

are found. Official records generally avoid these records as they consider them just part

of previous “missing persons” statistics. Of the total events covered by the media, on

average only 10.8% are reported in detail. Drug cartels take credit on average for 2.7%

of their homicides. There are cases in which credit taking is up to 250% larger than

homicides because many drug cartels take credit for a single action. Finally, on average

7.9% of credit-taking style homicides were accompanied with the display of bodies or

dismemberment.

Empirical test

To address possible bidirectionalities between media coverage and crime, we rely on re-

action functions and Granger causality tests as our main empirical specifications. These

methods are an efficient way to show co-movements of multiple time series. A large em-

pirical literature supports the use of Granger causality tests (Granger, 1969) for testing

cycles (Jaeger and Paserman, 2008; Huang et al., 2008; Gambacorta et al., 2014) and as a

relevant tool for crime studies with time series analysis (Saridakis, 2004; Ramirez, 2013;

Hsu and McDowall, 2017).

Our panel vector autoregression consists of p lags of each variable in the equation, in this

14

case empirical reaction functions for Criminogenic style (Credit-taking or Body exposure

style) and Media coverage (Media coverage or Media coverage quality):

xit = β0 +5∑

j=1

β1jxit−j +5∑

j=1

β2jzit−j + exit(1)

zit = β0 +5∑

j=1

β1jzit−j +5∑

j=1

β2jxit−j + ezit (2)

where the β1’s and β2’s are matrices of coefficients for each state i, on time t. The terms

z (for 1) and x (for 2) are vectors of exogenous variables that may shift the reaction

function up or down, and ex and ez are the vector error terms. These equations can also

be expressed in matrix form, letting Yit = xitzit and eit:

A(L)Yt = eit (3)

where eit are structural shocks which are, by definition, uncorrelated with each other, and

A(L) is a matrix lag polynomial of order A(L) = I −A1L1−A2L

2 + · · ·+ApLp where A1

is:

A1 =

βx1,1 βx

1,2

βz1,1 βz

1,2

(4)

Note that efficiency and consistency is achieved because all variables in the equations

15

are lagged (t − 1 or earlier), while the error term corresponds to time t, and because all

variables on the right side of the equations are the same.

This specification is advised to test whether two variables can be characterized as a

cycle, or as a unidirectional effect. It fits our purposes because we want to determine if

media coverage is reacting to past crime rates or crime styles, not to solve for a dynamic

equilibrium.

Every panel vector autoregression model is accompanied by a Granger test, a joint signif-

icance test to check the significance of the coefficient of our lagged impulse and response

variables. The null hypothesis that x does not Granger-cause y is accepted if and only if

no lagged values of x are individually significant according to their t-statistics, provided

that collectively they add explanatory power to the regression according to an F-test. The

null hypothesis of the F-test is that there is not explanatory power jointly added by the

x′s. If our coefficients for the impulse variable are significant we can reject the hypothesis

that the impulse variable is not causing the response variable to change. It is considered

that the impulse variable will react to the response variable if the lagged values of the

impulse variable have an effect on the current value of the response variable, controlling

for the lagged values of the response variable.

When estimating a panel vector autoregression, we are generally not interested in our co-

efficients. Instead, we are interested in the dynamic responses of the variables to shocks.

Therefore, we provide additional post-estimation analysis like nonparametric impulse reac-

tion functions and forecast error in variance decomposition. An impulse-reaction function

is a signal processing technique that allowed us to take our defined dynamic system with

16

an input signal, called an impulse, and to describe the reaction of the system to it as a

function of time.

We followed Love and Zicchino (2006) to transform the system of equations into a recursive

autoregressor vector with Choleski decomposition of variance-covariance matrix residuals.

Note that residuals need to be orthogonal to control for other shocks. The forecast error in

variance decomposition is useful to measure how relevant each shock is to the explanation

of each variable in the equation system. This is the same as obtaining the size of the

forecast error in variance for each variable, due to each shock at each period.

Results

Table 2, shows the results for the models that use combinations of variables for quality of

media coverage and crime rates for both reaction functions, estimated from a panel vector

autoregressor with five lags. Columns (1) and (2) show the relationship between quality

of media coverage and crime rates. The results show that exposure to crime media does

not trigger an increase in violent crime or aggression. Specifically, we can see that crime

rates show no reaction in any of the five weeks following media coverage. Additionally,

results show that quality of quality of media coverage also does not react to crime rates.

Our Granger test supports this result, showing that we fail to reject the hypothesis that

quality of media coverage does not “Granger cause” crime rates and vice versa.

Columns (3) and (4) show the relationship between the quality of media coverage and

crime rates. The results show that crime rates do not react to criminogenic media content.

Moreover, quality of media coverage has no reaction to crime rates. As expected, the

17

granger test shows that quality of media coverage and crime rates both fail to predict

each other.

(Table 2 about here)

In Tables 3 and 4, we present the same models as above, but for the two stylistic patterns:

credit-taking (columns 1 and 2) and body exposure (columns 3 and 4).

Columns (1) and (2) of table 3 show the relationship between quality of media cover-

age and observations of crimes that have a credit-taking style. The results indicate that

credit-taking styles of crime increase in a statistically significant way when similar crimes

have been covered in the prior one, two and four weeks. The Granger test supports this

result, showing that quality of media coverage helps to predict credit-taking behavior (5%

significance level), but credit-taking behavior fails to predict quality of media coverage

(35% level). Additionally, the forecast error in variance decomposition supports our find-

ings. We can see that as much as 53% of the error in the prediction of credit-taking crimes

is due to media coverage. In other words, about half of the variance of credit-taking style

is explained, not by previous cases of credit-taking, but by increased detailed media cov-

erage of cases of credit-taking. Drug cartels tend to take credit more for their crimes when

the media has previously covered their crimes with detail.

Columns (3) and (4) show the relationship between quality of media coverage and ob-

servations of crimes that have a body-exposure style. The results are consistent with

previous models, showing that body-exposure styles increase with quality of media cov-

erage of similar crimes during the prior first and fifth weeks. The Granger test supports

these findings that the quality of media coverage helps to predict body-exposure crime

18

styles but with lower significance than previous models. Accepting larger variance, the

model shows that as much as 26% of the error in the forecast of the body-exposure crime

style is due to quality of media coverage. A quarter of the variance of body-exposure style

is explained, not by previous trends in cases of body-exposure, but by increased detailed

media coverage of body-exposure crimes. In other words, rug cartels tend to expose bod-

ies more when the media has previously covered their crimes with detail, yet this effect is

less string than with credit-taking styles.

(Table 3 about here)

Overall, the models above provide evidence that media coverage functions as a crime

“rudder”, and not as crime “trigger”. Our empirical specifications show that there is not

any solid association between media exposure and crime rates. Also, that media coverage

influences the style of the crimes perpetrated by drug cartel organizations operating at the

US-Mexico border, but does not motivate more crime. In other words, we find compelling

evidence that detailed media coverage influences crime style, but not crime rates. It is

also noteworthy that the effects of detailed media coverage on credit-taking styles lasts

up to 4 weeks.

The impulse reaction functions for credit-style taking and media coverage are presented

in figure 1. They use 95 percent confidence intervals and 1000 repetitions of Monte-Carlo

simulations to create a 5 percent band error interval. The graph in the bottom-left corner

describes how the number of credit-taking style crimes reacts to quality of media coverage,

while holding all other shocks at zero. The graph in the top-right shows how quality media

coverage reacts to credit-taking style crimes, while holding all other shocks at zero. The

shocks and the impacts are presented as standard deviations.

19

(Figure 1 about here)

The bottom-left corner of Figure 1 shows that increased quality media coverage of credit-

taking crime styles increases similar crimes in the second and fourth week. This effect is

statistically significant. During the third week, the positive effect decreases and loses sta-

tistical significance; however, it increases during the fourth week and becomes statistically

significant again. After the fourth week, the graph shows a slightly decreasing tendency.

Figure 1 also shows (top-right corner) that when there is a shock in crime-taking style;

quality of coverage increases during the first and second week, decreases during the third

and fourth week, and slightly increases during the fifth week, and has practically no effect

after the fifth week. In this case, the effect of the credit-taking style on quality of media

coverage is only statistically significant for the second week. This pattern is intuitive and

it reflects the news cycle. It is unlikely that one incident would be in the news for more

than two weeks, but specialized opinion or monthly reviews can pick the story up again

once ‘it has been in the cooler’ for a while. This could explain the peaks of the coverage

function.

In Table 4, we present the same models as in Table 3, but instead examine media coverage,

independently it its quality. We have estimates for crime styles in the form of credit-taking

(columns 1 and 3) and body-exposure (columns 3 and 4).

(Table 4 about here)

The results show that we do find a statistically significant reaction to media coverage. As

in Table 4 above, both crime styles increase in a statistically significant way with media

coverage. The impact of both crime styles seems to be present during the first, third, and

20

fourth weeks. Additionally, the Granger test supports a rejection of the hypothesis that

media coverage does not cause the credit-taking criminal style (5% significance level), and

a rejection of the body-exposure hypothesis (5% significance level). As much as 1.9% of

the error in the forecast for credit-taking style, and 0.6% of the error in forecast for body

exposure style is explained by changes in media coverage.

Our results show that criminogenic media content provides criminals with inspiration

regarding the features of the crime, but does not cause crime. Interestingly, and quite

consistent with theories showing that criminogenic media contents influence the style of a

crime, we find that such influence is dramatically stronger when media covers crime with

full detail. In the case of the most detailed media coverage, we observe impacts that are

28 times larger.

Figure 2 presents the impulse-reaction functions with 95 percent confidence bands for

credit-taking and for media coverage (columns 1 and 2 in Table 4). The graph in the

bottom-left corner describes how credit-taking reacts to the media coverage, while holding

all other shocks at zero. The graph in the top-right shows how media coverage reacts to

crimes exhibiting the credit-taking style. The shocks and the impacts are presented as

standard deviations.

The bottom-left corner of Figure 2 shows that when there is an increase in media coverage,

credit-taking crimes increase in the following week. This effect is statistically significant.

During the second week, the positive effect decreases but keeps its statistical significance.

In the third week after coverage, the positive effect decreases again and loses significance.

During the fourth week, however, the positive effect increases and becomes statistically

21

significant again. After the fourth week period, the graph indicates the beginning of a

slightly decreasing tendency without statistical significance.

(Figure 2 about here)

Turning to the top-right corner of Figure 2, we see that credit-taking crimes cause coverage

to increase during the first week, decrease during the second and third week, peak during

the fourth week, but it has practically no effect after.. In this case, the effect of credit-

taking crime on media coverage is only statistically significant for the third week. Again,

this pattern reflects the news cycle.

Robustness tests

To interrogate our data further, we tested for heterogeneous effects driven by (a) the

strength of drug cartel territorial entrenchment, and (b) the degree in which drug cartels

follow media coverage. The logic of behind these tests is straightforward. First, if crim-

inogenic media coverage serves as inspiration for crime style, as we proved in our previous

models, then we should expect that media coverage will have a larger effect in are where

drug cartels are more entrenched and thus, more physically present to be influenced. Sec-

ond, if the media provides instructional cues to people intending to perform criminal acts,

we may see different effects depending on how much drug cartels follow the dissemination

of information by the media. We expect that media coverage will have a larger effect in

areas where traffickers more closely follow media coverage.

We began by testing heterogeneous effects by the strength of drug cartel territorial en-

22

trenchment. We determine whether a state is drug cartel territory by identifying the

presence of drug cartels in different territories over time. To measure drug cartel pres-

ence, we relied on a published big-data framework that uses a text-analysis algorithm

to extract web content about recorded criminal activities by subnational economy. The

algorithm “reads” digitalized records, news content, blogs and Google-News indexed con-

tent searching for instances in which drug cartel operations are mentioned. The Python

crawler was created to extract JSON objects using unambiguous query terms to perform

text analysis. The final data, cleaned using a hyper-geometric cumulative distribution

function, includes 2,449 sub-national economies, and 178 “actor terms” associated with

traffickers and drug cartel organizations. Each actor was classified according as part of

13 drug cartels and a residual category. We have information of 13 drug cartels operating

in Mexico for 19 years (1991 - 2010). Drug cartel organizations operate in 713 munici-

palities in Mexico, and in most of the US-Mexico border. A more detailed description of

the methodology that we followed can be found at the published paper (Coscia and Rios,

2012).

This framework allowed us to obtain information of a phenomenon that would otherwise

require large scale, expensive intelligence exercises. Most importantly, this procedure

helped us to disentangle drug cartels performing violent crimes from drug traffickers that

are present in peace. Many of the recorded drug cartel operations are non-violent, consis-

tent in peacefully trading, transporting, producing or cultivating illegal drugs. This data

set has been a source to study criminal activity in many published papers (Osorio, 2015;

Castillo et al., 2014; Dube et al., 2016; Holland and Rios, 2017).

Table 5 shows our estimates for the quality of media coverage and credit-taking style for

two subsets: areas that are geographical bastions for drug cartels, and areas that are not.

23

The results indicate that crimes with credit-taking style react in a statistically significant

way on the first, second and fourth, as detailed coverage increases. Meanwhile, locations

where drug cartels are not entrenched only exhibit association between criminogenic media

content and the stylistic decisions made by criminals during the second week. The Granger

test bolsters these findings, showing that media coverage helps to predict credit-taking

crimes (10% significance level) in places where drug cartels are entrenched, but fails to

do so otherwise (30% significance level). Furthermore, the forecast error in variance

decomposition supports our findings. We can see that as much as 52% of the error in the

forecast for credit-taking crimes is due to quality of media coverage.

(Table 5 about here)

A second robustness tests was conducted to evaluate heterogeneous effects the degree in

which drug cartels follow media coverage. Following the research of Holland and Rios

(2017), we used statistics of violence against the press as a proxy to measure how much

drug cartels care about the dissemination of information. The degree in which individuals

care about the dissemination of information is crucial in determining the violence that they

are willing to employ against potential informants (Reuter, 1983; Kalyvas, 2006). Only

drug cartels that are powerful and that care and pay attention to the media will incur in

the potentially quite large cost of inflicting violence against journalists to silence the press.

Violence against the press is costly because it may lead to increased federal attention and

enforcement on local illicit activity. Thus, incurring on the high cost of inflicting violence

to journalists is a proxy for how much drug cartels care and pay attention to media

contents.

To identify the existence of violence against the press, we used journalist homicide data

24

collected by the Committee to Protect Journalists (CPJ). This same data has been used

before in published academic articles like Holland and Rios (2017). The US-Mexico border

is one of the most dangerous areas in the Western Hemisphere to practice journalism, with

various sources reporting between 85 and 100 journalists having been killed or disappeared

since 2000 (Edmonds-Poli, 2013). In many recent years, journalists in Mexico have faced

levels of danger comparable to countries in war like Iraq and Pakistan (CPJ, 2014). In

fact, in 2010 and 2011, more journalists were assassinated in Mexico than in any other

country in the world, except Pakistan (CPJ, 2014).

Table 6 shows that credit-taking reacts to criminogenic media content in states where

cartels care the most about the dissemination of information. Specifically, there is a pos-

itive and statistically significant association during first, second, and fourth week after

the detailed coverage occurred. Meanwhile, states in which criminals do not follow media

coverage do not show any significant statistical reaction between credit-taking and the

quality of media coverage. The Granger test supports these findings, showing that me-

dia coverage helps to predict credit-taking crimes (1% significance level) in states where

criminals more closely follow press contents, but fails to do so in states where they do

not (65% significance level). Moreover, the forecast error in variance decomposition also

reinforces our findings. We can see that as much as 52% of the error in the forecast for

credit-taking crimes is due to media coverage.

(Table 6 about here)

In sum, our results have clearly defined the relationship between media coverage and

crime style. There is strong evidence that news coverage of stylized crimes is significantly

and predictably responsible for the replication of those styles. In other words, we find

25

that detailed media coverage of crime creates templates for crime. However, there is no

evidence that coverage impacts crime rates. Furthermore, there is a stronger and more

solid association between criminogenic media content and stylistic decisions than there

is between media and crime rates. We interpret these results as evidence of a world in

which the media influences the style of crimes for those already so inclined, but provides

no other kind of encouragement.

It is also notable that we find no evidence to suggest that the relationship between media

coverage and crime style is bidirectional. There exists a Granger causality that runs from

media coverage to criminal style, but not from criminal style to media coverage. We have

also found that the share of drug-related homicides that were publicly claimed increases

when the press covers similar credit-taking crimes. This means that publicity-seeking

criminals use more stylized crimes when they know that their messages will be better

spread by the press, but this does not mean that they commit more crimes. On this view,

media coverage and criminal behavior is not a vicious cycle, but instead a unidirectional

relationship that mostly flows from media coverage to criminal style.

The results above are further cemented by our finding of heterogeneous effects in states

where drug cartels operate. Specifically, we found a clear association between the media

and crime style in states with cartel operations, but not in states without them. Addition-

ally, we found that states in which the press was endangered by cartel operations exhibit

a strong association between media coverage and crime style, but this did not hold for

states without cartel operations. For both cases, we found evidence that the relationship

between media coverage and crime style is not bidirectional.

26

Conclusion

Our research was inspired by one of the most enduring controversies in criminology litera-

ture: whether exposure to criminogenic media content causes crime by “triggering” delin-

quency and aggression (Bushman and Anderson, 2002; Anderson and Carnagey, 2004;

Anderson and Bushman, 2001; Doley et al., 2013; Chadee et al., 2017), or just provides

“rudders” that guide the style of individuals already intending to commit crime (Ferguson

and Dyck, 2012; Ferguson et al., 2012; Grimes et al., 2008; Ferguson and Savage, 2012).

This debate has been difficult to resolve because measuring copycat crime presents major

empirical challenges (Surette, 2014), but also because criminogenic media content and

crime rates may be modeled as a self-reinforcing cycle. This would mean that media

coverage may induce criminal events, and criminal events may also induce increases in

criminogenic media content, if such events are interesting for audiences.

To provide insight into the controversy, and to overcome these empirical challenges, we

examined drug cartel operations at the US-Mexico border by leveraging two exceptionally

fruitful and unique datasets. To do so, we developed an empirical specification that

address the possibility that the relationship between media coverage and criminal behavior

may be bidirectional. Specifically, we defined reaction functions and Granger causality

tests to identify the relationships between media coverage, or media quality, and crime

rates, or crime styles.

Our major finding is that media coverage of crime provides criminals with stylistic inspi-

ration, but does not cause crime rates to increase. Specifically, we found evidence that

two “Criminogenic Styles” tend to react to media coverage: credit-taking banners and

displays of bodies/dismemberment. These results indicate that publicity-seeking crimi-

27

nals will intensify the most gruesome aspects of their crimes if they believe that doing so

means that their messages will get more attention. Additionally, contrary to suppositions,

a strong Granger causality from media coverage to crime style shows that this relationship

is unidirectional.

When media covers crimes in larger detail, criminals increase the use of credit-taking and

body exposure styles. As much as 53% of the variance in credit-taking styles of crime is

explained by changes in the quality of media coverage. The use of body-exposure styles

is also influenced by the quality of press coverage, but much less, only 26%, and with

higher variance. This may be related to the higher costs related to copycat different

styles. While credit-taking requires displaying a public banner next to a homicide, body-

exposure requires dismembering and displaying the body of the victim. Interestingly, we

also present evidence that detailed criminogenic contents (media coverage in which all

the details of the crime are covered) have up to 28 larger effects on crime style. Our

robustness tests also show that, as expected we find stronger in territories where drug

cartels are more entrenched, and where they follow media coverage with more attention.

Although this paper has contributed to a better understanding of the relationship between

the press and criminal behavior, future research would need to test for causal relationships

beyond Granger causality. Furthering this area of research is critical to the social sciences

and to democratic governance because it exposes the tension between two important

values: freedom of the press and the de-escalation of horrific publicity-seeking crime styles

(Doley et al., 2013)). In addition to guiding us as we consider this issue, this research will

assist with the development of better strategies to stifle terrorism, criminal organizations,

copycat criminals, and individuals who aspire to notoriety.

28

References

Abrahms, M. and Conrad, J. (2017). The strategic logic of credit claiming: A new theory

for anonymous terrorist attacks. Security Studies, 26(2):279–304.

Anderson, C. A. and Bushman, B. J. (2001). Effects of violent video games on aggressive

behavior, aggressive cognition, aggressive affect, physiological arousal, and prosocial

behavior: A meta-analytic review of the scientific literature. Psychological Science,

12(5):353–359.

Anderson, C. A. and Carnagey, N. L. (2004). Violent evil and the general aggression

model. In Fagerberg, J., Mowery, D. C., and Nelson, R. R., editors, The Social Psy-

chology of Good and Evil, chapter 10, pages 168–192. Oxford University Press.

Anderson, C. A., Carnagey, N. L., and Eubanks, J. (2003). Exposure to violent media:

The effects of songs with violent lyrics on aggressive thoughts and feelings. Journal of

Personality and Social Psychology, 84(5):960.

Anderson, C. A., Gentile, D. A., and Buckley, K. E. (2007). Violent video game effects on

children and adolescents: Theory, research, and public policy. Oxford University Press.

Anderson, C. A., Shibuya, A., Ihori, N., Swing, E. L., Bushman, B. J., Sakamoto, A.,

Rothstein, H. R., and Saleem, M. (2010). Violent video game effects on aggression,

empathy, and prosocial behavior in eastern and western countries: A meta-analytic

review. Psychological Bulletin, 136(2):151–174.

Atuesta, L. H. (2017). Narcomessages as a way to analyse the evolution of organised crime

in Mexico. Global Crime, 18(2):100–121.

29

Brown, J. M. (2017). Notes to the underground: Credit claiming and organizing in the

Earth Liberation Front. Terrorism and Political Violence, pages 1–20.

Brown, K. D. and Hamilton-Giachritsis, C. (2005). The influence of violent media on

children and adolescents: A public health approach. Lancet, 365(9460):702–710.

Bushman, B. J. and Anderson, C. A. (2002). Violent video games and hostile expectations:

A test of the general aggression model. Personality and Social Psychology Bulletin,

28(12):1679–1686.

Bushman, B. J. and Huesmann, L. R. (2014). Twenty-five years of research on violence

in digital games and aggression revisited. European Psychologist, 19(1):47–56.

Calderon, G., Robles, G., Dıaz-Cayeros, A., and Magaloni, B. (2015). The beheading

of criminal organizations and the dynamics of violence in Mexico. Journal of Conflict

Resolution, 59(8):1455–1485.

Campbell, H. and Hansen, T. (2014). Is narco-violence in Mexico terrorism? Bulletin of

Latin American Research, 33(2):158–173.

Carbray, P. (2002). Handkerchiefs on display as barcelona fans protest: South American

teams are biggest losers as world cup final spots are adjusted. The Gazette (Montreal).

Dec 23, 2002.

Castillo, J. C., Mejıa, D., and Restrepo, P. (2014). Scarcity without leviathan: The

violent effects of cocaine supply shortages in the Mexican drug war. Working paper.

Chadee, D., Surette, R., Chadee, M., and Brewster, D. (2017). Copycat crime dynamics:

The interplay of empathy, narrative persuasion and risk with likelihood to commit

future criminality. Psychology of Popular Media Culture, 6(2):142.

30

Coddington, M. (2014). Defending judgment and context in original reporting: Journalists

construction of newswork in a networked age. Journalism, 15(6):678–695.

Coleman, L. (2004). The copycat effect: How the media and popular culture trigger the

mayhem in tomorrow’s headlines. Simon and Schuster.

Coscia, M. and Rios, V. (2012). Knowing where and how criminal organizations oper-

ate using web content. In Proceedings of the 21st ACM international conference on

Information and knowledge management, pages 1412–1421. ACM.

Cowell, A. (1992). Big Palermo rally assails mafia grip. New York

Times. June 28, 1992. http://www.nytimes.com/1992/06/28/world/

big-palermo-rally-assails-mafia-grip.html.

Coyne, S. M. (2007). Does media violence cause violent crime? European Journal on

Criminal Policy and Research, 13(3-4):205–211.

CPJ (2014). Committee to protect journalists: Journalists killed since 1992. Accessed

May 20, 2014. cpj.org/killed/.

DeCamp, W. and Ferguson, C. J. (2017). The impact of degree of exposure to violent

video games, family background, and other factors on youth violence. Journal of Youth

and Adolescence, 46(2):388–400.

Dell, M. (2015). Trafficking networks and the mexican drug war. The American Economic

Review, 105(6):1738–1779.

Doley, R., Ferguson, C., and Surette, R. (2013). Copycat firesetting: Bridging two research

areas. Criminal Justice and Behavior, 40(12):1472–1491.

31

Dube, O., Garcıa-Ponce, O., and Thom, K. (2016). From maize to haze: Agricultural

shocks and the growth of the Mexican drug sector. Journal of the European Economic

Association, 14(5):1181–1224.

Duran-Martınez, A. (2015). To kill and tell? State power, criminal competition, and drug

violence. Journal of Conflict Resolution, 59(8):1377–1402.

Edmonds-Poli, E. (2013). The effects of drug-war related violence on Mexico’s press and

democracy. Working paper series on Civic Engagement and Public Security in Mexico.

Wilson Center. University of San Diego.

Elson, M. and Ferguson, C. J. (2014). Twenty-five years of research on violence in digital

games and aggression. European Psychologist, 19(1):33–47.

Ferguson, C. J. (2002). Media violence: Miscast causality. The American Psychologist,

57(6–7):446–447.

Ferguson, C. J. (2015). Does movie or video game violence predict societal violence? It

depends on what you look at and when. Journal of Communication, 65(1):193–212.

Ferguson, C. J. and Dyck, D. (2012). Paradigm change in aggression research: The time

has come to retire the general aggression model. Aggression and Violent Behavior,

17(3):220–228.

Ferguson, C. J. and Kilburn, J. (2009). The public health risks of media violence: A

meta-analytic review. The Journal of Pediatrics, 154(5):759–763.

Ferguson, C. J., Rueda, S. M., Cruz, A. M., Ferguson, D. E., Fritz, S., and Smith, S. M.

(2008). Violent video games and aggression: Causal relationship or byproduct of family

32

violence and intrinsic violence motivation? Criminal Justice and Behavior, 35(3):311–

332.

Ferguson, C. J., San Miguel, C., Garza, A., and Jerabeck, J. M. (2012). A longitudinal

test of video game violence influences on dating and aggression: A 3-year longitudinal

study of adolescents. Journal of Psychiatric Research, 46(2):141–146.

Ferguson, C. J., San Miguel, C., and Hartley, R. D. (2009). A multivariate analysis of

youth violence and aggression: The influence of family, peers, depression, and media

violence. The Journal of Pediatrics, 155(6):904–908.

Ferguson, C. J. and Savage, J. (2012). Have recent studies addressed methodological issues

raised by five decades of television violence research? A critical review. Aggression and

Violent Behavior, 17(2):129–139.

Fink, K. and Schudson, M. (2014). The rise of contextual journalism, 1950s–2000s. Jour-

nalism, 15(1):3–20.

Freedman, D. (2008). The politics of media policy. Polity Press, Cambridge.

Freedman, J. L. (1984). Effect of television violence on aggressiveness. Psychological

Bulletin, 96(2):227.

Gambacorta, L., Hofmann, B., and Peersman, G. (2014). The effectiveness of unconven-

tional monetary policy at the zero lower bound: A cross-country analysis. Journal of

Money, Credit and Banking, 46(4):615–642.

Gambetta, D. (2009). Codes of the Underworld: How Criminals Communicate. Princeton

University Press, Princeton, NJ.

33

Granger, C. W. (1969). Investigating causal relations by econometric models and cross-

spectral methods. Econometrica: Journal of the Econometric Society, 37(3):424–438.

Grimes, T., Anderson, J. A., and Bergen, L. (2008). Media violence and aggression:

Science and ideology. Sage Publications, Thousand Oaks.

Gunter, B. (2008). Media violence is there a case for causality? American Behavioral

Scientist, 51(8):1061–1122.

Helfgott, H. (2015). Growth in groups: Ideas and perspectives. Bulletin of the American

Mathematical Society, 52(3):357–413.

Hoffman, A. M. (2010). Voice and silence: Why groups take credit for acts of terror.

Journal of Peace Research, 47(5):615–626.

Hoffman, A. M., Jengelley, D. H., Duncan, N. T., Buehler, M., and Rees, M. L. (2010).

How does the business of news influence terrorism coverage? Evidence from the Wash-

ington Post and USA Today. Terrorism and Political Violence, 22(4):559–580.

Holland, B. E. and Rios, V. (2017). Informally governing information: How criminal

rivalry leads to violence against the press in Mexico. Journal of Conflict Resolution,

61(5):1095–1119.

Hsu, H. Y. and McDowall, D. (2017). Does target-hardening result in deadlier terrorist

attacks against protected targets? an examination of unintended harmful consequences.

Journal of Research in Crime and Delinquency, 54(6):930–957.

Huang, B.-N., Hwang, M. J., and Yang, C. W. (2008). Causal relationship between energy

consumption and gdp growth revisited: A dynamic panel data approach. Ecological

Economics, 67(1):41–54.

34

Huesmann, L. R. (1986). Psychological processes promoting the relation between exposure

to media violence and aggressive behavior by the viewer. Journal of Social Issues,

42(3):125–139.

Huesmann, L. R. and Taylor, L. D. (2006). The role of media violence in violent behavior.

Annu. Rev. Public Health, 27:393–415.

Ivory, A. H. and Kaestle, C. E. (2013). The effects of profanity in violent video games

on players’ hostile expectations, aggressive thoughts and feelings, and other responses.

Journal of Broadcasting & Electronic Media, 57(2):224–241.

Jaeger, D. A. and Paserman, M. D. (2008). The cycle of violence? An empirical analysis

of fatalities in the Palestinian-Israeli conflict. American Economic Review, 98(4):1591–

1604.

Kalyvas, S. N. (2006). The logic of violence in civil war. Cambridge University Press,

New York.

Kearns, E. M., Conlon, B., and Young, J. K. (2014). Lying about terrorism. Studies in

Conflict & Terrorism, 37(5):422–439.

Lantz, A. (2016). The performativity of violence: Abducting agency in Mexico’s drug

war. Journal of Latin American Cultural Studies, 25(2):253–269.

Love, I. and Zicchino, L. (2006). Financial development and dynamic investment behavior:

Evidence from panel var. The Quarterly Review of Economics and Finance, 46(2):190–

210.

Martin, C. (2012). Categorization of narcomessages in Mexico: An appraisal of the

35

attempts to influence public perception and policy actions. Studies in Conflict & Ter-

rorism, 35(1):76–93.

McCarthy-Latimer, C. and Kendrick Jr, J. R. (2016). How communication technologies

function as platforms and pathways to civic and political engagement: An agenda

for research and public policy. International Journal of Civic Engagement and Social

Change (IJCESC), 3(1):50–72.

Mendoza Rockwell, N. (2016). Narco-mantas o el confın de lo criminal. Acta poetica,

37(2):21–34.

Milenio (2017). Organized-crime group related homicides (2007-2015) from Milenio. Com-

piled by Justice in Mexico, San Diego.

Min, E. (2013). Taking responsibility: When and why terrorists claim attacks. APSA

2013 Annual Meeting Paper.

Molzahn, C., Rıos, V., and Shirk, D. A. (2012). Drug violence in Mexico. San Diego,

Trans-Border Institute, University of San Diego.

Murray, J. P. (2008). Media violence the effects are both real and strong. American

Behavioral Scientist, 51(8):1212–1230.

Nizza, M. (2013). Updates on Virginia Tech. New York Times. April 18, 2007. https:

//thelede.blogs.nytimes.com/2007/04/18/updates-on-virginia-tech/.

Ortiz, I. (2013). Family flees after possible cartel banner left at san juan home.

The Monitor (McAllen). May 21, 2013. http://www.themonitor.com/news/local/

family-flees-after-possible-cartel-banner-left-at-san-juan/article_

36cd5f0c-c262-11e2-b7ba-001a4bcf6878.html.

36

Osorio, J. (2015). The contagion of drug violence: Spatiotemporal dynamics of the Mex-

ican war on drugs. Journal of Conflict Resolution, 59(8):1403–1432.

Philip, G. and Berruecos, S., editors (2012). Mexico’s struggle for public security: Orga-

nized crime and state responses. Palgrave MacMillan, New York.

Ramirez, M. D. (2013). Punitive sentiment. Criminology, 51(2):329–364.

Reforma (2017). Organized-crime group related homicides (2007-2016) from Reforma’s

Ejecutometro. Compiled by Justice in Mexico, San Diego.

Reiner, R. and Newburn, T. (2007). Policing and the police. Oxford University Press,

Oxford.

Reuter, P. (1983). Disorganized crime: The economics of the visible hand. MIT Press,

Cambridge, MA.

Reyes, I. d. l. (2011). Mexico: entre la responsabilidad y la censura de los medios. BBC.

March 25, 2011. http://www.bbc.com/mundo/noticias/2011/03/110325_mexico_

asesinato_ao.shtml.

Rıos, V. (2015). How government coordination controlled organized crime: The case of

Mexico’s cocaine markets. Journal of Conflict Resolution, 59(8):1433–1454.

Salopek, P. (2011). Conflict graffiti. Foreign Policy, (189):94–96.

Saridakis, G. (2004). Violent crime in the United States of America: A time-series analysis

between 1960–2000. European Journal of Law and Economics, 18(2):203–221.

Savage, J. (2004). Does viewing violent media really cause criminal violence? A method-

ological review. Aggression and Violent Behavior, 10(1):99–128.

37

Savage, J. (2008). The role of exposure to media violence in the etiology of violent behavior

a criminologist weighs in. American Behavioral Scientist, 51(8):1123–1136.

Savage, J. and Yancey, C. (2008). The effects of media violence exposure on criminal

aggression a meta-analysis. Criminal Justice and Behavior, 35(6):772–791.

Schmid, A. P. and De Graaf, J. (1982). Violence as communication. Beverly Hills, CA:

Stage.

Shirk, D. and Wallman, J. (2015). Understanding Mexico’s drug violence. Journal of

Conflict Resolution, 59(8):1348–1376.

Slater, M. D. (2007). Reinforcing spirals: The mutual influence of media selectivity and

media effects and their impact on individual behavior and social identity. Communica-

tion Theory, 17(3):281–303.

SNSP (2011). Base de datos por fallecimientos por presunta rivalidad delincuencial, di-

ciembre de 2006 a septiembre de 2011. Data retrieved from Secretaria de Gobernacion.

Stack, S. (2000). Media impacts on suicide: A quantitative review of 293 findings. Social

Science Quarterly, pages 957–971.

Surette, R. (2013). Cause or catalyst: The interaction of real world and media crime

models. American Journal of Criminal Justice, 38(3):392–409.

Surette, R. (2014). Estimating the prevalence of copycat crime: A research note. Criminal

Justice Policy Review, 25(6):703–718.

Surette, R. (2015). Thought bite: A case study of the social construction of a crime and

justice concept. Crime, Media, Culture, 11(2):105–135.

38

Surette, R. and Gardiner-Bess, R. (2013). Media, entertainment, and crime. The Routledge

Handbook of International Crime and Justice Studies, page 373.

Trejo, G. and Ley, S. (2017). Why did drug cartels go to war in Mexico? subnational party

alternation, the breakdown of criminal protection, and the onset of large-scale violence.

Comparative Political Studies, https://doi.org/10.1177/0010414017720703.

Valenzuela, S., Arriagada, A., and Scherman, A. (2012). The social media basis of youth

protest behavior: The case of Chile. Journal of Communication, 62(2):299–314.

Wright, P. J. (2009). Sexual socialization messages in mainstream entertainment mass

media: A review and synthesis. Sexuality & Culture, 13(4):181–200.

Zhang, W., Johnson, T. J., Seltzer, T., and Bichard, S. L. (2010). The revolution will be

networked: The influence of social networking sites on political attitudes and behavior.

Social Science Computer Review, 28(1):75–92.

39

Tables

Table 1: Descriptive statisticsVariable Mean Std. Dev. Min MaxMedia coverage 1.29 6.11 -77.62 61.45Quality of media coverage 0.11 0.30 0 1Credit-taking style crimes 0.03 0.14 0 2.5Body-exposure style crimes 0.08 0.26 0 1

40

Table 2: Crime and media coverageVariable Coverage Homicides Coverage Quality HomicidesMedia coverage (t-1) -0.0046 0.0278Media coverage (t-2) 0.0022 -0.0211Media coverage (t-3) 0.0308 -0.0223Media coverage (t-4) -0.0040 -0.0038Media coverage (t-5) 0.0696 -0.0323Coverage quality (t-1) 0.1078*** 0.2056Coverage quality (t-2) 0.0770*** -0.0549Coverage quality (t-3) 0.0107 -0.0088Coverage quality (t-4) 0.0396* 0.2573Coverage quality (t-5) 0.0117 0.0465Homicides (t-1) -0.0992 1.2025*** 0.0053 1.1804***Homicides (t-2) -0.0347 -0.2813*** -0.0040 -0.2712***Homicides (t-3) 0.1742 0.1222*** 0.0066* 0.1259***Homicides (t-4) -0.2113 -0.2606*** -0.0008* -0.2637***Homicides (t-5) 0.1915 0.2153*** -0.0043 0.2208**Granger test (Prob>chi2) 0.249 0.301 0.015 0.442

Note: * p <0.1; ** p <0.05; *** p <0.01.

41

Table 3: Quality of media coverage and crime stylesVariable Coverage quality Credit-taking Coverage Quality Body exposureCoverage quality (t-1) 0.0683** 0.0683** 0.1020*** 0.0198**Coverage quality (t-2) 0.0213 0.0442* 0.0738*** 0.0096Coverage quality (t-3) -0.0092 0.0172 -0.0016 0.0033Coverage quality (t-4) 0.0282 0.0568** 0.0295 0.0026Coverage quality (t-5) -0.0053 0.0099 0.0177 0.0222**Credit-taking (t-1) 0.0424 0.0612*Credit-taking (t-2) 0.0692** 0.0838**Credit-taking (t-3) 0.0129 0.0373Credit-taking (t-4) -0.0017 0.0925*Credit-taking (t-5) 0.0076 -0.0182Body exposure (t-1) 0.0088 0.0612*Body exposure (t-2) -0.0024 0.0838**Body exposure (t-3) 0.0352 0.0373Body exposure (t-4) 0.0250 0.0925*Body exposure (t-5) -0.0418 -0.0182Granger test (Prob>chi2) 0.312 0.048** 0.786 0.155

Note: * p <0.1; ** p <0.05; *** p <0.01.

42

Table 4: Media coverage and crime stylesVariable Media coverage Credit-taking Media coverage Body exposureMedia coverage (t-1) 0.0153 -0.0033*** 0.0168 -0.0005*Media coverage (t-2) 0.0191 -0.0012 0.0193 -0.0003Media coverage (t-3) 0.0406 -0.0022** 0.0414 -0.0009***Media coverage (t-4) 0.0072 -0.0018* 0.0107 -0.0006*Media coverage (t-5) 0.0709 -0.0012 0.0710 -0.0002Credit-taking (t-1) 0.1689 0.1599***Credit-taking (t-2) -0.3733 0.1310***Credit-taking (t-3) -0.7232* 0.0609***Credit-taking (t-4) 0.5442 0.0498**Credit-taking (t-5) -0.2387 0.0462**Body exposure (t-1) 0.6198* 0.0815***Body exposure (t-2) 0.2679 0.0947***Body exposure (t-3) -0.3150 0.0425Body exposure (t-4) 0.6989** 0.0967**Body exposure (t-5) 0.2183 0.0043Granger test (Prob>chi2) 0.204 0.037** 0.101 0.012**

Note: * p <0.1; ** p <0.05; *** p <0.01.

43

Table 5: Strength of drug cartel territorial entrenchmentOperations No operations

Variable Coverage quality Credit-taking Coverage Quality Credit-takingCoverage quality (t-1) 0.0653 0.0754** 0.0722 0.0632*Coverage quality (t-2) 0.0194 0.0658* 0.0244 0.0029Coverage quality (t-3) -0.0193 0.0288 0.0108 -0.0102Coverage quality (t-4) 0.0241 0.0755** 0.0372 0.0172Coverage quality (t-5) -0.0511 -0.0103 0.0908** 0.0510*Credit-taking (t-1) 0.0472 0.1319*** 0.0334 0.0229Credit-taking (t-2) 0.0685* 0.0946** 0.0685 0.0933**Credit-taking (t-3) 0.0175 0.0356 0.0047 0.0819Credit-taking (t-4) 0.0015 0.0112 -0.0112 0.0071Credit-taking (t-5) 0.0287 0.0559 -0.0402 0.0147Granger test (Prob>chi2) 0.484 0.056** 0.775 0.293

Note: * p <0.1; ** p <0.05; *** p <0.01.

44

Table 6: Degree in which drug cartels follow media coverageViolence against the press No violence against the press

Variable Coverage quality Credit-taking Coverage Quality Credit-takingCoverage quality (t-1) 0.1303** 0.1028** 0.0411 0.0533Coverage quality (t-2) 0.0211 0.0724* 0.0221 0.0341Coverage quality (t-3) 0.0026 0.0295 -0.0127 0.0129Coverage quality (t-4) 0.0806 0.1342*** 0.0048 0.0243Coverage quality (t-5) 0.0165 -0.0141 -0.0130 0.0219Credit-taking (t-1) -0.0499 0.0443 0.0821** 0.1313Credit-taking (t-2) 0.0764 0.0751 0.0661 0.1085Credit-taking (t-3) 0.0344 0.0728 0.0030 0.0400Credit-taking (t-4) -0.0540 -0.0695 0.0226 0.0446Credit-taking (t-5) -0.0199 0.0258 0.0174 0.0476Granger test (Prob>chi2) 0.539 0.006*** 0.262 0.639

Note: * p <0.1; ** p <0.05; *** p <0.01.

45

Figures

46

Figure 1: Impulse reaction functions, Quality of media coverage and Credit-taking stylecrimes

47

Figure 2: Impulse reaction functions, Media coverage and Credit-taking style crimes

48