The impact of the criminal presence and
violence on wages: Evidence from Mexico
Preliminary Draft
Gustavo Iriarte∗
University of Sussex†
April 11, 2017
Abstract
This paper uses information on wages from the Mexican Family Life
Survey (2002-2010), data from homicide rates and a unique data set
created with a text analysis algorithm of web content that reflects the
presence of Drug Trafficking Organizations (DTO) in Mexican Munic-
ipalities from (Coscia and R´ıos, 2012). To address how the presence of
criminal organizations and violence associated with the War on Drugs
impacts municipal wages. Ex-ante the effect is unknown as the impact
of violence and criminal presence on labour markets is multidimensional
and can vary depending whether the worker is employed as formal or
informal. The estimation results of the prefferred specification after in-
strumenting violence and criminal presence to address reverse causality,
yields a positive effect of criminal presence on wages and no effect of
violence. More specifically, holding everything else constant, an addi-
tional DTO per municipality increases wages by 8.8 percentage points.
Further dividing the sample, wages for informal workers increases by
9.6 and 5.9 percentage points for formal workers. These results are ro-
bust to the correction for self-selection bias using the Heckman (1979)
two step procedure.
JEL classification: C26, D74, J24, J31, J46, O54.
Keywords: Violence, Crime, Wages, Mexico, Compensating wage differentials,
War on Drugs, Drug Trafficking Organizations.
∗
C.Iriarte-Rivas@sussex.ac.uk
†
Department of Economics, 230 Jubilee Building, BN1 9SL, Brighton, United Kingdom
1
1 Introduction
In the year 2006, after Felipe Calderon became president, the government of
Mexico launched an offensive to tackle organized crime in what was the be-
ginning of the so called ‘War on Drugs in Mexico’. The government pursued
an intensive policy of containment of Drug Trafficking Organizations (DTOs)
and this led to a dramatic increase in violence and the presence of criminial
organizations in Mexican municipalities measured by homicide rates and the
presence of DTOs respectively. The effects of violence in a country, if per-
sistent, has costly and profound negative effects on economic and social out-
comes, as this promotes illegality and discourages investment in infrastructure.
The literature on conflict has pointed out that crime “taxes” the econ-
omy and increases the cost of doing business (Abadie and Gardeazabal, 2003;
Gaibulloev and Sandler, 2008; Detotto and Otranto, 2010; Enamorado et al.,
2014) not just in the formal but in the informal sector as well(Camacho and
Rodriguez, 2013; BenYishay and Pearlman, 2014). The impacts of crime go
beyond factor accumulation and also have negative effects on economic diversi-
fication, increasing sector concentration and diminishing economic complexity
(R´ıos, 2015).
On the other hand, the effects of violence on socioeconomic aspects ranges
from inequality (Fajnzlber et al., 2002; Enamorado et al., 2016) to migration
(Kondylis, 2010; Calder´on et al., 2011; G´omez, 2012; Lozano and Aleman,
2013; R´ıos, 2014; Atuesta and Paredes, 2015; Calder´on-Mej´ıa and Ib´an ˜ez,
2015), human capital accumulation and school performance (Barrera et al.,
2004; Shemyakina, 2011; Rodriguez and Sanchez, 2012; Leon, 2012; Justino
et al., 2013; Brown and Vel´asquez, 2015; Romano et al., 2015), labour pro-
ductivity and outcomes (Bozzoli et al., 2013; Robles et al., 2013; BenYishay
and Pearlman, 2013; Fern´andez et al., 2014; Vel´asquez, 2014; Cabral et al.,
2016). However, the literature has generally neglected the effect on wages in
a developing economy setting. The only two studies that analyse the compen-
sating wage differential effects of high levels of crime are Smith Kelly (2011)
and Braakmann (2009) but for a developed country setting. The contribution
of this paper to the literature is twofold. First, it offers an explanation of
the impact that the presence of DTOs in Mexican municipalities and violence
have on the wages of individuals. Second, it offers an explanation of the im-
pact for both formal and informal workers as this sector functions differently
to the formal sector. To my knowledge, these are topics in the literature that
remain basically unexplored to date.
Given the availability of individual information from the Mexican Family
Life survey (2005-2010), data from homicide rates and a unique dataset that
reflects the presence of drug cartels in Mexican Municipalities (Coscia and
R´ıos, 2012), I am able to address how the violence associated with the ‘War
on Drugs’ and the presence of DTOs in Mexican municipalities impacts wages.
Ex-ante the effect is unknown as the impact of violence and criminal presence
on labour markets is multidimensional and can vary depending whether the
worker is employed as formal or informal. For example, the presence of such
2
groups can signal the absence of the rule of law in Municipalities pushing firms
to re-locate to avoid the risk of attacks, extortion or theft, thus pushing wages
down. It can also mean that because these groups inject illegal money into
the local economy, this could create employment opportunities and thus push
wages up.
The estimation results of the preferred specification after instrumenting
violence and the presence of DTOs to address reverse causality, yields a posi-
tive effect of the presence of DTOs, but no effect of violence. More specifically,
an additional DTO per municipality increases wages by 5.7%. On further dis-
sagregation, wages for informal workers increases by 4.9% and 3.4% for formal
workers. This paper is divided as follows: In section 2 the context behind the
war on drugs in Mexico is detailed, section 3 reviews the current literature on
the causes and effect of the increase in violence, and section 4 details the data
used and the empirical strategy. Results are discussed in section 5 and some
conclusions are detailed in section 6.
2 The Mexican context: Drug trafficking,
violence and the War on drugs
Mexico’s geographic location is crucial for terrestrial smuggling of drugs to
the US. Ensuring the safe passage of drugs and its commercialization is the
raison d’etre of Drug Trafficking Organizations (DTO) in Mexico and their
existence goes a long way back in Mexican history.1 It is argued that these
criminal organizations are so deeply rooted in society that they have made
pacts with the Mexican Government. Arrangements between Officials and
DTO were possible because there was willingness for this to take place from
some corrupt officials of the Institutional Revolutionary Party (PRI), an hege-
monic party that ruled the country for a long period of time after the Mexican
Revolution in 1910. The PRI used its “great patronage machine” to establish
a patron-client relationship with the DTOs (O’Neil, 2009).
The PRI allowed these organizations to operate following a strict code of
conduct and enforced its compliance using “extra-official mechanisms”, as ex-
plained by Ricardo Monreal, former Governor of the State of Zacatecas and
ex-PRI official. The most important rule to follow was the respect for DTO
territories. The size and borders were granted by the ruling party and all
DTO had to respect them(R´ıos, 2010).2 All this control over such groups
was possible because the PRI had a strict centralized control over state and
1
An example of how long this trafficking has existed in the Mexican Society is that
after the period of opium Prohibition in 1914, the U.S. customs officials recorded that the
Governor of Baja California (1916-1920) was responsible for the control of opium trafficking
(Astorga, 2001).
2
There were other rules that had to be strictly followed by the DTO. Such rules included
keeping visibility of DTO operations to a minimum (from media scandals to dead people
in the streets), periodic seizure of illegal drugs and imprisonment of lower level traffickers,
generation of economic revenues for small, poor communities, among others. To see the full
detail of rules see R´ıos (2010).
3
municipal governments. The outcome was a win-win situation in which the
government received a “tax” via bribes, information about dealings, associates
and competition (specially from those that traffic without permission)3 and,
as a quid pro quo the DTO would be permitted to operate without being
systematically prosecuted (R´ıos, 2010).
In the 1980s and 1990s, Mexico underwent a process of political opening
at the state level, this slowly weakened the PRI’s compacts with DTO. Elec-
toral competition ended understandings and pushed the DTO drug lords to
negotiate with parties at different levels of government (O’Neil, 2009; R´ıos,
2010). This encouraged rival traffickers to bid for new market opportuni-
ties. Mexico’s drug related violence rose first in states where the opposition
ruled,4 because the incoming political parties had relative less ruling experi-
ence and this created information asymmetries, hence increasing the cost of
negotiations. The increased violence in states governed by the opposition was
capitalised on politically by the party in the Federal Government. Governors
that didn’t belonged to the PRI had almost no support from the President
relative to those that did. (Astorga, 2001)
The continuous weakening of the DTOs-government agreement was com-
plemented by two important aspects: First, the increase demand for mari-
juana which became quite popular among U.S. consumers in the 1970s and
then shifted to cocaine in the 1980s. Second, an increasing involvement of the
Mexican DTOs with their Colombian counterparts. Colombian DTOs moved
cocaine into Miami, via the Gulf of Mexico and the Caribbean. Later, with
the disintegration of the major Colombian DTOs in the late 1980s and early
1990s, their Mexican counterparts gained entire control over smuggling routes
into the US. By 1991, Mexico reportedly accounted for an estimate 300-350
tons of cocaine and roughly 30 percent of all heroin and marijuana that en-
tered into the US (Astorga, 2001; Astorga and Shirk, 2010).
In the presidential elections in the year 2000, Vicente Fox, the right wing
opposition candidate was elected President. This event resulted in the termi-
nation of the arrangements between the PRI and DTOs. The DTOs strategy
then shifted towards gaining autonomy and ending their subordination to the
government. This was accomplished by buying off or intimidating local au-
thorities to secure the safe passage of drugs to the US (O’Neil, 2009; R´ıos,
2010). Given that the number of border crossings and ports of entry are
limited, the competition between DTOs becomes fierce and violent. Gaining
control over such ports of entry ensures a profitable business of drug traffick-
ing (Robles et al., 2013).5
3
This helped the officials in the Government in gaining credit, praise and promo-
tion(Snyder and Duran-Martinez, 2009).
4
For example, after the PRI lost its first governorship in Baja California in 1989, drug-
related violence surged there. The same happened in 1992 in Chihuahua. For a full discus-
sion see O’Neil (2009).
5
Skaperdas (2002) defines this as localized competition, where geography, transportation
difficulties and communication costs makes warlords use violence to establish control over
a limited area.
4
Despite the fact that agreements between government and DTOs were
weakening, violence (as measured by homicide rates) remained relatively sta-
ble over time as shown in figure 1. However, there is a period when the vio-
lence increased at levels never seen before.6 Soon after taking office in 2006,
President Felipe Calderon changed completely the strategy towards DTO.
Calderon’s government pursued an intensive policy of containment of DTO,
this involved the use of security forces at all three levels of government, in-
cluding the Army and Navy. The War against drugs which involved over
45,000 troops officially began in December 2006, when the Army was sent to
Michoacan. This was the launch of the first “Joint Operation”.7 Soon after,
the Army also arrived in Nuevo Leon, Guerrero and Tijuana(O’Neil, 2009;
Dıaz-Cayeros et al., 2011; Robles et al., 2013).
The increasing competition among Mexican DTOs created an atmosphere
of violence. The use of violence became the way in which these groups signaled
that they were in control (Castillo et al., 2014). This implied operations to
create fear, such as recruiting members in the streets, leaving messages in the
open that could be widely broadcast in the media (Dıaz-Cayeros et al., 2011).
Bodies were left mutilated in the streets with messages directed at politicians,
citizens and fellow criminals. Heads were thrown into the doors of primary
schools and mass executions replaced targeted murders (R´ıos, 2014).
The government military strategy targeted the drug hierarchy in a non-
selective way in what is commonly known as the “Kingpin Strategy”.8 It
is reported that after the offensive began, approximately 23 criminal leaders
were arrested or shot. Lindo and Padilla-Romo (2015) find that the capture
of a DTO leader in a municipality increases its homicide rate by 80% and
that this effect holds in the short run for up to a period of 12 months.9 In
this operation, major DTO such as Beltran-Leyva, La Familia Michoacana,
and the Cartel del Golfo were weakened. The fragmentation of DTO created
conditions for second and third generations of criminal leaders to compete for
territory, control and power. Soon, other groups emerged such as Cartel de
Jalisco Nueva Generacion and Caballeros Templarios. Violence emerged be-
cause many of their aspiring leaders worked as hitmen for the major DTO and
were keen on violence (O’Neil, 2009; Dıaz-Cayeros et al., 2011). One group
recruited former military officials and became widely known as Los Zetas,
6
There are alternative hypothesis of why the homicides increased so dramatically which
are worth mentioning. Dube et al. (2013) argues that the expiration of the U.S. Federal
Assault Weapon Ban in 2004 exerted a spillover of gun supply in Mexico, thus fueling the
violence between groups and against the Government. Castillo et al. (2014) on the other
hand, argues that violence increased as a consequence of cocaine supply shortages as a
result of a redefinition of the Colombian Government strategy towards DTO, which focused
more on the interdiction of drug shipments rather than targeting coca crops.
7
In total, nine “Joint Operations” have been enacted (Dıaz-Cayeros et al., 2011).
8
It has been documented that municipalities where elections were closely contested and
where eventually the PAN won, violence increased, as these mayors were more likely to
support the President’s strategy to combat crime in a direct way (Dell, 2015).
9
This study confirms that the strategy caused destabilization within organizations after
the capture.
5
characterised by the use of extreme, high profile violence (Astorga and Shirk,
2010).
Figure 1 shows the national homicide and drug-related homicide rates per
100,000 inhabitants from the period of 2000 until 2013.10 The homicide rate
remained almost constant before Calderon’s period, then it increased dramat-
ically from around 9 percent in 2006 up to almost 24 percent in 2011. It is
clear that the sudden increase in homicides after 2006 is largely attributed to
drug-related homicides. Furthermore, figure 2 shows the rate per 100,000 by
municipalities for different years before and after 2006.11 Violence is concen-
trated in areas that are close to the border and where marijuana is planted
and grown. Finally, it can be visually confirmed in figure 3 that the presence
of one DTO or more in a municipality is highly correlated with the homicide
rate in figure 2 for a set of selected years.
The proliferation of new groups created the conditions for the foray into
other illegal activities such as kidnappings, human trafficking, petroleum theft,
money laundering and arms trafficking. Perhaps, extortion was the most
widespread of these activities. It first targeted illegal business such as prosti-
tution and casinos, in which the probability of being denounced by the own-
ers to the authorities was low. Soon, extortions also extended into the legal
setting, creating an atmosphere of fear that affected local businesses. High
protection fees and intimidation forced many to close (R´ıos, 2014). Violence
increased more in municipalities with DTO presence and specially in those
that had more than one DTO (as shown in figure 3) as there is no monopoly
of violence and such groups engage in competition to win power (Castillo
et al., 2014). Incidentally, violence and DTO proliferated more where illegal
crops are grown and where there was trading and transit of drugs, money
laundering and potential markets for consumption (O’Neil, 2009).
3 Literature review
The analysis of how violence affects economic and social aspects is not new.
There are a number of studies which have found that crime has negative effects
on economic performance. The presence of crime acts like a tax on the whole
economy, as it increases the cost of doing business and creates uncertainty
(Abadie and Gardeazabal, 2003; Gaibulloev and Sandler, 2008; Detotto and
Otranto, 2010; Enamorado et al., 2014). It also affects firms, harming their
competitiveness and pushing them to close in locations that lack of the in-
stitutions to enforce the Rule of Law and property rights.12 The incentives
10
Using data from the National Institute of Statistics and Geography (INEGI) and data
from the National Security Commission (CNS). Data for drug-related homicides is only
available after 2006, when the government officially launched its operation against DTO.
11
In Mexico, there are three different levels of government. National level, state level and
Municipality level.
12
Firms that are exposed to violence may not always close, but instead shrink and go
through a process of “forgetting by not doing”, which has negative effects on productivity.
This process has important implications in the long run as the post conflict economic
6
to invest or to expand operations of a firm are low. This negative effect is
stronger in areas where the fear of victimization is high. Operational costs
increase due to the additional costs on security infrastructure incurred by the
firms. This affects both formal and informal sector, manufacturing or services
(Camacho and Rodriguez, 2013; BenYishay and Pearlman, 2014). The im-
pacts of crime go beyond factor accumulation and also exert negative effects
on economic diversification, increasing sector concentration and diminishing
economic complexity (R´ıos, 2015).
Alternatively, there are studies analyzing the effect of crime on foreign and
direct investment. Results are mixed and depend on the context of analysis
and more specifically, the industry. Ashby and Ramos (2013) find that in
Mexico crime deters foreign direct investment in financial services, commerce
and agriculture but not in oil and mining sectors, for which they found the
opposite effect. The latter is consistent with the findings for the Colombian
case by Maher (2015), which is that the presence of crime in certain industries
such as oil create the conditions that facilitate foreign direct investment flows.
This could occur because, according to Driffield et al. (2013), countries with
weaker institutions and less concern about corporate social responsibility are
more likely to invest in conflict regions.
The literature on the impact of crime on socioeconomic aspects range from
inequality to migration, human capital accumulation, labour productivity and
labour outcomes. A number of studies have pointed out that high inequality
creates the conditions for the proliferation of violence and illegal activities
(Fajnzlber et al., 2002; Enamorado et al., 2016). To some extent, the pop-
ulation in a given locality normalizes violence and learns to live with it.13
However, when violence becomes extreme, it disrupts the life of the popula-
tion and changes their behaviour. The first natural response to the presence of
high levels of violent crime on the place of residence (and one which has been
widely documented) is displacement. Both internal and external due to the
fear of victimization or threats by the groups causing the violence (Kondylis,
2010; Calder´on et al., 2011; G´omez, 2012; Lozano and Aleman, 2013; R´ıos,
2014; Atuesta and Paredes, 2015; Calder´on-Mej´ıa and Ib´an ˜ez, 2015).
One of the aspects that has received particular attention is the investment
in education. A number of studies have found that in the short run, exposure
to high level of violent crime, reduces school attendance and increases dropout
rates (Barrera et al., 2004; Shemyakina, 2011; Rodriguez and Sanchez, 2012).
This affects negatively child performance in school and increases failure rates
(Romano et al., 2015). The effects are largely visible in the long run, where
it has been found that individuals exposed to violence have, on average, less
years of schooling (Leon, 2012; Justino et al., 2013; Brown and Vel´asquez,
2015).
recovery of countries is slow and in some cases stagnates (Collier and Duponchel, 2013).
13
It has been documented that individuals change their behavior when their perception
of risk is high. They stop using public transport services, change commuting routes, stop
going to restaurants and coffee shops. In more extreme cases they arm themselves and even
suffer from sleep deprivation (Becker et al., 2004; Braakmann, 2012).
7
Violence can also alter the equilibrium in labour markets. The literature
has documented that in areas affected by high levels of violence productivity,
the proportion of employed and working hours fell (Robles et al., 2013; Cabral
et al., 2016). This reduction in working hours is largely attributed to the self
employed as the flexible nature of their jobs allows them to devote less time to
work and minimize their exposure to risk. This effect is stronger for women, as
they not only cut the number of hours worked but they also leave the labour
market and devote more time to household chores and caring for their family.
This results in a loss in hourly and total earnings (BenYishay and Pearlman,
2013; Fern´andez et al., 2014; Vel´asquez, 2014). Men, on the other side, spend
more time on other types of activities, which are often informal, to mitigate
significant loss of income, increasing the share of self employed and informal
individuals in the labour market (Bozzoli et al., 2013).
Another way of measuring how violence affects the labour market is look-
ing at how it influences wages and earnings. The effect here can be either
positive or negative depending on the context of analysis. The impact can
be positive in violent areas if the displacement of individuals to non-violent
areas results in a reduction in labour supply, thus pushing wages up. This is
commonly known in the literature as a compensating wage differential effect,
where firms offer a pay premium for risk to attract workers (Rosen, 1986).14
Smith Kelly (2011) analysed the compensating wage differential effect of crime
in Miami and found that a rise in crime rates led to high-crime-risk workers
earning a higher per hour relative wage than high-crime-risk workers in other
cities. But often the effect cannot be identified clearly as there might be un-
observables that affect wages which make it difficult to isolate the impact. For
example, Braakmann (2009) using three way component estimators to control
for individual and regional heterogeneity found that wages are not affected by
changes in both violent and non-violent crime rates. Alternatively, high levels
of violence can have a negative effect on wages and earnings in non-violent
areas as a result of displacement of individuals from violent ones, which cre-
ates an oversupply of labour (Atuesta and Paredes, 2015; Calder´on-Mej´ıa and
˜ez, 2015).
Ib´an
Considering that Mexico is a country that has a dual labour market, this
is that almost 60% of the workers are employed in the informal sector of
the economy, according to most recent statistics from the National Institute
of Statistics (INEGI).15 Analysing if violence and the presence of criminal
groups affects these sectors differently becomes an interesting exercise. One
can think that given the flexible nature of informal jobs, this sector would be
more elastic to any external influence, workers would simply reduce working
hours to reduce exposure. This mechanism works well in the case of self-
employed individuals (see Fern´andez et al. (2014); Vel´asquez (2014)), but the
effect is less clear for wage earners. On the other hand, formal wage earners
14
According to Rosen (1986) the actual wage under these conditions can be also consid-
ered a negative price for the job paid by the firms to workers.
15
Informal workers possess no social security or any of the job benefits that come with
being formally employed.
8
would not be able to reduce working hours as formal firms are less likely to
respond to episodes of violence in this way, but rather increasing wages when
the supply if labour decreases.
Finally, an alternative way of thinking about the impact of the presence
of Drug Trafficking Organizations is that these groups hire local labour force.
According to R´ıos (2010) two of the main DTOs in Mexico, Sinaloa and Golf,
opened their recruitment process to outsiders in the early 2000s.16 Such groups
transmitted radio adds and posted messages in the main border cities of Mex-
ico, encouraging “brave men” to join their organization. This has have a
positive spillover effect on wages that would most likely by reflected in the
informal sector of the labour market.17
The contribution of this paper then is to measure and explain the differ-
ential impact of violence and presence of Drug Trafficking Organizations on
wages within both the formal and informal labour market and to what extent
the presence of criminal groups affects the wages and working hours of indi-
viduals.
4 Data and empirical strategy
4.1 Homicides, Drug Trafficking Organizations and the
Mexican Family Life Survey
Data for the empirical analysis comes from three sources: First, individual
level information from the three waves of the Mexican Family Life Survey
(MxFLS). Second, number of monthly deaths by intentional homicides at the
municipal level available from 1990 to 2013 from the National Institute of
Statistics and Geography (INEGI). Third, the number of drug trafficking or-
ganizations by year per municipality is obtained from Coscia and R´ıos (2012).
The MxFLS is a longitudinal panel survey that covers different aspects
of household members and is representative at the municipal level and con-
ducted for the years 2002, 2005 and 2009.18 It includes information for 8,400
households and almost 35,600 individuals for 16 states throughout Mexico.19
The survey contains individual and household information, the type of job,
monthly wages, position in the job, industry, number of co-workers and if the
16
Before this, DTOs membership was reserved only for family and close friends of the
leaders.
17
This can be possible even if this is not directly expressed in the self-reported information
of wages in the household survey.
18
Some of the topics being covered are health, education, migration, labour, income and
access to government programs. One of the main characteristics of this survey is the low
attrition rate from one wave to the next. 89% of individuals were recontacted from 2002
to 2005 and 85% of individuals from 2005 to 2009. This was possible because the design
of the survey allowed interviewers to track individuals if they moved out of their original
place of residence after the first wave in 2002.
19
The total number of States is 31 plus Mexico City or Federal District.
9
person has access to social security. It also contains information on personal
characteristics such as if the person is the head of household, schooling level,
age, gender and marital status. Information at the household level is also
collected such as household size, number of children under 14 years and num-
ber of elderly in the household. Additional to this it contains information on
community size and location which makes possible to construct the identifier
to merge with the dataset on homicides and number of DTOs.
The timing of the MxFLS survey is particularly apposite for the purpose
of this analysis as the first and second waves were conducted in 2002 and 2005,
years before President Calderon took office, which is also a period of relative
stable levels of violence. The third wave was conducted between 2009 and
2010, the period where the homicide rates reached its highest point as ob-
served in figure 4. The analysis is carried out on the extensive margin rather
than intensive one; focusing on both formal and informal wage earners. Ex-
cluding non-paid workers, the self-employed, owners or employers, retired or
those working on agricultural activities for self-consumption. Table 1 contains
the descriptive statistics for the years of analysis. From this table is worth
pointing out several aspects. On average, 65% of the sample is comprised of
male workers, with elementary or secondary schooling. The share of informal
workers is large compare to their formal counterparts, although is decreasing
over time.
The information for homicides comes from INEGI. A detailed monthly re-
port of intentional homicides for all 2,457 Mexican municipalities from 1990
to 2015. The rate per 100,000 is calculated using yearly population figures
from the Mexican Census. This information is used to measure the presence
of violent crime in the place of residence as it has less issues of under-reporting
compared to other types of crime. The total number of homicides is not just
a result of the war on drugs and this information might not be providing the
most accurate effect of violence related to this phenomena, but given that
this information has been used in other studies, it is also used here as it is
disaggregated at the municipal level. Given that this information spans from
1990 to 2010, an average has been constructed covering from 1990 to 2001
and this would be used as an instrument for current levels of homicides. More
details of this in subsection 4.3.
The information regarding illegal activities carried out by Drug Trafficking
Organizations is either non-existent, restricted by the authorities or unreliable.
For this reason, often researchers rely on the homicide rates as a proxy for
the violence caused by the confrontations between different criminal groups
and as a way to measure the impact that these groups have on several aspects
of interest. However, the media reports the activities of such groups when
this implies a violent event. This reporting of activities have led to some re-
searchers to quantify the presence of DTOs, based on the content found on
the web. This is the case of Coscia and R´ıos (2012). This unique dataset is
the result of a text analysis algorithm designed to obtain information from
the web to identify where criminal groups operate. It is possible as it ex-
10
tracts information reading digitalized news, blogs and Google-News indexed
content, Google is used as it organizes reliable sources of information such as
newspapers and blogs that belong to the media.
The objective is to identify a number of hits or mentions per actor, the
actors are municipalities and DTOs, so the algorithm yields different com-
binations of hits of the actors. In this case when a local newspaper reports
the presence of a DTO in a given municipality. The outcome of the rigorous
analysis is a dataset containing information for 13 DTO’s in Mexico for the
period of 1990-2010, disaggregated up to the municipal level. According to
Coscia and R´ıos (2012) DTOs only operate in 713 of 2,441 municipalities in
Mexico. Leaving large areas of the country practically without the presence of
these criminal groups. There is temporal variation in the data, as some DTOs
appear in municipalities for most of the years in the period analyzed which is
the case for the large DTOs and others only appear until recently and these
groups were created when the fragmentation of the DTOs happened. Also,
according to Coscia and R´ıos (2012) there appears to be a clustering of the ar-
eas of operation for DTOs and as a result, many municipalities of the country
remain untouched by the DTOs throughout the period of analysis. Another
important aspect of the data is that it is precisely after 2006 that a growth of
the number of mentions takes place, approximately 10,000 articles to 100,000
articles in just four years (until 2010), which is consistent with the sudden
increase in homicide rates in Mexico during the same period.
Another aspect that is worth highlighting in this data is that as of 2010,
62% of the municipalities that have a presence of DTOs, have more than
one group operating simultaneously. Including information on the number of
DTOs per municipality allows to draw conclusions on the effect of the pres-
ence of these criminal organizations on the labour market. The effects are not
limited to the violence brought about by these organizations but more about
the sole presence of more than one group per municipality, compared to the
presence of just one or none. Additionally, an indicator of the average presence
of DTOs per municipality is constructed covering the period 1990-2001, and
it is used to control for the current presence of these criminal organizations
per municipality. This will be discussed with more detail in section 4.3.
4.2 Understanding how the fragmentation of DTOs
spreads violence
It is important to highlight that the presence of DTOs not necessarily means
violence. In fact, a municipality that has the presence of only one DTO tends
to have relatively less homicides compared to those that have more than one.20
One of the main assumptions made in this analysis relates to way DTOs be-
have before the year 2006 and after. According to Robles et al. (2013), DTO
20
See Castillo et al. (2013) for a detailed analysis of the increase in homicide rates in
Mexican municipalities when comparing the presence of one DTO against more than one.
In fact, they conclude that the presence of one DTO does not predict homicide rates whereas
for more than one the effect is strong and positive.
11
can behave as stationary bandits or roving bandits. The stationary bandits’
main characteristic is to hold control over a certain area in the long term.
The rationale behind this is that such groups pursue long term gains, they
might favour the growth and expansion of the criminal organization. In the
Mexican context, these groups are commonly those that are large in number
of members and have an important presence throughout the country, those
that had made historical pacts with the government and were abiding by the
rules that had been set from the beginning.21
Roving bandits on the other hand, just have temporary domain. They
extort, kidnap and murder to enhance short term gain.22 They behave this
way because they are interested in gaining immediate territorial power and
violence is their tool to acquire this (O’Neil, 2009; Dıaz-Cayeros et al., 2011).23
In this analysis it is assumed that until the year 2006 before the govern-
ment sent the army into the streets, DTOs behave like stationary bandits as
the equilibrium of power is kept by those involved in the traffic of drugs to the
United States. However, after the year 2006, they behave like roving bandits,
as the confrontation against the government and the consequent fragmenta-
tion of major DTOs brought about the sharp increase in violence. There is
empirical evidence by Osorio (2015) supporting that the government’s strat-
egy weakened major DTOs and motivated the invasion of neighboring ones.
This effect is particularly strong in areas with high density of these groups.
More specifically, both the intensification of the government strategy and the
increasing number of DTOs are positively associated with the severity of vio-
lence between the groups. This evidence is consistent with findings by Lindo
and Padilla-Romo (2015) who suggests that the capture of a DTO leader in-
creases the homicide rate in a municipality by 80%. However, the presence
of many groups per municipality also implies that these groups are in need of
labour and these communities can supply it.
4.3 Identification Strategy
The analysis starts by estimating an OLS regression to measure the impact of
homicides and presence of DTOs on individual wages in a given municipality.
This is specified in equation 1.
21
These are major DTOs such as Cartel de Sinaloa, Cartel del Golfo, Cartel Beltran-
Leyva. Also defined by Castillo et al. (2013) as the traditional groups.
22
These are commonly new groups such as Cartel de Jalisco Nueva Generaci´ on, Los
Caballeros Templarios and Los Zetas, to mention a few. Also defined as Castillo et al.
(2013) as the competitive and expansionary groups.
23
In their study, Robles et al. (2013) argue that there is evidence that DTO behave one
way or the other, citing the case of the Cartel de Tijuana in 2010 which split into two
factions. One faction was led by Teodoro Garc´ıa Simental (aka El Teo), who favoured
kidnappings in Tijuana. The other faction was led by Luis Fernando S´ anchez Arellano (aka
The Engineer ) who wanted to focus more on the traffic of drugs fearing that other types
of crime such as kidnapping local businessmen would attract too much attention from the
government. After the arrest of El Teo, the faction led by Arellano, regained control of the
group and the peace was reestablished in Tijuana after multiple confrontations between the
two factions.
12
LnWij = β0 + β1 Xij + β2 Sij + uij (1)
Where LnWij is the log of the real wage of the individual i in a municipal-
ity j 24 and it is regressed on Xij which contains personal characteristics such
as age, schooling level, gender, the industry classification of the worker and
Sij is a control for five Mexican regions in Mexico.25 uij is the error term.
The main concern when identifying the impact of violence and presence of
DTOs on wages as specified in equation 1 is the presence of reverse causality.
It can be argued that the presence of criminal groups is not exogenous to a
municipality, this would imply that the location of DTOs is correlated with
unobservables that also affect wages. Instead, they will locate in municipali-
ties with better economic performance as this would allow for more extraction
of rents. These conditions would attract more criminal groups. On the other
hand, homicide rates would increase only in those municipalities where DTOs
are located which are also the municipalities with better economic perfor-
mance.
However, homicides and the presence of DTO had remained relatively sta-
ble throughout the years at the national level (as can be observed in figure
1). Moreover, the location of DTOs rather than being determined by the eco-
nomic prosperity of a municipality is determined by the access to entry points
to the United States. 26 Additionally, after the Mexican government launched
the offensive to tackle such criminal groups in the year 2006, the confronta-
tions between the government and criminal groups on the one hand and the
fragmentation of the large DTO and the consequent fight to gain territorial
power on the other hand, brought about the drastic increase in violence and
the number of DTO trying to access power and the control of drug trafficking
routes, as a result many groups engaged in other activities such as kidnapping
and extortion, affecting the local population.
To overcome the issue of reverse causality, the instrumental variable pro-
posed by R´ıos (2015) is used. This is estimated by a Two-Stage Least Squares
estimation strategy (2SLS). The presence of DTO and the log homicide rate
in a municipality j is instrumented using the average presence of DTO and
average homicide rate per municipality; both instruments are for the period
of 1990-2001.27 The logic behind these instruments is that criminal groups
historically locate in municipalities that are important in terms of the traffic
of drug regardless of economic conditions. To access to such positions, the
new groups that flourished as a consequence of the war on drugs in 2006, had
24
The wage is deflated using the Mexican consumer price index for 2010.
25
The regions were classified based on the INEGI’s classification: north, south, centre,
east and west.
26
See Castillo et al. (2013) for a detailed explanation of location of DTOs in Mexican
municipalities.
27
The average is taken from 1990 until 2001, as this is a year before the first wave of the
MxFLS was conducted.
13
to fight and then violence increased. The exclusion restriction holds if the
historical location of DTO correlates with the current presence of more than
one criminal group per municipality j, but it is uncorrelated with unobserved
factors affecting wages in the current period i. The same criteria applies to
the homicide rates, as the average homicide rates in a municipality would be a
correlated with the current rate in a municipality, but is not with the current
level of wages for an individual or unobserved factors affecting it.
The first stage of the regression is then specified in equation 2.
Vij = π0 + π1 Z1990−2001 + π2 + εij (2)
Where V ij is the variable of primary interest and represents the log of
homicide rates per 100,000 inhabitants, as used commonly in the literature
it also represents the number of DTO, both at the municipality level.28 The
historical average of violence is measured by Z1990−2001 . Finally εij is the error
term.
Instrumenting current levels of violence with historical average would yield
non-biased estimates of the violence and presence of Drug Cartels on wage lev-
els. All of the regressions are clustered at the municipal level. To deal with the
potential issue of heteroskedasticity, a Limited Information Maximum Likeli-
hood (LIML) procedure is implemented, which is robust to clustering.
The second stage then is estimated via equation (3).
LnWij = δ0 + δ1 Vc
ij + δ2 Xij + υij (3)
where LnWij is the log of the wage of the individual i in a municipality j
and it is regressed on the predicted values of Vcij from the first stage and Xij
which contains personal characteristics such as age, schooling level, gender
and the industry classification. υij represents the error term.
Additionally, the instrument proposed by Castillo et al. (2013) is also used
in this exercise, which is the result of the interaction between two variables:
negative supply shocks of cocaine (which provides the temporal variation) and
distance to the nearest point of entry of cocaine to Mexico or point of exit
to the US (which provides the spatial variation), the major consumer of this
product. Cocaine supply shocks are measured by the amount of cocaine seized
by the Colombian Government, which shifted its drug interdiction strategy
from 2006 and this affected the value of the amount of cocaine supplied to the
28
Rios (2015) also uses the log of homicide rates and details that because the many zeros
in this variable are important in the estimation to compare between municipalities where
there is high presence of violence versus those that have almost none, a transformation is
made; where the log is calculated as the homicide rate plus one.
14
intermediaries or the DTOs in Mexico. According to Castillo et al. (2013) the
Colombian government shift in strategy, in 2006, targeted less the eradication
of coca crops which are considered low value added and instead focused on the
interdiction of drug shipments and destruction of coca processing labs. Under
the assumption that the demand for drugs is inelastic,29 a contraction of sup-
ply derives in an increase in drug trafficking activities because the DTOs will
try to access the product even if it is scarce and costs more, this ultimately
affects the levels of violence. In figure 5 the homicide rate in Mexico is plotted
against cocaine seizures in Colombia, it can be observed that there is a pos-
itive correlation between cocaine seizures and homicide rates, specially after
2006.
The increase in drug trafficking activities and levels of violence as a result
of the reduction of supply of cocaine, happens only in localities that are valu-
able for the smuggling of cocaine into Mexico and to the United States. This
localities are the ones close to the north and south border and ports in the Pa-
cific Ocean and Gulf of Mexico, these localities give a comparative advantage
for the trade of drugs, so these locations are valuable for the DTOs. Figure
6 shows the geographic coordinates of such points of entry in the south bor-
der, ports in the Pacific Ocean and Gulf of Mexico and the exit points to the
United States. Comparing this map with the location of DTOs in 3 specially
in the year 2010 and the homicide rates shown in 2 it can be observed that
there is a correlation between the location of these points and the location of
DTOs and homicide rates. Identification of the effect of supply shocks on the
number of DTOs and homicide rates comes from the interaction of the two
variables.
5 Preliminary Results
When analyzing the impact of violence on local wages, the interest lies in the
the impact on those that remain in a municipality and absorb the violence
shock by adapting to the new circumstances. In a way it can be perceived as a
supply and demand problem. This would be true if the movement of workers
out of the municipalities with high levels of violence is enough to push wages
up for those that remain. Since the survey is successful in following individu-
als from the first to the last wave, those that move to a different municipality
from one wave to the next are dropped from the sample, which only accounts
for 1.4% of the total sample. In this way it is ensured that the analysis is
done on the stayers.
29
As detailed in Castillo et al. (2014) there is enough evidence to support the claim that
demand for drugs is inelastic, this assumption is central to the use of this instrumental
variable.
15
5.1 Instrumentation of violence and presence of Drug
Trafficking Organizations
I will first investigate how the average violence from previous years predicts
current levels of violence, adding controls for state income per capita and re-
gional controls. The results of the first stage are presented in table 2, the
values for the F statistic are displayed at the bottom of the table and all are
well above 10, which confirm that the instrument is good. Columns (1) and
(4) present the results for the full sample, columns (2) and (5) present results
for informal workers and columns (3) and (6) present results for formal work-
ers. From the results of the first stage it can be concluded that instruments
are both strong predictors of homicides and the presence of DTOs. State in-
come per capita does not seem to be have any effect on the number of DTOS
which is consistent with the assumption that location of criminal groups and
homicides do not relate to economic performance but to other factors. More
specifically, everything else constant, an increase in State income per capita
can be associated with a decrease in the homicide rate per 100,000 for Mexican
municipalities. States with larger income per capita are less prone to have high
levels of homicides at least in this context and economic performance is not
statistically significant when determining the location of the DTOs, although
the coefficient is negative. This might be the case because better economic
performing states have better functioning institutions and more enforcement
of the rule of law.
Additionally, these results are consistent with the literature on the rela-
tionship between crime and economic performance which has found that this
relationship is negative. Detotto and Otranto (2010) suggest that the effect
is stronger during recessions. Enamorado et al. (2016) shows that this effect
might be larger when analyzing drug-related crimes compared to more ‘com-
mon’ types of crime.
The results of the second stage of the 2SLS regression are presented in
table (3). The the value of Kleibergen-Paap test for underidentification is
displayed at the bottom of the table, the values confirm that the estimation is
not underidentified. Columns (1) to (3) show that, after the instrumentation
of the average homicides per municipality, the effect of homicides on wages is
not statistically significant. There is no evidence at least under this setting
that high levels of homicides have any significant effect on individual wages.
Comparing this result with existing ones in the literature, Braakmann (2009)
finds no significant impact of crime on wages for Germany for both violent and
non violent types of crime. Wages under this setting, seem to be determined
more by personal characteristics, schooling and the industry of the worker.
On the other hand, after the instrumentation, the effect of the presence
of DTOs remains robust to the specification. Holding everything else con-
stant, an additional DTO per municipality increases the wages of individuals
by 8.8 percentage points.30 A word of caution must be put here and it re-
30
The results of the OLS are presented in table (10). The results yield a positive impact
16
lates to the interpretation of this coefficient. This variable does not explicitly
measure violence related to DTOs, it is merely representing a count of the
number of these organizations per municipality. However, it can be assumed
that it is only after 2006 when these groups engage in confrontations with
the government and with other groups. Only after 2006 is when it can be
considered as a measure of violence as a result of the war on drugs. The pro-
liferations of DTOs in Mexican municipalities can have spillover effects on the
local economy as an expanding organization would need to hire labour force
or simply have an influence on the economy via consumption. For this rea-
son, the interpretation of the effect of this variable must be done with caution.
Further dividing the sample between formal and informal workers, columns
(5) and (6) of table (3) present the results for this estimation, it can be ob-
served that an additional DTO per municipality increases wages of informal
workers by 5.6 and the wage of formal workers by 9.6 percentage points. This
coefficient is larger for informal workers and this can be the case because the
flexibility associated to the type of jobs in this sector makes wages more elastic
to external shocks. It is often the case that most contracts in informal jobs
are a mere verbal arrangement between the employer and employee. This
flexibility is also applied when the wage is agreed. So compared to the formal
sector, informal jobs are less resilient.
Homicides and the presence of DTOs are now instrumented combining the
historic average from 1990-2001 and the instrument from Castillo et al. (2013)
which interacts the supply shock with the location of entry points in the south
border and for the entry ports in the Pacific Ocean. Furthermore, given the
comparative advantage for the traffic of drug, it is expected that the negative
shock to the supply of cocaine will affect more those municipalities that are
closer to the entry points, compared to those that are far. This would imply
more homicides and more presence of DTOs in these municipalities. The re-
sults of the first stage of the preferred specification are presented in table (4).
It can be observed that the three instruments predict positively the homicide
rates and the number of DTOs. The interaction of supply shocks and entry
points in the Pacific ocean is not statistically significant for the case of homi-
cide rates. Additionally, it can be observed that the F-statistic is well above
10 (according to the rule of thumb) only for the case of number of DTOs but
not for homicide rates.
Results for the second stage of the estimations are presented in table (5).
Values for the test of underidentification and overidentification are presented
at the bottom of the table. On the one hand, all the values for the Kleibergen-
Paap test suggest that the models are not underidentified. On the other hand,
the values for the Hansen-J statistic and its respective p-value, suggest that
the models are not overidentified. Results are also presented for the full sam-
ple, and for both formal and informal workers. It can be observed that the
of DTOs of 4.2 percentage points for the number of DTOs and no effect for homicide rates.
Furthermore, for informal workers, ceteris paribus, the effect is an increase in wages of 3.7
percentage points. But no effect for formal workers.
17
results are similar to those presented in table 3, homicides do not seem to have
any statistically significant effect on wages, whereas the presence of DTOs in
a municipality do affect wages. More specifically, an additional DTO per
municipality has a positive effect on wages of 3.4 for formal workers and 4.9
percentage points for informal workers.
A precision about the estimations has to be made here. The estimation
was conducted using the instrument by Castillo et al. (2013) without combin-
ing it with historical averages and no effects appear. It was also conducted
using different combination of the instruments, but only the combination of
historical averages with supply shocks in the south, Pacific and Atlantic Ocean
give statistically significant results. The exercise was also conducted using the
cocaine seizures and distance to the nearest border without interacting them.
When the cocaine seizures is used as an instrument alone, wages experience a
positive effect of 1.9 percentage points but this effect does not hold when the
regression is separated by formal and informal workers. It is acknowledged
that given that the analysis is done only on three periods of time, out of
which only the last period experiences high levels of violence and the presence
of many DTOs per municipality the variation attributed to the rise in levels
of violence might not be enough to affect the wage levels.
If we believe that the impact of DTOs presence on wages is true, we can
think that this is not only affecting the wages, but also the number of hours
worked. For this reason, the model suggested in equation (3) is also used here
but in this case the dependent variable is the log of weekly hours worked. The
results of this estimation are presented in table 6. The table shows the same
pattern as with wages, no effect of homicides on working hours but there is an
effect of the number of DTOs. Using the instrumental variable of historical
average, columns (3) to (6) show that the number of DTOs increases weekly
working hours by 3.2 percentage points for informal workers but no effect for
formal workers. When using the historical averages instrument combined with
the supply shocks instruments as done in table (5) columns (10) to (12) show
that working hours increase by 2.8 percentage points for formal workers but no
effect for informal workers. This results is somewhat surprising given that it
has been established that informal jobs are more elastic to shocks compared to
formal ones. However, it can be the case that many informal workers search
for more formal jobs in firms as a coping mechanism. These formal firms
presumably react to changes in violence or the presence of DTOs, increasing
expenditure in security measures. However I cannot formally test for this as
there is no firm information available in the survey.
5.2 Selection of individuals into formal and informal
jobs
It can be argued that there is an issue of selection of individuals into both for-
mal and informal jobs and this affects the wages of individuals. Some workers
might prefer be formal as often the wages and benefits associated are higher
compared to informal jobs. On the other hand, individuals might prefer infor-
18
mal jobs due to the flexibility of working hours, the proximity to their homes
even if this means sacrificing income. Moreover, in the sample used here, in-
dividuals switch from a job in one sector to another sector from one period
to the other. There is a possibility that at the household level some mem-
bers join the labour force as a result of the levels of violence and as a coping
strategy, commonly known as the added worker effect. This is validated by
some studies where the informal sector has proven to be a good alternative
when individuals live in areas affected by high levels of violence, although this
evidence refers to the self-employment of individuals (Vel´asquez, 2014; Bozzoli
et al., 2013).
Moreover, if there is an issue of selection of individuals here, this might
be biasing the results and it can be argued that the positive impact of the
presence of DTOs on wages is just a result of the selection that drives wages
up. For this reason and as as robustness check, a Heckman (1979) two step
procedure is proposed to correct for this issue. The first stage is defined as:
Pj =f (Age, Gender, Schooling level, Marital status,
Number of children under 14 years in the HH, HH size, (4)
Number of older that 65 in the HH), j = 1, 2
The second stage is then:
ln(wages) =g(Age, Gender, Schooling,
Violence corrected, regional controls, industry controls, (5)
Correction terms), iff,
Where equation (4) represents the first stage of the estimation and is the
employment selection function. It contains variables that are commonly used
in the literature to predict employment decisions such as the number of chil-
dren under 14 years of age in the household, the household size, the number
of adults over 65 years of age that are unemployed, the marital status, age,
and gender. Equation (5) is the wage equation, this specification includes
the variable violence instrumented from equation (1) and the correction term
from the first stage and personal characteristics controls.
Results for the first stage of the estimation are presented in table (8). It
can be observed, that age and schooling all have the expected sign. From the
controls used, only being married and living in a household with elderly have a
negative impact on the probability of being informal. The coefficients for age
indicate that the probability of being informal is negative for younger individ-
uals at early stages of their working career and then becomes positive with age.
The results for the wage equation after controlling for selection selection
bias to either formal or informal jobs are presented in table (9). The coef-
ficient of the inverse mills ratio represented as lambda reveals that there is
19
evidence of strong positive selection of individuals into informal jobs. This
means that individuals have a strong preference for the sector they want to
work in and this biases the results of the estimation. The interpretation of
this coefficient follows J. (1988) and Reilly (1991) and it refers to the effect of
the selection variable on the wage. The effect is obtained multiplying minus
the selection variable coefficient by the mean value of the selection variable.
For the case of informal jobs, the calculation suggests that those self-selecting
into the informal sector earn on average 51.0% lower wages that an individual
drawn at random from the labour force with identical observable characteris-
tics would be expected to earn. It is not surprising that workers have a strong
preference for informal jobs in Mexico, as many firms (that are often formal)
hire workers informally to avoid paying benefits to workers that come with a
formal contract. In the context of this estimation, the sign of the coefficient of
the impact of the presence of DTOs holds, the corrected term from (3) is also
used here. The magnitude of the coefficient, on the other hand, is marginally
larger without correcting for selection bias. The positive effect of the presence
of DTOs on wages persist confirming what was found when using the IV pro-
cedure. Ceteris paribus, an additional DTO in a municipality, increases the
wages of individuals by 8.7 percentage points.
6 Preliminary Conclusions
It has been the aim of this analysis to measure to what extent episodes of
high levels of violence and presence of Drug Trafficking Organizations affect
the local wages of individuals. The compensating wage differential argument
states that workers that face unfavorable working conditions, such as the risk
of death, will be compensated by an increase in wages (see Rosen (1986) for
full details). Such increase is the price firms pay to attract workers under
difficult conditions. I have tried to explore whether the sudden increase in
violence brought about after the start of the “war on drugs in Mexico”, had a
compensating wage differential effect for individuals in Mexican municipalities.
The results of this analysis yield many interesting results. From the first
stage of the IV procedure it can be concluded that better economic perfor-
mance of a state does not solely determine the location of DTOs, there are
other things that are influencing this location. After instrumenting the pres-
ence of Drug Trafficking Organizations (DTOs), confirms that this increases
the wages of workers by 8.8 percentage points. Further dividing the sample
between formal and informal workers The results yield an increase of 9.6 for
informal workers and 4.3 percentage points for formal workers. As expected
the impact is larger for informal workers as the flexibility of the working con-
ditions for this individuals make wages more elastic. On the other hand, no
effect is found of homicides on wages. At least under this setting the increase
in homicides does not seem to be affecting the wages of individuals in Mexico.
The fact that the presence of DTOs do have a positive impact on wages might
be signaling that these groups to some extent boost economic activity and this
has an impact on wages, mostly in the informal sector of the labour market.
20
On the other hand, after correcting the model for the presence of self-
selection that results from the occupational choice of individuals using a Heck-
man (1979) two step procedure, the results show that there is strong evidence
of self-selection into informal jobs. However, the sign of the coefficient pre-
sented after instrumenting DTOs presence, holds. Everything else constant,
an additional DTO in a municipality increases wages of informal workers by
7.0 and the wage of formal workers by 4.6 percentage points.
21
References
Abadie, A. and Gardeazabal, J. (2003) The economic costs of conflict: A case
study of the basque country, American Economic Review, 93, 113–132.
Ashby, N. J. and Ramos, M. A. (2013) Foreign direct investment and indus-
try response to organized crime: The mexican case, European Journal of
Political Economy, 30, 80 – 91.
Astorga, L. (2001) The limits of anti-drug policy in mexico, International
Social Science Journal, 53, 427–434.
Astorga, L. and Shirk, D. A. (2010) Drug trafficking organizations and
counter-drug strategies in the u.s.-mexican context, Center for U.S.-
Mexican Studies. UC San Diego: Center for U.S.-Mexican Studies. Re-
trieved from: http://escholarship.org/uc/item/8j647429.
Atuesta, L. H. and Paredes, D. (2015) Do mexicans flee from violence? the
effects of drug-related violence on migration decisions in mexico, Journal of
Ethnic and Migration Studies, pp. 1–20.
Barrera, F., Ib´anez, A. M. et al. (2004) Does violence reduce investment in
education?: A theoretical and empirical approach, Tech. rep., UNIVERSI-
DAD DE LOS ANDES-CEDE.
Becker, G. S., Rubinstein, Y. et al. (2004) Fear and the response to terrorism:
an economic analysis, University of Chicago mimeo.
BenYishay, A. and Pearlman, S. (2013) Homicide and work: The impact
of mexico’s drug war on labor market participation, Available at SSRN
2302437.
BenYishay, A. and Pearlman, S. (2014) Crime and microenterprise growth:
Evidence from mexico, World Development, 56, 139–152.
Bozzoli, C., Br¨
uck, T. and Wald, N. (2013) Self-employment and conflict in
colombia, Journal of Conflict Resolution, 57, 117–142.
Braakmann, N. (2009) Is there a compensating wage differential for high crime
levels? first evidence from europe, Journal of Urban Economics, 66, 218–
231.
Braakmann, N. (2012) How do individuals deal with victimization and vic-
timization risk? longitudinal evidence from mexico, Journal of Economic
Behavior & Organization, 84, 335–344.
Brown, R. and Vel´asquez, A. (2015) The effect of violent conflict on the human
capital accumulation of young adults.
Cabral, R., Mollick, A. V. and Saucedo, E. (2016) Violence in mexico and its
effects on labor productivity, The Annals of Regional Science, 56, 317–339.
22
Calder´on, V., G´afaro, M. and Ib´an
˜ez, A. M. (2011) Forced migration, female
labor force participation, and intra-household bargaining: does conflict em-
power women?, Documento CEDE.
Calder´on-Mej´ıa, V. and Ib´an
˜ez, A. M. (2015) Labour market effects of
migration-related supply shocks: evidence from internal refugees in colom-
bia, Journal of Economic Geography, p. lbv030.
Camacho, A. and Rodriguez, C. (2013) Firm exit and armed conflict in colom-
bia, Journal of Conflict Resolution, 57, 89–116.
Castillo, J. C., Mejia, D. and Restrepo, P. (2013) Illegal drug markets
and violence in mexico: The causes beyond calder´on, Versi´on electron-
ica: http://cie. itam. mx/SEMINARIOS/Marzo Mayo 2013/Mejia. pdf, 3,
2013.
Castillo, J. C., Mej´ıa, D. and Restrepo, P. (2014) Scarcity without leviathan:
the violent effects of cocaine supply shortages in the mexican drug war,
Available at SSRN 2409268.
Collier, P. and Duponchel, M. (2013) The economic legacy of civil war firm-
level evidence from sierra leone, Journal of Conflict Resolution, 57, 65–88.
Coscia, M. and R´ıos, V. (2012) Identifying where and how criminal orga-
nizations operate using web content, CIKM-12, octubre-noviembre, Maui,
USA.
Dell, M. (2015) Trafficking networks and the mexican drug war, American
Economic Review, 105, 1738–79.
Detotto, C. and Otranto, E. (2010) Does crime affect economic growth?, Kyk-
los, 63, 330–345.
Dıaz-Cayeros, A., Magaloni, B., Matanock, A. and Romero, V. (2011) Living
in fear: Mapping the social embeddedness of drug gangs and violence in
mexico, Manuscript: University of California at San Diego.
Driffield, N., Jones, C. and Crotty, J. (2013) International business research
and risky investments, an analysis of {FDI} in conflict zones, International
Business Review, 22, 140 – 155.
Dube, A., Dube, O. and Garc´ıa-Ponce, O. (2013) Cross-border spillover: Us
gun laws and violence in mexico, American Political Science Review, 107,
397–417.
Enamorado, T., Lpez-Calva, L. F. and Rodrguez-Casteln, C. (2014) Crime
and growth convergence: Evidence from mexico, Economics Letters, 125, 9
– 13.
Enamorado, T., Lpez-Calva, L. F., Rodrguez-Casteln, C. and Winkler, H.
(2016) Income inequality and violent crime: Evidence from mexico’s drug
war, Journal of Development Economics, 120, 128 – 143.
23
Fajnzlber, P., Lederman, D. and Loayza, N. (2002) Inequality and violent
crime, JL & Econ., 45, 1.
Fern´andez, M., Ib´an
˜ez, A. M. and Pe˜
na, X. (2014) Adjusting the labour supply
to mitigate violent shocks: Evidence from rural colombia, The Journal of
Development Studies, 50, 1135–1155.
Gaibulloev, K. and Sandler, T. (2008) Growth consequences of terrorism in
western europe, Kyklos, 61, 411–424.
G´omez, E. A. (2012) Drug-related violence and forced migration from mexico
to the united states.
Heckman, J. J. (1979) Sample selection bias as a specification error, Econo-
metrica, 47, 153–161.
J., G. (1988) An analysis of public- and private-sector wages allowing for
endogenous choice of both government and union status, Journal of Labor
Economics, 6, 751.
Justino, P., Leone, M. and Salardi, P. (2013) Short-and long-term impact of
violence on education: The case of timor leste, The World Bank Economic
Review, p. lht007.
Kondylis, F. (2010) Conflict displacement and labor market outcomes in post-
war bosnia and herzegovina, Journal of Development Economics, 93, 235–
248.
Leon, G. (2012) Civil conflict and human capital accumulation the long-term
effects of political violence in per´
u, Journal of Human Resources, 47, 991–
1022.
Lindo, J. M. and Padilla-Romo, M. (2015) Kingpin approaches to fighting
crime and community violence: Evidence from mexico’s drug war, Working
Paper 21171, National Bureau of Economic Research.
Lozano, H. G. and Aleman, S. O. (2013) Does violence affect migration flow?
Maher, D. (2015) The fatal attraction of civil war economies: Foreign direct
investment and political violence, a case study of colombia, International
Studies Review, 17, 217–248.
O’Neil, S. (2009) The real war in mexico: How democracy can defeat the drug
cartels, Foreign Affairs, 88, 63–77.
Osorio, J. (2015) The contagion of drug violence: Spatiotemporal dynamics
of the mexican war on drugs, Journal of Conflict Resolution.
Reilly, B. (1991) Occupational segregation and selectivity bias in occupational
wage equations: an empirical analysis using irish data, Applied Economics,
23, 1–8.
24
R´ıos, V. (2010) To be or not to be a drug trafficker: Model-
ing criminal occupational choices, Unpublished working paper. De-
partment of Government, Harvard University. www. gov. harvard.
edu/files/Rios MPSA2010 TobeOrNotToBe. pdf.
R´ıos, V. (2014) Security issues and immigration flows: Drug-violence refugees,
the new mexican immigrants, Latin American Research Review, 49, 3.
Rios, V. (2015) How government coordination controlled organized crime: The
case of mexicos cocaine markets, Journal of Conflict Resolution.
R´ıos, V. (2015) The impact of crime and violence on economic sector diversity,
Unpublished manuscript.
Robles, G., Calder´on, G. and Magaloni, B. (2013) The economic consequences
of drug trafficking violence in mexico, Poverty and Governance Series Work-
ing Paper, Stanford University.
Rodriguez, C. and Sanchez, F. (2012) Armed conflict exposure, human capital
investments, and child labor: Evidence from colombia, Defence and peace
economics, 23, 161–184.
Romano, P. et al. (2015) Crime exposure and educational outcomes in mexico,
Tech. rep.
Rosen, S. (1986) Chapter 12 the theory of equalizing differences, Elsevier,
vol. 1 of Handbook of Labor Economics, pp. 641 – 692.
Shemyakina, O. (2011) The effect of armed conflict on accumulation of school-
ing: Results from tajikistan, Journal of Development Economics, 95, 186–
200.
Skaperdas, S. (2002) Warlord competition, Journal of Peace Research, 39,
435–446.
Smith Kelly, C. (2011) Crime and compensating wage differentials: Evidence
from miami in 1980, Journal of Applied Business and Economics, 12, 79–89.
Snyder, R. and Duran-Martinez, A. (2009) Does illegality breed violence?
drug trafficking and state-sponsored protection rackets, Crime, Law and
Social Change, 52, 253–273.
Vel´asquez, A. (2014) The economic burden of crime: Evidence from mexico,
Job Market Paper, Department of Economics, Duke University, Durham,
NC.
25
Tables and figures
Figure 1: National Homicide rate per 100,000 (2000-2013)
Source: Elaborated using data from INEGI and CNS
26
Figure 2: Homicide rate per 100,000 inhabitants by municipality
27
Source: Elaborated using data from INEGI
Figure 3: Number of Drug Trade Organizations per municipality 2004-2010
28
Source: Author’s elaboration with data from Coscia and R´ıos (2012)
Figure 4: National Homicide rate per 100,000 and MxFLS (2000-2013)
Source: Elaborated using data from INEGI and MxFLS
29
Figure 5: Homicide rate in Mexico and cocaine seizures in Colombia
Source: Elaborated using data from INEGI and ODC
30
Figure 6: Geographic coordinates of entry and exit points in the Mexican
borders
Source: Elaborated using data from INEGI and World Port Source
31
Table 1: Descriptive statistics for the sample of workers by year (means)
Variable 2002 2005 2009
Log of monthly wages 8.11 8.25 8.25
(0.85) (0.78) (0.76)
Age 34.0 36.0 36.0
(13.04) (13.47) (13.41)
*
Male 0.66 0.66 0.62
No education* 0.06 0.06 0.05
Elementary school 0.34 0.32 0.30
Secondary school 0.32 0.32 0.33
High school 0.13 0.15 0.17
More than high school 0.16 0.15 0.16
Share of informal* 0.72 0.65 0.58
Number of DTOs 0.15 0.97 2.28
(0.41) (1.13) (1.78)
Homicide rate per 100,000 8.34 9.45 19.19
(7.56) (10.96) (26.46)
Cocaine seizures (mt) 95.27 162.66 186.09
Observations 3,103 4,925 5,361
Additional Variables
Mean homicide rate (1990-2001) 14.58
(13.51)
Mean DTOs (1990-2001) 0.093
(0.17)
Distance to the US (km) 630.83
(252.97)
Distance to South (km) 1,254.07
(602.47)
Distance to Atlantic (km) 523.80
(454.56)
Distance to Pacific (km) 309.62
(201.62)
*
These values refer to shares of the total. Standard deviations in parenthesis.
32
Table 2: First stage regression for IV
Firts stage: (1) (2) (3) (4) (5) (6)
Violence on instruments Homicides Homicides Homicides DTO DTO DTO
Average homicides 1990-2001 0.0347*** 0.0567*** 0.0303***
(0.0078) (0.0084) (0.0066)
Average number of DTO p/municipality 1990-2001 3.0338*** 3.099*** 3.0079***
(0.6611) (0.7058) (0.6368)
North region 0.363*** 0.324** 0.430*** 0.261 0.268 0.142
(0.133) (0.146) (0.121) (0.275) (0.285) (0.261)
West region 0.0296 -0.0512 0.285* -0.304 -0.145 -0.659*
(0.164) (0.189) (0.145) (0.318) (0.321) (0.344)
East region -0.101 -0.130 0.00485 -0.756*** -0.712** -0.884***
(0.160) (0.168) (0.165) (0.287) (0.296) (0.276)
South region -0.794*** -0.898*** -0.478** -0.598* -0.648** -0.595*
(0.242) (0.274) (0.202) (0.306) (0.291) (0.309)
State income per capita -0.00599** -0.00694*** -0.00314 -0.00102 -0.000689 -0.00112
(0.00232) (0.00262) (0.00210) (0.00288) (0.00278) (0.00305)
Constant 1.661*** 1.865*** 1.211*** 0.273 0.308 0.311
(0.299) (0.307) (0.349) (0.329) (0.323) (0.404)
Observations 13269 8550 4717 13269 8550 4717
F-test 19.52 44.99 20.54 21.06 19.28 22.31
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 3: Impact of Violence on Wages (2SLS Estimation)
(1) (2) (3) (4) (5) (6)
2SLS 2SLS 2SLS 2SLS 2SLS 2SLS
Second Stage Homicides Formal Informal DTO Formal Informal
Male 0.314*** 0.211*** 0.380*** 0.319*** 0.216*** 0.382***
(0.0181) (0.0232) (0.0213) (0.0195) (0.0239) (0.0231)
Elementary School 0.146*** 0.0975 0.128*** 0.156*** 0.121* 0.130***
(0.0305) (0.0623) (0.0352) (0.0291) (0.0664) (0.0344)
Secondary School 0.352*** 0.312*** 0.285*** 0.350*** 0.335*** 0.274***
(0.0377) (0.0630) (0.0418) (0.0332) (0.0677) (0.0401)
High School 0.465*** 0.434*** 0.324*** 0.457*** 0.452*** 0.311***
(0.0427) (0.0691) (0.0468) (0.0404) (0.0740) (0.0459)
More than High School 0.834*** 0.733*** 0.737*** 0.845*** 0.761*** 0.751***
(0.0508) (0.0681) (0.0577) (0.0457) (0.0726) (0.0556)
Directors & Chiefs 0.929*** 0.764*** 0.945*** 0.898*** 0.743*** 0.915***
(0.0541) (0.0823) (0.0691) (0.0514) (0.0831) (0.0692)
Manufacture &Industry 0.394*** 0.256*** 0.380*** 0.377*** 0.239*** 0.367***
(0.0322) (0.0541) (0.0327) (0.0309) (0.0520) (0.0319)
Commerce & Sales 0.337*** 0.268*** 0.301*** 0.254*** 0.198*** 0.228***
(0.0383) (0.0605) (0.0420) (0.0446) (0.0595) (0.0502)
Services 0.336*** 0.292*** 0.304*** 0.301*** 0.277*** 0.258***
(0.0360) (0.0565) (0.0370) (0.0361) (0.0507) (0.0409)
Professional Services 0.666*** 0.543*** 0.619*** 0.582*** 0.490*** 0.532***
(0.0322) (0.0509) (0.0404) (0.0413) (0.0540) (0.0532)
Log of homicide rates -0.0434 -0.0256 -0.0204
(0.0272) (0.0248) (0.0276)
Number of DTOs 0.0888*** 0.0594** 0.0959***
(0.0270) (0.0236) (0.0334)
Constant 6.300*** 7.004*** 6.269*** 6.151*** 6.880*** 6.171***
(0.0966) (0.128) (0.102) (0.0819) (0.102) (0.0953)
Regional Controls Yes Yes Yes Yes Yes Yes
Age controls Yes Yes Yes Yes Yes Yes
Kleibergen-Paap 13.14 8.089 13.139 14.95 12.718 15.947
R-squared 0.314 0.302 0.256 0.303 0.286 0.242
Observations 13269 4717 8550 13269 4717 8550
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
33
Table 4: First stage regression using IV from Castillo et al. (2013)
Firts stage: (1) (2) (3) (4) (5) (6)
Violence on instruments Homicides Homicides Homicides DTO DTO DTO
Average homicides 1990-2001 0.0338*** 0.0293*** 0.0578***
(0.0080) (0.0070) (0.0093)
Average number of DTO p/municipality 1990-2001 2.8139*** 2.8701*** 2.8143***
(0.6832) (0.6651) (0.7459)
Cocaine seizures and south border 4.61e-06 *** 4.73e-06*** 4.84e-06*** 0.0000105*** 0.000011*** 9.78e-06***
(1.44e-06) (1.56e-06) (1.48e-06) (2.00e-06) (1.94e-06) (2.29e-06)
Cocaine seizures and pacific border -1.37e-06 -2.43e-06 2.04e-06 0.0000158*** 0.0000145*** 0.0000168***
(2.19e-06) (2.34e-06) (2.73e-06) (2.82e-06) (2.64e-06) (3.57e-06)
State income per capita -0.0073** -0.0042 0.0011 -0.1780*** -0.1566*** -0.2271***
(0 .0179) (0.0177) (0.00210) (0.0266) (0.0230) (0.03844)
Constant 1.0400*** 1.1592*** 1.211*** -1.4225*** -1.4104 -1.1600**
(0.4224) (0.4110) (0.349) (0.2875) (0.2364) (0.5140)
Region controls yes yes yes yes yes yes
F-statistic 9.12 9.51 18.64 23.75 25.39 17.06
34
Observations 13269 8550 4717 13269 8550 4717
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 5: Impact of Violence on Wages using Castillo et al. (2013)
(1) (2) (3) (4) (5) (6)
2SLS 2SLS 2SLS 2SLS 2SLS 2SLS
Second Stage Homicides Formal Informal DTO Formal Informal
Male 0.314*** 0.211*** 0.380*** 0.318*** 0.214*** 0.381***
(0.0181) (0.0230) (0.0213) (0.0190) (0.0233) (0.0221)
Elementary School 0.149*** 0.102 0.129*** 0.154*** 0.114* 0.129***
(0.0301) (0.0629) (0.0348) (0.0290) (0.0660) (0.0344)
Secondary School 0.354*** 0.317*** 0.285*** 0.351*** 0.329*** 0.279***
(0.0367) (0.0639) (0.0413) (0.0341) (0.0665) (0.0406)
High School 0.466*** 0.437*** 0.323*** 0.460*** 0.447*** 0.316***
(0.0417) (0.0695) (0.0464) (0.0405) (0.0725) (0.0460)
More than High School 0.835*** 0.736*** 0.737*** 0.842*** 0.752*** 0.744***
(0.0492) (0.0688) (0.0568) (0.0459) (0.0717) (0.0552)
Directors & Chiefs 0.938*** 0.767*** 0.951*** 0.913*** 0.754*** 0.932***
(0.0530) (0.0817) (0.0697) (0.0517) (0.0818) (0.0690)
Manufacture &Industry 0.401*** 0.259*** 0.385*** 0.386*** 0.248*** 0.375***
(0.0319) (0.0533) (0.0327) (0.0301) (0.0515) (0.0315)
Commerce & Sales 0.338*** 0.267*** 0.303*** 0.284*** 0.227*** 0.264***
(0.0376) (0.0599) (0.0417) (0.0377) (0.0582) (0.0433)
Services 0.342*** 0.295*** 0.307*** 0.316*** 0.285*** 0.282***
(0.0350) (0.0554) (0.0366) (0.0346) (0.0521) (0.0374)
Professional Services 0.666*** 0.543*** 0.619*** 0.612*** 0.512*** 0.574***
(0.0318) (0.0502) (0.0404) (0.0329) (0.0509) (0.0425)
Log of homicide rates -0.0122 -0.0117 0.00105
(0.0239) (0.0195) (0.0254)
Number of DTOs 0.0573*** 0.0347** 0.0496***
(0.0141) (0.0137) (0.0173)
Constant 6.223*** 6.968*** 6.217*** 6.166*** 6.904*** 6.194***
(0.0912) (0.121) (0.0961) (0.0796) (0.0998) (0.0906)
Regional Controls Yes Yes Yes Yes Yes Yes
Age controls Yes Yes Yes Yes Yes Yes
R-squared 0.317 0.304 0.257 0.314 0.299 0.256
Kleibergen-Paap 16.753 10.839 17.299 19.178 19.697 21.531
Hansen J statistic 1.894 0.98 1.503 2.036 1.964 3.713
p-value 0.388 0.6128 0.4716 0.3614 0.3747 0.1562
Observations 13269 4717 8550 13269 4717 8550
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
35
Table 6: First stage regression for Hours worked
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Homicides Formal Informal DTO Formal Informal Homicides Formal Informal DTO Formal Informal
Average homicides 1990-2001 0.0346*** 0.0565*** 0.0302*** 0.0340*** 0.0577*** 0.0294***
(0.0078) (0.0084) (0.0066) (0.0079) (0.0092) (0.0069)
Average dtos 1990-2001 3.0227*** 3.0864*** 2.9971*** 2.8084*** 2.8082*** 2.8637***
(0.6640) (0.7091) (0.6389) (0.6871) (0.7498) (0.6686)
Supply shock-South 4.78e-06*** 4.81e-06*** 4.84e-06*** 0.0000105*** 9.73e-06*** 0.0000109***
(1.20e-06) (1.23e-06) (1.31e-06) (2.00e-06) (2.28e-06) (1.95e-06)
Supply shock-Pacific -1.13e-06 -1.97e-06 -2.24e-06 0.0000158*** 0.0000167*** 0.0000145***
(1.98e-06) (2.46e-06) (2.14e-06) (2.84e-06) (3.60e-06) (2.65e-06)
Male -0.0365 -0.0611 -0.0258 -0.0113 -0.0111 0.0065 0.02710 -0.0455 -0.0206 -0.0055 0.0138 -0.0107***
(0.0271) (0.0410) (0.0353) (0.0415) (0.0582) (0.0484) (0.0255) (0.0404) (0.0315) (0.0345) (0.0552) (0.0371)
Constant 1.5260*** 1.3405*** 1.6360*** 0.4282 0.7148* 0.4849* 0.9623 0.5435 1.1097*** -1.4114*** -1.1303** -1.4010***
(0.1923) (0.2269) (0.1815) (0.2372) (0.4045) (0.2431) (0.2834) (0.3323) (0.2769) (0.2870) (0.5188) (0.2330)
Schooling controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Age controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Region controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
36
Occupation controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
F-test 19.41 44.47 20.43 20.72 18.94 22 9.8 19.32 9.94 23.46 16.7 25.18
Observations 13269 4717 8550 13269 4717 8550 13269 4717 8550 13269 4717 8550
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 7: Second stage regression for Hours worked
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS
Homicides Formal Informal DTOs Formal Informal Homicides Formal Informal DTOs Formal Informal
Male 0.192*** 0.132*** 0.226*** 0.194*** 0.134*** 0.226*** 0.192*** 0.133*** 0.226*** 0.194*** 0.135*** 0.226***
(0.0118) (0.0126) (0.0192) (0.0124) (0.0128) (0.0200) (0.0118) (0.0127) (0.0191) (0.0123) (0.0127) (0.0196)
Elementary School 0.0157 -0.0112 0.0189 0.0204 -0.000705 0.0200 0.0165 -0.00730 0.0186 0.0202 0.00245 0.0196
(0.0273) (0.0345) (0.0318) (0.0268) (0.0350) (0.0320) (0.0270) (0.0345) (0.0317) (0.0269) (0.0346) (0.0320)
Secondary School 0.0554* 0.000791 0.0495* 0.0557** 0.0110 0.0462 0.0560** 0.00491 0.0496* 0.0558** 0.0139 0.0483
(0.0287) (0.0402) (0.0298) (0.0270) (0.0389) (0.0295) (0.0284) (0.0403) (0.0299) (0.0271) (0.0379) (0.0295)
High School 0.0169 -0.0372 -0.0240 0.0139 -0.0286 -0.0288 0.0170 -0.0337 -0.0235 0.0143 -0.0262 -0.0262
(0.0330) (0.0447) (0.0369) (0.0320) (0.0445) (0.0370) (0.0328) (0.0448) (0.0370) (0.0320) (0.0441) (0.0368)
More than high school -0.0766*** -0.130*** -0.115*** -0.0719*** -0.118*** -0.110*** -0.0762*** -0.126*** -0.115*** -0.0723*** -0.114*** -0.113***
(0.0262) (0.0365) (0.0325) (0.0248) (0.0362) (0.0327) (0.0260) (0.0362) (0.0325) (0.0249) (0.0352) (0.0325)
Directors & Chiefs 0.171*** 0.0468 0.197*** 0.162*** 0.0426 0.188*** 0.173*** 0.0496 0.195*** 0.164*** 0.0373 0.196***
(0.0349) (0.0515) (0.0425) (0.0363) (0.0551) (0.0426) (0.0347) (0.0516) (0.0416) (0.0368) (0.0557) (0.0426)
Manufacture & Industry 0.114*** 0.0114 0.110*** 0.110*** 0.00843 0.106*** 0.116*** 0.0139 0.108*** 0.111*** 0.00426 0.110***
(0.0183) (0.0281) (0.0216) (0.0183) (0.0318) (0.0214) (0.0182) (0.0274) (0.0213) (0.0185) (0.0327) (0.0212)
37
Commerce & Sales 0.109*** 0.0409 0.0903*** 0.0797*** 0.0199 0.0663* 0.110*** 0.0400 0.0896*** 0.0827*** 0.00717 0.0823**
(0.0247) (0.0303) (0.0343) (0.0256) (0.0376) (0.0351) (0.0248) (0.0298) (0.0340) (0.0257) (0.0362) (0.0347)
Services 0.0423** 0.0141 0.0134 0.0311 0.0115 -0.00136 0.0437** 0.0166 0.0124 0.0326 0.00754 0.00893
(0.0210) (0.0256) (0.0274) (0.0214) (0.0285) (0.0282) (0.0207) (0.0247) (0.0272) (0.0209) (0.0291) (0.0270)
Professional services 0.0113 -0.114*** 0.0196 -0.0191 -0.130*** -0.00893 0.0113 -0.114*** 0.0196 -0.0161 -0.140*** 0.00924
(0.0212) (0.0278) (0.0273) (0.0242) (0.0352) (0.0299) (0.0212) (0.0276) (0.0273) (0.0236) (0.0342) (0.0283)
Log of homicide rates -0.0252* -0.0185 -0.0101 -0.0170 -0.00557 -0.0168
(0.0152) (0.0142) (0.0156) (0.0138) (0.0140) (0.0157)
Number of DTOs 0.0321*** 0.0174 0.0319*** 0.0290*** 0.0285*** 0.0116
(0.00842) (0.0121) (0.0122) (0.00774) (0.0100) (0.00972)
Constant 3.231*** 3.810*** 3.129*** 3.154*** 3.744*** 3.088*** 3.211*** 3.776*** 3.145*** 3.155*** 3.733*** 3.098***
(0.0568) (0.0670) (0.0673) (0.0510) (0.0588) (0.0604) (0.0523) (0.0671) (0.0613) (0.0509) (0.0587) (0.0593)
Age controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Regional controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
R-squared 0.055 0.097 0.059 0.056 0.105 0.056 0.056 0.101 0.059 0.057 0.102 0.059
Kleibergen-Paap 13.109 8.041 13.104 15.047 12.802 16.055 16.556 11.114 17.452 23.713 19.608 26.653
Hansen J statistic N/A N/A N/A N/A N/A N/A 5.784 11.213 0.928 4.111 3.524 6.918
p-value N/A N/A N/A N/A N/A N/A 0.0555 0.0037 0.6289 0.1282 0.1717 0.0315
Observations 13269 4717 8550 13269 4717 8550 13269 4717 8550 13269 4717 8550
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Appendix
38
Table 8: First stage regression for self-selection
(1) (2)
Homicides DTOs
Age -0.0629*** -0.0629***
(0.00664) (0.00664)
Age squared 0.000706*** 0.000706***
(0.0000781) (0.0000781)
Male 0.0670 0.0670
(0.0422) (0.0422)
Elementary School -0.106 -0.106
(0.0943) (0.0943)
Secondary School -0.415*** -0.415***
(0.104) (0.104)
High School -0.723*** -0.723***
(0.114) (0.114)
More than high school -0.812*** -0.812***
(0.123) (0.123)
Directors & Chiefs -1.044*** -1.044***
(0.130) (0.130)
Manufacture & Industry -0.853*** -0.853***
(0.0993) (0.0993)
Commerce & Sales -0.799*** -0.799***
(0.0914) (0.0914)
Services -0.719*** -0.719***
(0.0923) (0.0923)
Professional services -0.928*** -0.928***
(0.0934) (0.0934)
Married -0.141*** -0.141***
(0.0415) (0.0415)
Head of household -0.0524 -0.0524
(0.0391) (0.0391)
Children under 14 years in the hh 0.0186 0.0186
(0.0163) (0.0163)
Household size 0.00443 0.00443
(0.0102) (0.0102)
Elderly in the hh -0.0453*** -0.0453***
(0.0171) (0.0171)
Constant 2.912*** 2.912***
(0.219) (0.219)
Observations 13267 13267
Pseudo R-squared 0.123 0.123
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
39
Table 9: Effect on wages controlling for self-selection
(1) (2)
Homicides DTOs
Average homicides 1990-2001 -0.0384
(0.0243)
Average dtos 1990-2001 0.0912***
(0.0218)
Age 0.0257*** 0.0231***
(0.00499) (0.00515)
Age squared -0.000289*** -0.000265***
(0.0000540) (0.0000552)
Male 0.327*** 0.333***
(0.0187) (0.0192)
Elementary School 0.114*** 0.123***
(0.0318) (0.0310)
Secondary School 0.188*** 0.180***
(0.0472) (0.0475)
High School 0.153** 0.133*
(0.0689) (0.0713)
More than high school 0.460*** 0.458***
(0.0740) (0.0732)
Directors & Chiefs 0.523*** 0.475***
(0.0757) (0.0789)
Manufacture & Industry 0.112** 0.0826
(0.0527) (0.0550)
Commerce & Sales 0.0744 -0.0202
(0.0537) (0.0609)
Services 0.116** 0.0709
(0.0465) (0.0502)
Professional services 0.332*** 0.233***
(0.0597) (0.0693)
lambda 0.842*** 0.872***
(0.124) (0.127)
Constant 6.830*** 6.711***
(0.118) (0.112)
Observations 13267 13267
Pseudo R-squared 0.322 0.325
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
40
Table 10: OLS regression for the effect of violence and DTOs presence on
wages
(1) (2) (3) (4) (5) (6)
OLS OLS OLS OLS OLS OLS
Homicides Formal Informal Homicides Formal Informal
Male 0.311*** 0.210*** 0.375*** 0.313*** 0.210*** 0.376***
(0.0184) (0.0230) (0.0223) (0.0187) (0.0231) (0.0225)
Elementary school 0.163*** 0.101 0.146*** 0.160*** 0.0981 0.142***
(0.0321) (0.0655) (0.0382) (0.0325) (0.0655) (0.0384)
Secondary school 0.383*** 0.320*** 0.318*** 0.375*** 0.316*** 0.309***
(0.0399) (0.0663) (0.0453) (0.0397) (0.0659) (0.0453)
High school 0.483*** 0.433*** 0.342*** 0.474*** 0.429*** 0.335***
(0.0443) (0.0722) (0.0503) (0.0448) (0.0718) (0.0504)
More than high school 0.852*** 0.732*** 0.757*** 0.851*** 0.730*** 0.760***
(0.0507) (0.0715) (0.0592) (0.0501) (0.0716) (0.0582)
Log of homicide rates 0.0317** 0.0193 0.0353**
(0.0134) (0.0141) (0.0166)
Number of DTOs 0.0375*** 0.0101 0.0426***
(0.00871) (0.00722) (0.0127)
Constant 6.111*** 6.899*** 6.133*** 6.174*** 6.940*** 6.199***
(0.0833) (0.117) (0.0936) (0.0800) (0.0966) (0.0905)
Age Controls Yes Yes Yes Yes Yes Yes
Regional Controls Yes Yes Yes Yes Yes Yes
R-squared 0.297 0.299 0.232 0.301 0.299 0.236
Observations 13269 4717 8550 13269 4717 8550
Standard errors in parentheses
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
41
