^{1}

The scaling up of violent crime in México is often characterized as detrimental to the Mexican tourism industry. However, no econometric study so far challenges this claim with data. This paper therefore empirically analyzes the impact of crime on the arrivals of tourists in México for the period 1990 to 2010. Using a panel data set for the 31 Mexican federal states and México City, I find a negative and significant effect of homicides on the number of tourists arriving. This finding is robust to alternative estimation techniques and samples. Furthermore, when disaggregating the tourist arrival data into local and international, I find that international tourists seem to be more intimidated from homicides than locals.

El aumento del crimen violento en México ha sido considerado como negativo para la industria turística mexicana. Sin embargo, no se cuenta con un estudio econométrico que confronte este argumento con datos estadísticos. El presente artículo analiza empíricamente el impacto que el crimen violento tiene en las llegadas de turistas a México para el periodo 1990 a 2010. Utilizando datos panel para los 31 estados mexicanos y la Ciudad de México, encuentro un efecto negativo y significativo de los homicidios sobre las llegadas de turistas. Este hallazgo es robusto a diferentes técnicas de estimación y muestras. Además, cuando se desagrega la información de llegadas de turistas en locales e internacionales, encuentro que los turistas internacionales se intimidan más por la presencia de crimen violento que los turistas locales.

Does violent crime deter tourists from visiting México? According to the United Nations World Tourism Organization (UNWTO 2016), México was ranked in 2016 as the 8^{th} place to visit in the preferences of international tourists. Conversely, the country was ranked 142 out of 163 countries by the Global Peace Index (2017), with 163 being the most violent country. In the year 2006 the Mexican government decided to give a frontal fight to the different drug trafficking organizations (henceforth DTOs) operating all across the Mexican territory. As a result of this strategy violent crime in the form of homicides started to dramatically increase (Ríos 2012). Thus, it was not uncommon to read since the end of 2006 the headlines of international and national newspapers reporting the increasing wave of violence in México. This has had a negative impact on the Mexican society. For instance,

Moreover, after the intensification of violence from early 2007 onwards, analysts in the U.S. and México argued that there was a strong similarity between terrorism and attacks by the DTOs in México.^{2}^{3}^{4}^{5}^{6}

Due to the availability of tourism flow data, the period of study is restricted to 1990-2010. However this period takes into account the scaling up of crime during the years 2007-2010 when the Mexican government started to directly fight organized crime. The findings show that international tourist flows are more affected than local tourist flows after controlling for violent crime, income, price level, urbanization, weather, and infrastructure. As a starting point I propose a dynamic panel data model with fixed effects. According to

The rest of the paper is organized as follows: “Leterature review” provides a review of the literature on tourism demand and crime. “Data and method” explains the data selection based on the literature on tourism demand and presents the empirical methodology. “Empirical results” discusses the results, while the last section concludes. The conclusion is followed by an appendix including graphs and robustness checks.

The literature on crime and tourism is small. Most work on the impact of crime on tourism concentrates on qualitative evidence as for instance,

Several developing countries have seen tourism as a strategy for economic development. As the United Nations World Tourism Organization documents, tourism provides about 9.6% of the world´s total employment. This includes jobs indirectly supported by the industry. Furthermore, it accounts for 28.2% of the world´s exports of services (UNWTO 2017). More specifically, for México the tourism industry contributed 7.4% of the country´s GDP in the year 2016 and is after oil exports and remittances the third source of foreign currency for the country (WTTC, 2017). As of now there is no empirical evidence arguing that organized crime is targeting the tourism industry in México as a way to exert political pressure on the Mexican government. According to

The data used in the paper is a panel dataset across 31 Mexican states and México City during the 1990-2010 period. The following specification estimates the tourists arrivals (_{
it
} ) (logged), in state _{
it-1
} , homicides In_{
it
} and a vector of control variables _{
it
} :

where _{
i
} denotes state fixed effects to control for unobserved state specific heterogeneity in the panel dataset, ^{7}^{8}

Having described the two main variables of interest I turn now to the vector of control variables (Z_{
it
} ) which includes other potential determinants of tourist arrivals reported in state i during year t. I select these control variables from the existing literature on the subject.

The literature on tourism demand has focused on the study of international tourism while neglecting the study of national/local tourism. This literature can broadly be divided in two groups: The first group corresponds to contributions whose aim is to forecast tourism statistics as number of nights of stay, expenditures by tourists and /or the number of tourists arriving. For instance, the work by

I use the natural logarithm of the gross domestic product per capita in state i during year t as a proxy. I expect a positive and significant effect. A better economic environment enhances appropriate conditions for the stay of tourists. Furthermore researchers have used a wide variety of variables to represent prices in their models. In the context of international tourism demand, the variables used to represent prices have been foreign currency prices of tourist goods and services in destinations, the cost of transportation between origin and destination country and the effect of exchange rate variations on purchasing power. Put differently, as consumers, tourists also decide where to go based on the price of the goods they want to purchase; for instance holiday packages, which in some cases include flights and hotel reservations. In order to account for the differences in prices I use the price levels^{9}^{10}^{11}

It can potentially be the case that the number of tourists visiting a country originates more crime. Tourists are new to the destination they visit; this lack of information puts them in a riskier situation more easily than local people. Thus, criminals may see in them an easier prey. This applies to both national and international tourists. Moreover, while I am not aware of any variable which at the same time exerts any form of variation in the number of tourist arrivals and the number of homicides and is omitted from my specification, in general, the endogeneity problem in an econometric model can not only be due to the reverse causality as outlined above but also due to third omitted variables which affect both of the variables involved.

In order to account for potential endogeneity I employ a Two Stage Least Squares (2SLS) model. The validity of an estimation based on this method relies on the choice of a proper instrument. The instrumental variable must fulfil two criteria. The first one refers to the relevance of the instrument, i.e., it must induce sufficient exogenous variation in the explanatory variable in question, in particular, ^{12}

I propose the use of two instruments in an attempt to control for endogeneity in the model: The first instrument is the adult illiteracy rate within the population older than 15 years across the 31 Mexican states and México City. The data come from the Ministry of Education of México. This variable is intended to be a proxy for social exclusion. The rationale here is that social exclusion directly affects the increase in violent outcomes. For instance, the work by Caldeira (2000); Heinemann and Verner (2006), Borjas (1995); Katzman (1999), Buvinic, Morrison and Orlando (2002) and Beato (2002) show that socially excluded communities have higher illiteracy rates, higher numbers of homicides, higher percentages of employment in the informal sector and higher child mortality. Following on this, illiteracy impedes the opportunities for participation in the labour market and thus reduces the income of individuals and their chances to be included in society. For instance, using data from two groups of British adults born in 1958 and 1970, Parsons (2002) found a significant association between repeated offending and poor literacy or numeracy scores, particularly among young men.

In addition to these arguments, the work by

The second instrument is a proxy for the severity of punishment of committing a homicide. According to Becker´s model of crime and punishment (^{13}

^{14}^{15}

(1)
(2)
(3)
Variables
Tourists Arrivals: Total
Tourists Arrivals: Foreign
Tourists Arrivals: National
LDV (log) t-1
0.580*** (0.0629)
0.474*** (0.0860)
0.635*** (0.0521)
Homicide (log)
-0.123** (0.0570)
-0.307*** (0.105)
-0.0944* (0.0537)
Price level
-0.0265 (0.0225)
-0.0424 (0.0410)
-0.0336* (0.0194)
State per capita GDP (log)
0.0187 (0.318)
0.490 (0.550)
-0.0161 (0.300)
Urbanization
0.0282** (0.0126)
0.0603** (0.0268)
0.0296** (0.0113)
Storms
-0.0293 (0.0415)
-0.0357 (0.0809)
-0.0338 (0.0391)
Roads (log)
-0.00795 (0.0826)
-0.133 (0.220)
0.0183 (0.0777)
Constant
5.038** (1.825)
3.631 (3.389)
3.960** (1.776)
Hausmant test p > chi2
0.00
0.00
0.00
Year and State dummies
YES
YES
YES
Number of States
31
31
31
Number of Observations
497
492
494
R-squared
0.535
0.326
0.603
Method
Fixed Effects
Fixed Effects
Fixed Effects

Robust standard errors clustered by state in parentheses *** p<0.01, ** p<0.05, * p<0.1

Source: Own elaboration

Relying on these results, it is not possible yet to give a definitive answer to the research question of the paper. According to

The bias-correction procedure involves consistent estimates as a first step. These consistent estimates are based on one out of the three following estimators, namely the Anderson-Hsiao, Arellano-Bond and Blundell-Bond estimators. I choose the ^{16}

(1)
(2)
(3)
Variables
Tourists Arrivals: Total
Tourists Arrivals: Foreign
Tourists Arrivals: National
LDV (log) t-1
0.697*** (0.0483)
0.567*** (0.0515)
0.730*** (0.0461)
Homicide (log)
-0.118* (0.0678)
-0.295** (0.134)
-0.0896 (0.0673)
Price level
-0.0317 (0.0276)
-0.0485 (0.0512)
-0.0383 (0.0329)
State per capita GDP (log)
0.101 (0.405)
0.481 (0.645)
0.0977 (0.333)
Urbanization
0.0332* (0.0201)
0.0662* 0.0338
0.0355* (0.0193)
Storms
-0.0124 (0.151)
-0.0278 (0.0957)
0.0302 (0.0506)
Roads (log)
0.0124 (0.151)
-0.108 (0.296)
-0.0457 (0.130)
Year and State dummies
YES
YES
YES
Number of States
31
31
31
Number of Observations
497
492
494
Method
LSDVC
LSDVC
LSDVC

Bootstrapped standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Source: Own elaboration

The estimates of homicides coefficients in ^{17}

With respect to the control variables,

Arguably, the higher the concentration of people in cities, the higher are the victimization rates of crimes as pointed out by ^{18}

So far the LSDVC estimator has taken into account the bias inherent in the model due to the inclusion of the lagged dependent variable. However there is still a further issue to be dealt with, namely the potential reverse causality of the variables tourism and homicides. Since the dynamic bias-corrected estimator does not account for this problem,^{19}

(1)
(1a)
(2)
(2a)
(3)
(3a)
Variable
Total Arrivals
First stage regression
Foreign Arrivals
First stage regression
National Arrivals
First stage regression
Homicide (log)
Homicide (log)
Homicide (log)
Dependent Variable- (log) t-1
0.569*** (0.0549)
-0.071 (0.0488)
0.478*** (0.0808)
-0.063** (0.0299)
0.623*** (0.0475)
-0.071 (0.0499)
Homicide (log)
-0.218** (0.0901)
-0.241 (0.2021)
0.207** (0.0886)
Illiteracy
0.0149*** (0.0305)
0.147*** (0.0312)
0.144*** (0.0302)
Imprisonment rate
-0.004*** (0.0011)
-0.004*** (0.0011)
-0.004*** (0.0011)
Price level
-0.031 (0.0216)
-0.017 (0.0237)
-0.046 (0.0382)
-0.015 (0.0244)
-0.0381*** (0.0193)
-0.018 (0.0248)
State per Capita GDP (log)
0.132 (0.3098)
-0.049 (0.5019)
0.612 (0.5023)
0.021 (0.4885)
0.0832 (0.288)
-0.025 (0.4964)
Urbanization
0.029** (0.0122)
0.039* (0.0205)
0.057** (0.0269)
0.042 (0.0200)
0.0310*** (0.0108)
0.039* (0.0206)
Storms
-0.029 (0.0394)
-0.011 (0.0320)
-0.038 (0.0759)
-0.008 (0.0329)
-0.033 (0.0369)
-0.011 (0.0326)
Roads (log)
-0.016 (0.0741)
0.062 (0.0925)
-0.100 (0.1911)
0.060 (0.0938)
-0.00029 (0.0722)
0.056 (0.0948)
F-statistic
22.51
21.51
21.82
Hansen J (p-value)
0.8827
0.1904
0.8187
Kleibergen Paap LM test
11.71
11.14
11.27
Endogeneity test (p-value)
0.3462
0.9893
0.2376
Year and State Dummies
YES
YES
YES
YES
YES
YES
Number of States
31
31
31
31
31
31
Number of Observations
494
494
489
489
491
491
R-squared
0.5309
0.5802
0.3273
0.5837
0.598
0.5793
Method
FE-2SLS
FE-2SLS
FE-2SLS
FE-2SLS
FE-2SLS
FE-2SLS

Robust standard errors clustered at the state level in parentheses ***p<0.01, **p<0.05, *p<0.1.

Source: Own elaboration

Column 1 of

Interestingly, as column 2 shows, there is no significant effect of homicides on the arrival of international tourists. However, the endogeneity test for all three models at the bottom of table I3 shows that the null hypothesis of exogeneity of the homicides variable cannot be rejected (with p-values of 0.35, 0.99 and 0.24). According to this test there is no reverse causality going from tourism to homicides. In this sense, the results of the LSDVC Bruno estimator provide the preferred estimation since this method is superior to the (2SLS) fixed effects estimation which does not control for the

This paper has investigated whether there is an effect of violent crime on tourism in México for the 1990-2010 period. The contribution of the paper is twofold. First, addressing endogeneity the paper finds that the impact of violent crime on tourism in México is negative and significant. Second, this paper investigates whether international tourists or local tourists are more affected by violent crime. Due to the lack of data, previous research has concentrated only on the analysis of international tourism flows. First, my findings show that tourist arrivals in Mexican states are reduced by increased violent crime. Second, international tourists appear to be more intimidated by violent crime than local tourists. As argued in previous research by

In terms of tourism policy, the findings suggest that better information and promotion of tourism in México abroad could positively affect the image of the country itself. Indeed, the Mexican Federal government promotes tourism in México abroad. Further studies might look at whether these tourism promotion investments have been effective by using impact evaluation techniques as for instance Difference in Differences estimation. Moreover, it would be important for tourism policy to know whether tourists move to different locations in order to avoid dangerous regions.

The time span of the paper is one of its main limitations. Due to data availability, it is only possible to build up the panel data set for all Mexican states and México City up to the year 2010. For several control variables data are not yet available. Examples of these variables are recent data on GDP and data derived from the not yet available 2020 census. Furthermore, the panel data results presented here show only average effects at the state level. The paper is not able to look deeper into more disaggregated data at the municipal level since these data are not available. Data disaggregation is relevant since not all regions of the country experience crime in the same way and not all regions in the country are tourism destinations. Further research might look at the relationship tourism and crime using municipal or county level data.

Finally yet importantly, the study of how crime impacts tourism in México can better be determined if more disaggregated data on tourism and crime at the municipal are available.

I thank Axel Dreher, Hannes Öhler, Nils-Hendrik Klann, Andreas Fuchs, Alexandra Rudolph, Edgar J. Sánchez Carrera, Krishna C. Vadlamannati, Willy Cortez Yactayo, Abraham Zepeda, Salvador O. Rodríguez for their helpful comments. I also thank the participants at the research seminars at Universidad de Guadalajara and at the Universidad de las Americas Puebla and two anonymous referees for their helpful comments. The kind support from INEGI in Aguascalientes, México for all questions about the data at the state level is also acknowledged. All remaining errors are mine.

See: The Economist, November 15^{th} 2010,

See: The Economist, May 27^{th} 2012,

See: El Sol de Hidalgo, September 17^{th} 2008,

Travel Warning as of February 8^{th} 2012 U.S. Department of State. Bureau of Consular Affairs. Travel Warning as of April 4^{th} 2012 Foreign Affairs and International Trade Canada.

See: Australian Department of Foreign Affairs and Trade, May 12^{th} 2012,

See:

For details on mortality statistics see:

These data are measured as regional consumer price indexes with base year 2010. Indeed the real exchange rate U.S. dollar- Mexican Peso is another control variable for prices. Please note that its use imposes limitation in the variation across groups in the panel data set. To allow for variation I multiplied this real exchange rate with the price level across states. The results of the control variables of interest are not changed when using this product. Since this variable is not the control variable of interest, I do not delve further in to it. These results are available upon request.

See:

See:

See

See: El Universal,

According to the Hausman test, the fixed effects model is preferred over the random effects model. The test result is available upon request.

The difference of the coefficients is statistically significant at the 5% level. I tested for the significance of the difference in a nested model, interacting an international tourist dummy with all explanatory variables and the state and year dummies.

As in

In the LSDVC Bruno estimation, the difference of the coefficients is statistically significant at the 10% level.

In the nested model of both the fixed effects model and the LSDVC Bruno estimation, the difference of the coefficients is statistically significant at the 10% level.

For details see

These graphs are available upon request and are not shown in order to save space.

The difference in the nested model is statistically significant at the 5% level.

In order to identify outliers in the previous estimations I implement graphs showing the linear relationship between homicides and (total, international, national) tourist arrivals, controlling for all other explanatory variables. Using these graphs and coding in the Stata do-file, I identified those observations which lie far away from the regression line and removed them from the dataset^{20}

^{21}

Looking now at

(1)
(2)
(3)
Variables
Tourists Arrivals: Total
Tourists Arrivals: Foreign
Tourists Arrivals: National
Dependent Variable- (log) t-1
0.589*** (0.0518)
0.554*** (0.0534)
0.630*** (0.0531)
Homicide (log)
-0.112* (0.0563)
-0.229*** (0.0959)
-0.106** (0.0487)
Price level
-0.016 (0.0202)
-0.0515 (0.0441)
-0.0305* (0.0168)
State per capita GDP (log)
-0.349 (0.288)
0.390 (0.522)
-0.322 (0.278)
Urbanization
0.0377*** (0.0122)
0.0604*** (0.0231)
0.0359*** (0.0113)
Storms
-0.0329 (0.0343)
-0.0353 (0.0649)
-0.025 (0.0369)
Roads (log)
0.038 (0.0660)
0.001 (0.204)
0.0517 (0.0592)
Constant
4.321** (1.678)
-0.0154 (2.928)
3.866** (1.753)
Hausmant test p > chi2
0.00
0.00
0.00
Year and State dummies
YES
YES
YES
Number of States
31
31
31
Number of Observations
492
485
490
R-squared
0.914
0.914
0.915
Method
Fixed Effects
Fixed Effects
Fixed Effects

Robust standard errors clustered by state in parentheses *** p<0.01, ** p<0.05, * p<0.1

Source: Own elaboration

(1)
(2)
(3)
Variables
Tourists Arrivals: Total
Tourists Arrivals: Foreign
Tourists Arrivals: National
LDV (log) t-1
0.669*** (0.0406)
0.463*** (0.0611)
0.714*** (0.0373)
Homicide (log)
-0.107* (0.0583)
-0.219*** (0.0821)
-0.102* (0.0575)
Price level
-0.020 (0.0293)
-0.057 (0.0598)
-0.033 (0.0254)
State per capita GDP (log)
-0.285 (0.336)
0.409 (0.667)
-0.229 (0.345)
Urbanization
0.042*** (0.0165)
0.067** (0.0262)
0.041*** (0.0128)
Storms
-0.029 (0.0405)
-0.026 (0.0659)
-0.0239 (0.0439)
Roads (log)
0.057 (0.130)
0.053 (0.287)
0.071 (0.129)
Time dummies
YES
YES
YES
Number of States
31
31
31
Number of Observations
492
485
490
Method
LSDVC
LSDVC
LSDVC

Bootstrapped standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Source: Own elaboration

(1)
(1a)
(2)
(2a)
(3)
(3a)
Variable
Total Arrivals
First stage regression
Foreign Arrivals
First stage regression
National Arrivals
First stage regression
Homicide (log)
Homicide (log)
Homicide (log)
Dependent Variable- (log) t-1
0.572*** (0.0443)
-0.071 (0.0488)
0.548*** (0.0478)
-0.07** (0.0307)
0.616*** (0.0492)
-0.068 (0.0500)
Homicide (log)
-0.251*** (0.0802)
-0.305* (0.1821)
-0.242*** (0.0787)
Illiteracy
0.15*** (0.0303)
0.15*** (0.0308)
0.145*** (0.0301)
Imprisonment rate
-0.005*** (0.0011)
-0.005*** (0.0011)
-0.004*** (0.0011)
Price level
-0.022 (0.0198)
-0.020 (0.0237)
-0.06 (0.0397)
-0.017 (0.0244)
-0.035** (0.0174)
-0.020 (0.0248)
State per Capita GDP (log)
-0.239 (0.2622)
-0.068 (0.4996)
0.583 (0.4862)
0.026 (0.4845)
-0.238 (0.2490)
-0.056 (0.4937)
Urbanization
0.040*** (0.0112)
0.04* (0.0204)
0.060*** (0.0219)
0.041** (0.0199)
0.038*** (0.0101)
0.040* (0.0205)
Storms
-0.032 (0.0320)
-0.009 (0.0317)
-0.035 (0.0606)
-0.007 (0.0326)
-0.024 (0.0340)
-0.009 (0.0326)
Roads (log)
0.020 (0.0600)
0.060 (0.0926)
0.008 (0.1758)
0.0589 (0.0927)
0.027 (0.0551)
0.057 (0.0950)
F-statistic
22.43
21.61
22.09
Hansen J (p-value)
0.8486
0.4525
0.8801
Kleibergen Paap LM test
11.7
11.11
11.21
Endogeneity test (p-value)
0.1075
0.6083
0.1098
Year and State Dummies
YES
YES
YES
YES
YES
YES
Number of States
31
31
31
31
31
31
Number of Observations
489
489
482
482
487
487
R-squared
0.6312
0.5825
0.4529
0.5879
0.6736
0.5797
Method
FE-2SLS
FE-2SLS
FE-2SLS
FE-2SLS
FE-2SLS
FE-2SLS

Robust standard errors clustered at the state level in parentheses ***p<0.01, **p<0.05, *p<0.1.

Source: Own elaboration

(1)
(2)
(3)
Variables
Arrivals: Total
Arrivals: Foreign
Arrivals: National
LDV (log) t-1
0.665***
0.570***
0.723***
(0.0510)
(0.0455)
(0.0520)
Homicide (log) (fitted values) *
-0.179*
-0.288*
-0.173*
(0.0995)
(0.172)
(0.0952)
Price level
-0.0336
-0.0510
-0.0406
(0.0306)
(0.0543)
(0.0285)
State per Capita GDP (log)
0.188
0.626
0.179
(0.439)
(0.732)
(0.369)
Urbanization
0.0329
0.0635**
0.0359**
(0.0212)
(0.0282)
(0.0176)
Storms
-0.0275
-0.0309
-0.0308
(0.0601)
(0.0929)
(0.0519)
Roads (log)
0.0142
-0.0743
0.0346
(0.159)
(0.265)
(0.146)
Time dummies
YES
YES
YES
Number of States
31
31
31
Number of Observations
494
489
491
Method
LSDVC
LSDVC
LSDVC

Bootstrapped standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

* The fitted values of the potentially endogenous variable Homicide (log) are taken from the first stage regression from the 2SLS estimation

Source: Own elaboration.

Aguascalientes
Ciudad de México
Morelos
Sinaloa
Baja California
Durango
Nayarit
Sonora
Baja California Sur
Estado de México
Nuevo León
Tabasco
Campeche
Guanajuato
Oaxaca
Tamaulipas
Chiapas
Guerrero
Puebla
Tlaxcala
Chihuahua
Hidalgo
Querétaro
Veracruz
Coahuila
Jalisco
Quintana Roo
Yucatán
Colima
Michoacán
San Luis Potosí
Zacatecas

Source: Own elaboration.

Variables
Mean
Standard Deviation
Minimum
Maximum
Observations
Tourist Arrivals (Total) log
14.15577
1.098107
7.233455
16.31798
589
Tourist Arrivals (Total) log t-1
14.13433
1.107204
7.233455
16.31798
562
Tourist Arrivals (International) log
12.00465
1.668351
5.826
15.62604
586
Tourist Arrivals (International) log t-1
12.00123
1.675661
5.826
15.62604
559
Tourist Arrivals (National) log
13.92622
1.072491
6.952729
16.0395
588
Tourist Arrivals (National) log t-1
13.90193
1.080445
6.952729
16.02846
561
Homicides (log)
5.382426
1.150092
2.484907
8.737774
672
Price Level
57.02784
27.58268
10.48747
98.55759
637
State per Capita GDP (log)
4.146175
.513901
3.386864
6.176142
672
Urbanization
72.61502
14.94279
39.45287
99.76386
672
Storms
.1622024
.368911
0.00
1.00
672
Roads (log)
8.883446
.6622945
7.247081
10.16591
651
Illiteracy Rate
9.54375
5.673183
2.1
29.20
640
Imprisonment Rate
46.57406
39.92938
0.00
247.8261
665

Source: Own elaboration.

Variables
Definitions and data sources
Total Tourist Arrivals
The logarithm of
International Tourist Arrivals
The logarithm of
National Tourist Arrivals
The logarithm of
Homicides
The logarithm of total number of homicides committed in state in year
Urbanization
Share of the total population living in urban areas in state
Price Level
Price level of the main cities in each state and México City. The data were obtained from the Mexican Central Bank. The period is 1990 till 2010.
State per Capita GDP (log)
Own calculation using data on each State GDP and Population in each State. Values are in Mexican pesos, constant prices 2003. The data on State GDP are form the National Accounting System and the Population data are from the population censuses 1990, 2000, 2010 and population counting 1995, 2005. All data are provided by INEGI.
Storms
A dummy variable which takes the value of one if a hurricane hit in state i in year t and zero otherwise. The data are from the Meteorological National Service and the Institute of Engineering at the National Autonomous University (UNAM) in México City.
Roads
The logarithm of the number of kilometres of highways and paved roads in state i in year t. The data are from the Ministry of Transport and Communication (SCT México).
Illiteracy Rate
Illiteracy rate of population older than 15 years. Data are provided by the Mexican Education Ministry.
Imprisonment Rate
Rate of imprisonment of people who have committed homicide. Data provided by INEGI.

Source: Own elaboration.