The significance of age structure,education,and youth unemployment for explaining subnational variation in violent youth crime in Mexico

Abstract

Violent crime in Mexico occurs at a rate that dwarfs the human costs of most contemporary civil wars, and the drug cartels responsible for the violence exercise de facto control over significant geographical territories. In this respect, the Mexican “drug wars” resemble conflicts over the control of rich natural resources in Sub-Saharan Africa and elsewhere, blurring the distinction between “political” and “social” or “criminal” violence. In the civil war literature, a young age structure has been argued to provide inexpensive rebel labor and thus increase opportunities for a rebel group to wage war against a government. Similarly, relatively large groups of “idle” young men could arguably be a factor that reduces recruitment costs for criminal enterprises through the abundant supply of youth with low opportunity cost. Acknowledging organized crime around drug trafficking as a major cause of crime and violence in Mexico, we ask whether the availability of large young male cohorts, or male “youth bulges”, low education and high youth unemployment ease the recruitment to these organizations and may contribute to explain variance in violent crime rates across Mexican states over time. Using panel data covering 32 states in Mexico during the 1997–2010 period, we find that, while a coarse measure of regional youth bulges is not associated with patterns of violent youth crime, high youth unemployment in low-education strata is, in particular in the context of large male youth bulges. These results remain robust against alternative data, sample size, estimation techniques and controls for potential endogeneity concerns.

Keywords

Education Mexico unemployment violent crime youth bulge

Introduction

The resemblance between the “drug wars” in Mexico and many contemporary civil wars over access to natural resources is striking. However, while some scholars have noted the similarities between factors explaining armed conflict and violent crime (e.g. Neumayer, 2003: 619), the two phenomena are usually studied separately. This article addresses the issue of violent youth crime in Mexico, employing a theoretical framework, the “opportunity perspective”, which has been a dominating narrative in the civil war literature. The framework emphasizes structural factors providing opportunities for potential rebel organizations to launch an insurgency against a state, such as large youth cohorts, or “youth bulges”, as well as other factors that determine economic opportunities for youth, like education and unemployment (Brett and Specht, 2004). In the political violence literature it has been noted that “youth bulges” have historically been associated with times of political crisis and upheaval (Goldstone, 1991, 2001) and it has generally been observed that young males are the main protagonists of criminal (Neapolitan, 1997: 92, Neumayer, 2003: 621) as well as political (Elbadawi and Sambanis, 2000: 253; Mesquida and Wiener, 1996; Urdal, 2006) violence. Generally, the increasing acknowledgement of the role of demographic factors in shaping conflict and international political developments is underscored by recent contributions in the field of political demography (e.g. Goldstone et al., 2012).

Studies of violent crime, particularly studies of homicide rates, have long employed cross-national time-series research designs. Most of these cross-national studies have included few developing countries, however. A much-cited homicide study, Fajnzylber et al. (2002), included only 39 countries, of which the minority were developing countries, citing problems with low data availability for developing countries as well as underreporting. Underreporting, the authors argue, should not be considered random noise, but measurement error that is systematically correlated with factors assumed to affect crime rates (Fajnzylber et al., 2002: 14).

Furthermore, while subnational panel studies have recently become more prominent in the civil war literature (e.g. Buhaug and Rød, 2006; Hoelscher et al., 2012; Østby et al., 2011; Urdal, 2008; Vadlamannati, 2011), as well as in social violence studies of global south contexts (e.g. Dreze and Khera, 2000 on India; Hoelscher, 2015 on Brazil; and Widner et al., 2011 on Mexico), such studies are still rare compared with cross-national analyses. By assessing variation in violent crime within Mexico over time, this study is less prone to measurement error stemming from differences in data collection and reporting procedures across countries, although we acknowledge several potential sources of bias. Furthermore, the subnational focus enables the use of data sources—in particular youth unemployment—that are not available for a large number of countries, and thus may not be used in cross-national studies.

Mexico provides an ideal case for testing propositions about the significance of youth opportunities for violent crime. Demographically, Mexico is a relatively young country with about a third of its current population falling into the age range of 12–29 years. The period of study, 1997–2010, covers a time of significant youth population growth in Mexico. According to the Mexican Institute of the Youth, the Mexican population aged 12–29 years increased by 40.6% between 1990 and 2000 (Instituto Mexicano de la Juventud y Secretaría de Educación Pública, México, 2008). While the overall growth in youth population is slowing down, regional differences in growth rates still exist owing to migration and geographic fertility differentials. Detailed demographic, social, and crime data further allows us to use econometric methods to consider how large youth cohorts in the context of limited education and employment opportunities affect violent crime.

This study adds to the existing literature in several ways. The article identifies and discusses youth opportunities and their potential implications for violent youth crime and tests these propositions empirically in one of the first subnational studies of violent crime in a developing country. It is also the first study to look at youth bulges and violence, either political or criminal, in the context of both education and employment, a unique opportunity arising from the rare availability of such data for Mexican states. Our results suggest that while youth crime and high homicide rates in Mexico are not associated with the ebb and flow of the male youth population, both high youth unemployment and low youth education are associated with higher levels of crime and homicide. And in this context, the relative size of the male youth population does matter. We also report further results of notable significance. In particular, there is an increasing concern that rapid urban population growth around the globe could lead to increasing levels of criminal as well as political violence. While this study, generally, finds some support for urban environments being more conducive to violent crime in Mexico, the pace of growth in the urban population does not appear to be associated with crime levels.

Theory

The literature on youth bulges and violence has particularly focused on the role of large youth cohorts in facilitating spontaneous and low-intensity political violence. Two different explanatory frameworks have primarily informed the discussion: one focusing on opportunities, and the other on motives for political conflict. The opportunity framework is particularly relevant for explaining criminal violence and has a parallel expression in the literature on violent crime (Neapolitan, 1997). Neumayer (2003) notes that “opportunity theory”“tries to understand variation in violent crime rates in terms of different opportunities or favourable conditions for committing crime”. Basing their approach primarily on Gary Becker”s (1968) opportunity framework, Fajnzylber et al. (2002: 1–2) contend that, “crime rates depend on the risks and penalties associated with apprehension and also on the difference between the potential gains from crime and the associated opportunity cost”.¹ Berman et al. (2011: 499) note that the relevance of the opportunity cost theory for crime is generally supported by subnational level evidence.

Youth bulges and violent crime

The opportunity literature, often referred to as the “greed” perspective (e.g. Collier, 2000), has its roots in economic theory and focuses on structural conditions that facilitate an organization’s engagement in violent activity: whether that be a rebel group or a criminal organization. These are conditions that either provide the organization with financial means, such as rents from drug trafficking, or reduce the costs of operation, including costs of recruitment. Relatively large youth cohorts can reduce recruitment costs for insurgent groups through the abundant supply of “rebel labor” with low opportunity cost, increasing the risk of armed conflict (Collier, 2000: 94). Similarly, large youth bulges may depress the cost of recruitment to criminal organizations. Opportunities for violence may be further boosted by a weak government with limited capabilities (Collier and Hoeffler, 2004; Fearon and Laitin, 2003).

A key assumption is that organizational structures that may be used for illegal purposes, whether political or criminal, exist exogenously, and that recruits join these organizations in order to obtain a private good. Hence, the collective action problem is presumed to be negligible. Organizations are able to recruit successfully only when the potential gain from joining is so high and the expected costs so low that potential recruits will favor joining over alternative income-earning opportunities. Collier (2000: 94) argues that the mere existence of an extraordinarily large pool of youth is a factor that lowers the cost of recruitment since the opportunity cost for a young person is generally low. Hence, our expectation is that:

Hypothesis 1: In regions with large youth populations relative to the adult population, violent crime rates are higher, everything else being equal.

However, Hirschi and Gottfredson (1983) argue that age, in and of itself, is an insufficient explanation for violence, and that shifting attention toward the meaning or interpretation of the relationship is required. Hence, in the following we consider two factors that are key determinants of youth opportunities: educational attainment and youth unemployment.

Educational opportunities and violent crime

Governments can expand educational opportunities in response to youth bulges and hence ease demographic transition problems. Higher levels of education among men arguably act to reduce the risk of political violence, resulting from the higher opportunity cost of rebellion for educated men (Collier, 2000). Since educated men generally have better income-earning opportunities than uneducated men, their alternative cost is higher, and they will be less likely to be recruited to criminal organizations.

Hence, higher levels of education are expected to be associated with a reduced risk of violence. While for “criminal entrepreneurs”, a high level of education may in fact lead to higher rewards if it enables more efficient management of illicit activities (Barakat and Urdal, 2009), the argument that the involvement of young people in criminal activity is economically less attractive the more highly educated a person is refers to mass participation. In areas with large potential pools of recruits, increasing education can act to reduce this pool. Although the argument that education increases the opportunity cost for young people takes a broad form, we focus here on secondary education for young males since they are the primary target for recruitment to criminal organizations, and secondary education is typically an entrance requirement to modern-sector employment.

Hypothesis 2: In regions with low secondary male education levels, violent crime rates are higher, everything else being equal.

Youth unemployment and violent crime

Central to the opportunity cost framework is the availability of youth employment opportunities. If the ability in the labor market to absorb a sudden surplus of young job seekers is limited, a large pool of unemployed and frustrated youth with low opportunity cost arises, providing potential recruits for either political or criminal violence (Braungart, 1984; Choucri, 1974; Cincotta et al., 2003; Dell et al., 2019; Goldstone, 1991; Moller, 1968).

The expectation that exceptionally large youth cohorts increase the supply of cheap recruits for criminal enterprises is further supported by studies in economic demography suggesting that the alternative cost of individuals belonging to larger youth cohorts is generally lower compared with members of smaller cohorts owing to higher unemployment and thus increased pressure on male wages (Easterlin, 1987; Machunovich, 2000: 236). Increases in relative cohort size arguably result in a reduction in male relative income. Such a direct relationship has been found in several studies using wage data for smaller samples of countries (Machunovich, 2000: 238; see also Korenman and Neumark, 1997). Berman et al. (2011: 500) note that, according to opportunity cost theory, recruits to violent crime are drawn not only from among the unemployed, but also from among individuals in low-wage employment.

So not only do youth bulges provide an unusually high supply of individuals with low opportunity cost, as anticipated by Collier (2000), but an individual belonging to a relatively large youth cohort generally also has a lower opportunity cost relative to a young person born into a smaller cohort. While labor markets differ substantially with regard to flexibility, but also within countries, empirical evidence suggests that on average, large youth cohorts are substantially more likely to experience both lower relative wages and higher unemployment rates (Korenman and Neumark, 1997). Hence, we expect that:

Hypothesis 3: In regions with large unemployment among young males, crime rates are higher, everything else being equal.

Finally, we consider the possible impact of violent crime on the factors leading to low opportunities for young people. Given the expectations that low education and high unemployment among male youth should be associated with increased levels of violent crime, we would further expect that high unemployment in low-education male strata should be particularly strongly associated with violence, and that the economic opportunities for this group of males may be particularly limited in the context of large male youth bulges. Rogers and Pridemore (2016: 259) note the lack of previous tests of this contextual relationship.

Hypothesis 4: The association between large youth cohorts and violent crime is particularly strong in regions where education levels are low and unemployment rates among young males are high, everything else being equal.

Existing research

Previous studies have found mixed evidence for a relationship between age structure, or “youth bulges”, and violent crime. Hansmann and Quigley (1982) and Pampel and Gartner (1995) both find a significant impact of age structure on homicide rates in cross-national studies, while Gartner and Parker (1990) find a strong age structure effect on homicide in two (US and Italy) out of five countries, acknowledging that differential patterns within countries may still have affected internal variation in homicide among the remaining three countries. On the other hand, Avison and Loring (1986), Fajnzylber et al. (2002), Neumayer (2003), Cole and Gramajo (2009), Pridemore (2011), and Rogers and Pridemore (2016) do not find statistically significant effects of age structure on crime among country-level panel data analyses.² In a meta-analysis of cross-national homicide studies, Nivette (2011) reports that static population indicators were among the group of variables that exerted the weakest effect on homicide. Fox and Hoelscher (2012) find some initial and strong support for the youth bulge hypothesis, although the relationship washes away once socioeconomic factors are controlled for. A possible reservation here is that introducing socioeconomic variables also reduces the sample considerably. However, both Fox and Hoelscher’s (2012) results, as well as Neumayer’s (2003) finding that economic growth reduces homicides, point to the salience of socioeconomic factors. Hence, what we should be looking for are conditional factors determining youth opportunities.

There appears to be somewhat stronger, albeit by no means unequivocal, evidence for a link between education and violent crime. Cole and Gramajo (2009) find that increasing male education reduces homicide, Fajnzylber et al. (2002) conclude that higher education levels overall are associated with less homicide, while Dreze and Khera (2000) found that higher literacy levels moderated criminal violence levels in India. However, some results appear more puzzling: Cole and Gramajo (2009) found that higher female education was associated with higher homicide levels, while Fajnzylber et al. (2002) unexpectedly found that higher education was associated with higher levels of robbery. Furthermore, Pridemore (2011) reports inconclusive results with regards to education, while Robbins and Pettinicchio (2012) only finds weak support for the assumed beneficial effects of social capital on homicide.

While previous studies have identified a theoretical link between youth unemployment and violence, the lack of reliable youth unemployment data for many developing countries has made cross-national assessment of this relationship difficult. Several studies have rather tested the opportunity cost framework using measures of overall economic performance, assuming that youth unemployment will generally be affected by poor economic performance. Low economic growth has been identified as a robust predictor of both homicide (Neumayer, 2003) and the onset of civil war (Collier et al., 2003; Sambanis, 2002: 229). Two cross-national studies including unemployment data for a limited number of developing countries and studying the impact of national-level unemployment rates on crime came to different conclusions. Pampel and Gartner (1995) found no effect of unemployment on homicide rates, while Neumayer (2005) reported that higher unemployment rates were found to increase levels of both robbery and homicide. In a rare meso-level analysis of subnational-level unemployment and violence data spanning Afghanistan, Iraq, and the Philippines, Berman et al. (2011) found no relationship between local-level unemployment rates and insurgent attacks that kill civilians. While empirically sophisticated and a significant improvement over national-level analyses, the study is limited by the lack of age-specific unemployment rates. Providing more supportive evidence for the opportunity cost framework, Blattman and Annan (2016) found that, among Liberian ex-fighters, illicit and mercenary activities declined as their engagement in ordinary, peaceful work increased.

In the civil war literature, there has been a certain discussion about the measurement of age structure (Barakat and Urdal, 2009; Urdal, 2006). Like two authoritative civil war studies by Fearon and Laitin (2003) and Collier and Hoeffler (2004), most of the studies above employ suboptimal age structure measures. The commonly used operationalization counts 15–24 (or 29) year-old cohorts relative to the total population, including all cohorts under the age of 15 years in the denominator. Such definition is highly problematic both theoretically and empirically. First, most theories about youth revolt and crime assume that violence arises because youth cohorts experience institutional “bottlenecks” in the education system or in the labor market owing to their larger size compared with previous cohorts. Second, when using the total population in the denominator, youth bulges in countries with continued high fertility will be underestimated because the large under-15 populations deflate the youth bulge measure. At the same time, countries with declining fertility and relatively smaller under-15 populations—which are in a position to experience economic growth driven by age structural change, a so-called demographic bonus, which may reduce both criminal and political violence—score relatively higher. The issue of measurement appears not to have been discussed in the homicide literature, with the lone exception of Fox and Hoelscher (2012).

Data and methods

In this section, we describe the data covering all 32 Mexican states, including the former Federal district now officially known as Mexico City, during the 1997–2010 period. We further describe the estimation specifications.

Estimation specification

The baseline specification estimates the number of federal crimes committed by Mexican males in the age cohort 18–24 years in state i during year t measured in per capita (log) ( $Y C_{it}$ ), as a function of a set of youth opportunity variables viz., $Y {E_{it}}_{- 1}$ , and control variables ${Z_{it}}_{- 1}$ :

Y C_{it} = γ Y {E_{it}}_{- 1} + β {Z_{it}}_{- 1} + ν_{i} + λ_{t} + {ω_{i}}_{t}

(1)

where $ν_{i}$ denotes state-fixed effects to control for unobserved state-specific heterogeneity in the panel dataset, $λ_{t}$ is year-specific dummies and ${ω_{i}}_{t}$ is the error term. Note that the Hausman (1978) test overwhelmingly favours a fixed effect over a random effect estimator. For the dependent variable we use the number of federal crimes committed by Mexican males in the age cohort 18–24 years³ in state i in Mexico in year t measured in per capita (log). This data is reported by the National Institute for Statistics and Geography (Instituto Nacional de Estadística y Geografía, INEGI hereafter) for the 32 states (including Mexico City) for the 1997–2010 period (INEGI, 2012). Federal crimes include all counts of drug-related crime and other violent organized criminal activity, but exclude “common crime”, providing for an appropriate proxy for violent crime to be tested specifically against youth opportunities (see Online Appendix 4 for details and limitations of the dataset). Figure 1 shows the annual mean value of youth federal crime incidents reported across Mexican states during the 1997–2010 period. As seen, the states with the highest mean values of youth federal crimes are Baja California, Sonora, Jalisco, Mexico City, Chihuahua, and Sinaloa, many of which are heavily affected by drug-related violence. These states also have the highest relative shares of youth crimes in the country as illustrated in the right-hand side of the panel in Figure 1. Additionally, Map 1 shows that some of these states are close to the border with the United States. Other states in the middle of the country (Jalisco, Guanajuato and Michoacán) that have high federal crime rates are home to major drug trafficking organizations.⁴

Figure 1.

Federal crime incidents by males (18–24 years) in Mexico during 1997–2010.

Map 1.

Geographical distribution of federal crime incidents by males (18–24 years) during 1997–2010.

Our main variables in the vector of youth opportunity ( $Y {E_{it}}_{- 1}$ ) in equation (1) are male youth bulge, male youth education attainment rate, and male youth unemployment rate. We define the male youth bulge of 18–24-year-old males as a share of all males aged 18 years and above, capturing the dynamics in the younger working-age segments.⁵ The demographic data is sourced from Mexican population censuses carried out by INEGI across all the 32 Mexican states once every 10 years. Once every five years INEGI also conducts random surveys known as population counts. Thus, the data used to construct the youth bulge is sourced from the censuses of 1990, 2000, and 2010 (INEGI, 1990, 2000, 2010), and from the population surveys of 1995 and 2005 (INEGI, 1995, 2005). The youth education variable also originates from the census data, as well as the 2005 survey. This measures the proportion of males aged 18–24 years with at least a secondary education attainment normalized by the total male population aged 18–24 years. Youth unemployment is defined as the number of males aged 18–24 years who are reportedly unemployed divided by the total male labor force aged 18–24 years. The unemployment and labor force data are available from the Mexican census files for 1990, 2000 and 2010 only (INEGI, 1990, 2000, 2010). Missing years between the reported census and survey observations are interpolated. We believe this is defendable given that demographic and education variables normally change relatively slowly. We do acknowledge, however, that unemployment figures are likely to be much more volatile, and that the interpolation between the census observations is likely to miss considerable variation. While this is unfortunate, unemployment data based on census records is clearly preferable to less reliable survey data, given our aim to study age-, gender- and education-specific unemployment across all Mexican states over time.

We further disaggregate the youth unemployment data by the category of education, which is only possible given the use of census information, constructing data that as far as we know has not previously been used to test the youth opportunity and violence nexus. We specifically use unemployment rate in low education and high education strata, respectively, in specification (2):

Y C_{it} = γ URlowY {_{it}}_{- 1} + δ URhighY {_{it}}_{- 1} + β {Z_{it}}_{- 1} + ν_{i} + λ_{t} + {ω_{i}}_{t}

(2)

where, $URlowY {_{it}}_{- 1}$ denotes unemployment rate in low-education stratum, while $URhighY {_{it}}_{- 1}$ denotes unemployment rate in high-education stratum in state i and year t − 1 respectively. We first condense the categories for “no”, “primary” and “incomplete secondary” education into the low-education stratum, defined as those males aged 18–24 years with lower education than completed secondary level. We then divide the number of males who are unemployed in this category by the total male population aged 18–24 years with low education. Note that data on employment by education is available only from the 1990, 2000 and 2010 population censuses. Likewise, we categorize male youth in the high education strata as those aged 18–24 years who have completed secondary schooling or higher (including tertiary education). We then construct a measure for unemployment rate in high education strata by dividing unemployed male youth with high education, with total male population with high education, in the age group of 18–24 years. We also control for year- and state-fixed effects in equation (2).

We further examine under what conditions youth bulge can be associated with an increase in youth crimes using the specifications (3) and (4):

Y C_{it} = ζ {(URlow Y \times YB)_{it}}_{- 1} + γ URlowY {_{it}}_{- 1} + φ Y {B_{it}}_{- 1} + β {Z_{it}}_{- 1} + ν_{i} + λ_{t} + {ω_{i}}_{t}

(3)

Y C_{it} = ξ {(URhighY \times YB)_{it}}_{- 1} + δ URhighY {_{it}}_{- 1} + φ Y {B_{it}}_{- 1} + β {Z_{it}}_{- 1} + ν_{i} + λ_{t} + {ω_{i}}_{t}

(4)

where ${(URlow Y \times YB)_{it}}_{- 1}$ denotes unemployment rate in low-education stratum coupled with youth bulge in state i and year t − 1 in equation (3), while ${(URhighY \times YB)_{it}}_{- 1}$ is the interaction between unemployment rate in the high-education stratum and youth bulge in state i and year t − 1 in equation (4). These interactions help deduce whether the effect of youth bulge on violent crime is conditional upon unemployment rate in low- or high-education strata. As before, along with control variables we also include both year- and state-fixed effects.

Finally, the vector of control variables (Z_{it −}₁) includes other potential determinants of youth crime in state i during year t − 1, which we obtain from the extant literature on the subject. In selecting the controls, we follow earlier studies by Barakat and Urdal (2009), Demombynes and Ozler (2005), Fajnzylber et al. (2002), Hashimoto (1987), Miron (2001) and Urdal (2006). We were wary of the potential traps of the “garbage can model” (Achen, 2005) or “kitchen sink models” (Schrodt, 2014), in which all sorts of variables are dumped onto the right-hand side of the equation, making interpretation more difficult. Thus, we follow a conservative strategy of accounting only for known factors that affect youth crime. Accordingly, we include state per capita GDP (log) in US$ 2003 constant prices⁶ in state i during year t − 1 to proxy for the level of economic development. The income data is available from the National Accounts System of INEGI. Likewise, we also use state population (log), which is drawn from the population census data compiled by INEGI. We further include a measure on urban population, namely urbanization, which captures urban population as share of total population in state i during year t − 1. Urdal and Hoelscher (2012) point out that managing urban development sustainability poses significant challenges for the respective governments and therefore a large youth bulge in urban centres could be a source of instability and violence. We then include a measure of state governor elections. We follow Schneider (2013) to generate an indicator for the timing of elections that varies between 0 and 1. For all non-election years, the value is 0. For election years we make use of the following measure: [12 − (Mn− 1)]/12, wherein Mn is the month in which the state governor election took place. The data on the exact date and month in which the elections are held in each state was obtained from the state elections results and information published by the Institute of Marketing and Opinion (Instituto de Mercadotecnia y Opinión, 2012). Accordingly, for election years this indicator takes smaller values the later the election takes place within the year.⁷ The descriptive statistics are reported in Online Appendix 1 and the correlation matrix is in Online Appendix 2. The data sources and definitions are reported in Online Appendix 3. We estimate all our models using generalized least squares (GLS) controlling for two-way fixed effects. Using GLS over a simple ordinary least squares (OLS) estimator allows estimations in the presence of AR (1) autocorrelation within panels and cross-sectional heteroskedasticity across the panels.

Empirical results

Baseline results

Table 1 presents the baseline results estimated using specification (1) capturing the effects of youth bulge, youth education and youth unemployment rate on youth crime incidents. In Table 2 estimating specification (2), we disaggregate the youth unemployment rate by category of education, i.e. unemployment in the low-education strata. In Table 3, we estimate specification (3) by introducing the interaction between unemployment rate amongst low- and high-education strata and youth bulge. Finally, in Table 4 we replace the two youth unemployment measures for education with a measure, “Density of low opportunity cost youth”, capturing the overall “density” of unemployed male youth with low education as a percentage of all male youth. Beginning with column 1 in Table 1, the results show that the youth unemployment rate is positive and statistically significant at the 1% level. At the mean value of youth unemployment rate (2%) there is a 0.6% increase in youth crime incidents per capita (log), independent of a lagged dependent variable (we retain this lagged dependent variable hereafter in all our models). An increase by a standard deviation in youth unemployment rate (1.37%) above the mean increases the youth crime incidents per capita (log) by roughly 25%. These effects remain similar when we introduce various control variables in a stepwise manner in columns 2–4. Notice that these results marginally lose statistical significance in column 5 when we include all control variables into the model.

Table 1.

Effect of youth bulge and youth opportunity on youth crime.

Dependent variable: Federal youth crime incidents per head (log).

	(1)	(2)	(3)	(4)	(5)
Lagged dependent variable	0.357***	0.345***	0.341***	0.337***	0.315***
	(0.0428)	(0.0428)	(0.0427)	(0.0427)	(0.0430)
Male youth unemployment rate_t−1	0.0605***	0.0578**	0.0452*	0.0422*	0.0318
	(0.0226)	(0.0225)	(0.0235)	(0.0236)	(0.0234)
Male youth secondary school enrolment_t−1		−0.0182**	−0.0203***	−0.0159*	−0.0178**
		(0.00775)	(0.00782)	(0.00845)	(0.00895)
State per capita GDP (log)_t−1			0.457*	0.435*	0.632**
			(0.257)	(0.257)	(0.268)
Male youth bulge_t−1				−0.0557	−0.0393
				(0.0409)	(0.0404)
State population (log)_t−1					0.765**
					(0.325)
Urbanization_t−1					0.0224*
					(0.0116)
Timing of state governor elections					−0.114***
					(0.0402)
Constant	−6.351***	−5.823***	−9.648***	−8.431***	−23.03***
	(0.458)	(0.508)	(2.205)	(2.375)	(5.741)
Hausman test (P-value)	0.87	0.00	0.00	0.00	0.00
State-specific dummies	Yes	Yes	Yes	Yes	Yes
Year-specific dummies	Yes	Yes	Yes	Yes	Yes
Number of states	32	32	32	32	32
Observations	448	448	448	448	448

Notes: Robust standard errors in parentheses. ***P < 0.01, **P < 0.05, *P < 0.1.

Table 2.

Effect of youth unemployment rate by education category on youth crime.

Dependent variable: Federal youth crime incidents per-capita (log).

	(1)	(2)	(3)	(4)
Lagged dependent variable	0.354***	0.351***	0.340***	0.323***
	(0.0427)	(0.0426)	(0.0426)	(0.0429)
Unemployment rate in uneducated youth (males)_t−1	0.0652***	0.0574***	0.0542**	0.0373*
	(0.0214)	(0.0221)	(0.0220)	(0.0225)
State per capita GDP (log)_t−1		0.346	0.335	0.570**
		(0.252)	(0.251)	(0.270)
Male youth bulge_t−1			−0.0883**	−0.0740**
			(0.0376)	(0.0371)
State population (log)_t−1				0.847***
				(0.325)
Urbanization_t−1				0.0146
				(0.0109)
Timing of state governor elections				−0.113***
				(0.0403)
Constant	−6.385***	−9.345***	−7.444***	−22.80***
	(0.450)	(2.204)	(2.335)	(5.896)
Hausman test (P-value)	0.00	0.00	0.00	0.00
State-specific dummies	Yes	Yes	Yes	Yes
Year-specific dummies	Yes	Yes	Yes	Yes
Number of states	32	32	32	32
Observations	448	448	448	448

Notes: Robust standard errors in parentheses. ***P < 0.01, **P < 0.05, *P < 0.1.

Table 3.

Effect of youth unemployment rate—interactions with education category.

Dependent variable: Federal youth crime incidents per-capita (log).

	(1)	(2)	(3)	(4)
Lagged dependent variable	0.335***	0.325***	0.347***	0.326***
	(0.0423)	(0.0426)	(0.0429)	(0.0430)
Youth unemployment rate in low-education stratum (males)_t−1× youth bulge_t−1	0.0252***	0.0239**
	(0.00822)	(0.00954)
Youth unemployment rate in low-education stratum (males)_t−1	−0.459***	−0.455**
	(0.172)	(0.198)
Youth unemployment rate in high-education stratum (males)_t−1× youth bulge_t−1			0.00921	0.000827
			(0.00718)	(0.00807)
Youth unemployment rate in high-education stratum (males)_t−1			−0.146	0.00326
			(0.148)	(0.166)
Male youth bulge_t−1	−0.163***	−0.146***	−0.122***	−0.0764
	(0.0444)	(0.0467)	(0.0458)	(0.0477)
State per capita GDP (log)_t−1		0.686**		0.649**
		(0.272)		(0.270)
State population (log)_t−1		0.504		0.950***
		(0.351)		(0.340)
Urbanization_t−1		0.00930		0.0134
		(0.0110)		(0.0112)
Timing of state governor elections		−0.113***		−0.112***
		(0.0400)		(0.0404)
Constant	−3.145***	−17.13***	−3.775***	−24.66***
	(0.995)	(6.279)	(1.026)	(6.164)
Hausman test (P-value)	0.00	0.00	0.00	0.00
State-specific dummies	Yes	Yes	Yes	Yes
Year-specific dummies	Yes	Yes	Yes	Yes
Number of states	32	32	32	32
Observations	448	448	448	448

Notes: Robust standard errors in parentheses. ***P < 0.01, **P < 0.05, *P < 0.1.

Table 4.

Effect of the density of low-opportunity cost youth on youth crime.

Dependent variable: Federal youth crime incidents per-capita (log).

	(1)	(2)	(3)
Lagged dependent variable	0.354***	0.342***	0.333***
	(0.0426)	(0.0420)	(0.0424)
Youth unemployment density in low-education stratum (males)_t−1	0.199***	−1.746***	−2.033***
	(0.0648)	(0.537)	(0.613)
Youth unemployment density in low-education stratum(males)_t−1× youth bulge_t−1		0.0936***	0.104***
		(0.0255)	(0.0292)
Male youth bulge_t−1		−0.169***	−0.158***
		(0.0405)	(0.0418)
State per capita GDP (log)_t−1			0.680***
			(0.263)
State population_t−1			0.297
			(0.349)
Urbanization_t−1			0.0180*
			(0.0108)
Timing of state governor elections			−0.114***
			(0.0396)
Constant	−6.348***	−2.958***	−14.61**
	(0.443)	(0.913)	(6.110)
Hausman test (P-value)	0.00	0.00	0.00
State-specific dummies	Yes	Yes	Yes
Year-specific dummies	Yes	Yes	Yes
Number of states	32	32	32
Observations	448	448	448

Notes: Robust standard errors in parentheses. ***P < 0.01, **P < 0.05, *P < 0.1.

In column 2, we also include the youth education attainment ratio. As expected, higher levels of education have a negative effect on youth crime. The finding is significantly different from zero at the 5% level. The substantive effect suggests that at mean value (30.38%) an increase in youth education is associated with 0.2% decline in youth crime incidents per capita (log). However, an increase in youth education by a standard deviation (4.29%) lowers the average youth crime incidents per capita (log) by 63%. Interestingly, our crude measure for male youth bulge has a negative association with youth crime (columns 4 and 5), although this relationship is statistically insignificant. These results do not lend support to those who attribute crimes committed by youth in Mexico to a surge in youth bulge. In all of the tables reported here, the lagged dependent variable remains significantly different from zero at the 1% level. While the results for unemployment and education remain similar to those in column 4, the youth unemployment rate becomes marginally insignificant in column 5 when all control variables are included in the model. These results do not provide clear-cut evidence on the effect of youth unemployment on youth crime. We therefore disaggregate the unemployment levels among youth by low and high education in Table 2.

As seen in column 1 of Table 2, we find a positive effect of youth unemployment rate in the low-education stratum, which is statistically significant at the 1% level in column 1. The substantive effects suggest that a standard deviation increase in youth unemployment rate in the low-education stratum is associated with a 24% increase in youth crime incidents per capita (log), which is about 4% of the standard deviation of the dependent variable.⁸ Note that these results remain robust when we add other control variables in an incremental manner in columns 2–4. These results broadly support our hypothesis that the opportunity cost of engaging in violent crimes is lower among young unemployed men in the low-education stratum.

Conditional effects

Table 3 presents the interactive effects between youth bulge and youth unemployment rate by education category. First, in columns 1 and 2, we interact youth bulge and unemployment rate in the low-education stratum, and in columns 3 and 4 youth bulge and unemployment rate are coupled in the high-education stratum. As seen in column 1, we find that the interaction between youth bulge and unemployment in the low-education stratum has a positive effect on youth crime incidents per capita (log) and is significantly different from zero at the 1% level. This suggests that states with a higher percentage of male youth in their populations are more vulnerable to crime if the unemployment rate in the low-education stratum increases. In other words, a youth bulge is not a problem in itself, but rather the risk of violence is conditional upon higher levels of youth within the low-education stratum and thus scant employment opportunities.

To better understand the interaction effect, we rely on margins plot in Figure 2 (Greene, 2012). To calculate the marginal effect of an increase in the youth bulge variable, we take into account both the conditioning variable (unemployment rate in the low-education stratum) and the interaction outcome and display graphically the total marginal effect conditional on unemployment rate in the low-education stratum. The y-axis of Figure 2 displays the marginal effect of a unit increase of the youth bulge and the marginal effect is evaluated on the unemployment rate in the low-education stratum along the x-axis. Note that we include the 90% confidence interval. As seen in Figure 2, a unit increase in youth bulge decreases youth crime per capita (log) when the unemployment rate in the low-education stratum is lower than 3.72%. For instance, at 0.72% of unemployment rate in the low-education stratum crime per capita (log) is reduced by 16%, which is significantly different from zero at the 1% level (the marginal effects are significant and negative when the upper bound of the confidence interval is below zero). However, the margin plots also show that the impact of youth bulge on crime becomes positive but statistically insignificant once unemployment rate in the low-education stratum is over and above 3.72%, i.e. at moderate to high levels of the unemployment rate. These results suggest that states with low unemployment rate among low-educated youth are far less likely to witness crime by youth independently of variables such as state income, population, and urbanization, among other factors. Note that the three terms are all jointly significant (P < 0.0004).

Figure 2.

Youth bulge, unemployment rate in low-education stratum and marginal effect on youth crime.

We now turn to the interaction between youth bulge and unemployment rate in the high-education stratum in columns 3 and 4, Table 3. As seen, the interactive effect is not significantly different from zero. At the first glance, this suggests that larger youth bulges do not appear to increase the risk of violent youth crime even when the unemployment rate in the high-education stratum is increasing. However, as suggested above, the interaction results can be best assessed with a margins plot presented in Figure 3. The y-axis in Figure 3 displays the marginal effect by a unit increase of the youth bulge variable and along the x-axis with the unemployment rate in the high-education stratum at which the marginal effect is evaluated. Again, we include the 90% confidence interval. As seen there, an increase in youth bulge decreases the youth crime per capita (log) when the unemployment rate in the high-education stratum is lower than 5%. However, the margin plot also shows that the impact of youth bulge on youth crime incidents per capita (log) is positive, albeit statistically insignificant, when the unemployment rate in the high-education stratum is higher than 5%. These results suggest that the opportunity costs of engaging in crime are markedly higher for unemployed youth in the high-education stratum.

Figure 3.

Youth bulge, unemployment rate in high-education stratum and marginal effect on youth crime.

Lastly, in Table 4, we use “Density of low opportunity cost youth” capturing the “density” of unemployed male youths aged 18–24 with low education measured as the share of the total male youth population in that age group. We restrict our specification to only include unemployment in the low-education stratum since the relative number of unemployed youth with low education is the quantity of greatest theoretical relevance to the opportunity cost perspective. As seen in column 1, the density of unemployed youth with low education is positive and significantly different from zero at the 1% level. In columns 2 and 3, we interact youth bulge with the density variable wherein column 2 is a parsimonious model while in column 3 all control variables are included. As seen, the interactive effect is positive and significantly different from zero at the 1% levels in both columns 2 and 3. The marginal effects of the interaction variable are shown in Figure 4. Figure 4 shows that a unit increase in youth bulge would decrease the youth crime incidents per capita (log) (at the 90% confidence level) when the density of unemployed male youths with low education is lower than 0.9%. However, when the density of unemployed male youths with low education is high, at 2.7 and 3%, the impact of youth bulge on the youth crime incidents per capita (log) is positive and statistically significant at the 5% level. For instance, at 3% of unemployed male youth with low education, an additional point increase in youth bulge variable is associated with a 16% increase in youth crime per capita (log). These results highlight that, irrespective of the measure we use (unemployment rate or density), unemployment in the low-education stratum is the best predictor of youth crime in Mexico.

Figure 4.

Low opportunity costs (males), youth bulge and marginal effect on youth crime.

Before moving forward toward robustness checks, we will briefly discuss the results of the control variables in Tables 1 –4. Interestingly, we find evidence of a positive association between per capita state GDP and violent youth crime. These results are contrary to general expectations that higher levels of income are associated with lower levels of crime. It is noteworthy that urbanization, education and unemployment variables are controlled in the models, while the impact of income on these variables is not accounted for. For instance, as per capita income increases governments have more resources to spend on public services, such as crime prevention. It is then plausible that an increase in income per capita is associated with lower levels of actual crime but higher levels of reported crime as public expenditure on law and order allows for police to enforce the law more effectively. The other plausible explanation could be that it is not the level of income that matters for crime but how the income is distributed in the society. Next, after controlling for time- and state-fixed effects we find our population (log) to increase youth crime. Naturally, states with higher levels of population tend to witness more incidences of crime than sparsely populated states. Also, like others, we do find a strong positive impact on youth crime of the level of urbanization, which is consistent with the idea that urban environments are more conducive to violent crime (e.g. Moura and Silveira Neto, 2016; Urdal and Hoelscher, 2012). The variable capturing the timing of elections is associated with fewer crime incidents during the run-up toward governor elections. This might be due to two reasons. Firstly, there is every possibility of under-reporting of crime incidents during the run-up toward elections by the incumbent government. Second, it is also plausible that the incumbent governor would impose measures aimed at reducing violence during the election period, signaling to voters her/his commitment to control crime and restore law and order. This is similar to the political budget cycles theory for corruption suggested by Vadlamannati (2015) wherein an incumbent, in states within India, considers controlling corruption based purely on political considerations. We also cannot rule out the possibility that the result is driven by a combination of the two factors.

Endogeneity

Finally, we address the question of whether causality runs from youth unemployment and education measures to youth crime incidents per capita (log) or the other way around. It is quite possible that our key explanatory variables capturing youth opportunity are endogenous. That is, it might be that criminal activities attract more youth with low opportunity cost toward areas with high crime rates, and especially toward drug-related activities that might maximize their returns in the short run. This could affect the education and unemployment measures. It could also be that high levels of crime deter local investment, driving up unemployment levels. Although the case for reverse causality is indirect and presumably relatively weak, not taking this endogeneity into account might induce bias in our estimates. To determine the direction of causality, we use a dynamic model of Granger causality (Granger, 1969). Accordingly, once the past influence of y has been accounted for, the variable x is said to “Granger cause” the variable y if the past values of x help explain y (Engle and Granger, 1987). It is noteworthy that while the Granger causality investigates causality between two variables in time series data, it does not examine the true cause-and-effect relationship. Rather, it tests whether a particular variable comes before another in the time series. In other words, if we find Granger causality in the data it means variable x“Granger-causes” (i.e. precedes) variable y. We follow Dreher et al. (2012) to account for Granger causality in a panel setting as:

y_{it} = \sum_{j = 1}^{ρ} ψ_{j} y_{i, t - j} + \sum_{j = 1}^{ρ} ξ_{j} x_{i, t - j} + δ_{i} + ζ_{t} + {ω_{i}}_{t}

(5)

where the parameters are denoted as ψ_it and ξ_it for state i during the year t, and the maximum lag length is represented by ρ, while δ_i are unobserved individual effects, ζ_t are unobserved time effects and ω_it denotes the error term. Under the null hypothesis, the variable x is assumed to not Granger cause y, while the alternative hypotheses allow for x to Granger cause y after controlling for past influence of the variable y. We use three lags to estimate the models. Note that the joint F-statistic is used to gage the joint significance of the youth crime incidents per capita (log) on youth unemployment, education measures, and vice-versa. Following de Soysa and Vadlamannati (2012), we estimate four sets of Granger causality models in which set 1 captures estimations of youth unemployment Granger causing youth crime and vice-versa. In set 2 we test whether unemployment rate in low education strata Granger causes youth crime and vice-versa. Set 3 estimates whether unemployment rate in high education strata Granger causes youth crime and vice-versa. Lastly, set 4 tests whether the “density” of low opportunity costs of youth Granger causes youth crime and vice-versa. Granger causality test results for all four sets are reported in Tables A–D in the Supplemental Material. Our findings from all four sets reveal that we do not find any statistically significant effects of youth crime incidents per capita (log) on youth unemployment and education measures. The joint F-statistics show that none of the three lags in the youth crime display correlation with youth unemployment and education measures. Likewise, we do not find youth unemployment and education measures to explain increases in youth crime incidents per capita (log) as joint F-statistics is less than the thumb rule of 10 and remains statistically insignificant at all three lags (the results on Granger causality are available in the Online Appendix). Hence, our results reveal no significant reverse causality.

Robustness checks

We examined the robustness of our main findings in the following ways. First, we used alternative measures for the youth bulge, youth unemployment, and education variables. Departing from the measure of 18–24-year-old males, we used 18–30-year-old males as a share of all males aged 18 years and above. We also computed the federal crime incidents per capita registered in the age group of 18–30 years. Likewise, we also used the 18–30 year age group to compute unemployment rates by level of education. Using our alternative measures does not alter our results significantly. We still find that the unemployment rate in the low-education stratum matters the most. The results for the interaction between youth bulge and unemployment rate in the low-education stratum remain robust. These results are reported in Table E in the Supplemental Material. Second, we re-estimated our GLS fixed effects models with negative binomial models where we used the dependent variable as an event count of youth federal crime incidents in the male 18–24-year-old category. We also control for year- and state-specific dummies. The results (in Table F in the Supplemental Material) estimated using negative binomial methods remain similar to those reported in baseline models in Tables 1 –4. Third, as an additional test for robustness, we exclude the few observations with extreme values in youth crime incidents reported. Excluding outliers, the baseline results remain robust (Table G, Supplemental Material), suggesting that results are not driven by extreme values. Next, we replicate the GLS fixed effects models with a simple OLS fixed effects estimator. Our results (reported in Table H) remain robust to using the OLS estimator.

Finally, we attempt to control for some form of governance and institutional quality, which can significantly affect crime. Unfortunately, measures like rule of law, corruption or deployment of police, which are usually associated with governance strength, are not available at the subnational level in Mexico. Therefore, we use two crude proxy measures for governance. First, we use a measure of federal crime agencies per capita (log). The crime agencies are independent of state police and report the federal crime incidents. They also assist the state police force in carrying out investigations into these crimes. We use the data on number of federal crime agencies located in each state during 1998–2010 period. Thus, per capita crime agencies (log) is our measure of institutional effectiveness. Our baseline estimations remain robust to controlling for crime agencies, which in itself remains statistically insignificant (Table I, Supplemental Material). Second, we use a GINI coefficient index available across states as a proxy for governance. While not a conventional measure of governance, there is broad support among academics and policy makers that good governance provides the fundamental basis for economic development (e.g. Kaufmann et al., 2005). Good governance matters not only for income per capita but also for substantially reducing illiteracy and infant mortality. Hence, through reduction in illiteracy and health outcomes, good governance can help to change the access to opportunities for the most disadvantaged in society, leading to a reduction in inequality and hence in crime. It is important to note that inequality has been regarded as a determinant of violent crime and drug war in Mexico (Enamorado et al., 2016). Hence, in the absence of a more direct measure, we use GINI index as a crude proxy for governance. Our results remain robust to controlling for GINI measure, while the effect of GINI on youth crime remains statistically insignificant (Table J, Supplemental Material). The results of these robustness checks are available in the Online Appendix.

Conclusion

This article investigates potential causes for variation in violent youth crime across Mexican states, with a particular focus on the role of youth opportunities. Building on an opportunity framework prominent both in the civil war and criminology literatures, we initially hypothesized that violent crime should vary with demographic age structure, so that Mexican states with large youth bulges should have higher levels of violent crime, everything else being equal. This expectation is not borne out by the empirical models, however, as our measure for male youth bulge is consistently negatively associated with violent crime rates. We further hypothesized that the two factors that arguably most strongly determine the actual opportunity cost for youth, levels of education and employment, should be associated with crime levels, and particularly so when low education levels and high unemployment levels occur in states with large male youth bulges. These much more specific expectations regarding youth opportunities are not easily tested for global cross-national samples owing to data limitations. The availability of reliable and comparable census data for Mexico providing age- and gender-specific educational attainment and unemployment rates at the state level allows for a detailed subnational panel study of youth opportunities and violent crime. Our empirical models, also taking into account possible confounding factors and endogeneity, find strong support for the importance of youth opportunities. This pertains in particular to educational attainment as our models consistently find low levels of education to be strong predictors of high levels of violent crime. We further find that high unemployment among males with low education is clearly associated with higher crime rates, and that this effect is amplified by an interaction with large male youth bulges. No similar effect is found for high unemployment among males with higher education levels, suggesting that the higher opportunity cost of youth with at least completed secondary education may inhibit recruitment to criminal organizations.

This study provides some crucial insights into the complex root causes for the high levels of violent crime in emerging economies undergoing extensive economic, social, and demographic change. While being a middle-income country with relatively well-developed institutions, Mexico is experiencing a de facto lack of territorial control over certain geographical areas to drug cartels, and levels of violence that vastly surpass most contemporary armed conflicts. As such, improving knowledge of structural factors determining violent crime and ultimately building increased capacity to reduce crime have implications for understanding the security situation of the greater region as challenges pertaining to gang violence and drug trafficking extend beyond the Mexican context. Furthermore, the findings reported here may have implications for understanding the drivers of violent crime beyond the Latin American context and should spur more detailed data collection and empirical study of youth opportunities and violence elsewhere.

The extensive challenge with youth across Latin America falling out of education as well as employment, generally referred to as “ninis”, has been noted in recent policy analyses (de Hoyos et al., 2016). With ninis accounting for one in five youth between the ages of 15 and 24 in the region, Latin America is on a path to potentially passing on their “demographic window of opportunity” (Birdsall and Sinding, 2001) offered by the very low dependency ratios projected for the 2015–2035 period (de Hoyos et al., 2016: 6–7). With Mexico being one of the countries in the region with levels of ninis above the regional average, implications of the findings of this study strongly resonate with policy recommendations for extending education and reducing drop-outs through demand-side (e.g. financial incentives) as well as supply-side (e.g. pedagogical and early warning) interventions. More challenging yet are youth employment programs, with formal sector employment programs delivering mixed success, calling for context-sensitive diagnostics and implementation (de Hoyos et al., 2016: 31–32). With the combined potential for a lost demographic dividend and detrimental developmental consequences of political and criminal violence (Benyishay and Pearlman, 2014; World Bank, 2011), the underinvestment in young people’s human capital represents a double developmental challenge.

Supplemental Material

SUPPLEMENTARY_FILES – Supplemental material for The significance of age structure, education, and youth unemployment for explaining subnational variation in violent youth crime in Mexico

Supplemental material, SUPPLEMENTARY_FILES for The significance of age structure, education, and youth unemployment for explaining subnational variation in violent youth crime in Mexico by Nicolás Corona Juárez, Henrik Urdal and Krishna Chaitanya Vadlamannati in Conflict Management and Peace Science

Footnotes

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Krishna Chaitanya Vadlamannati

Supplemental material

Supplemental material for this article is available online.

Notes

References

Achen

(2005) Let’s put garbage can regressions and garbage can probits where they belong. Conflict Management and Peace Science 22(4): 327–339.

Avison

Loring

(1986) Population diversity and cross-national homicide: The effects of inequality and heterogeneity. Criminology 24(4): 733–749.

Barakat

Urdal

(2009) Breaking the waves? Does education mediate the relationship between youth bulges and political violence. World Bank Policy Research Working Paper 5114.

Braungart

(1984) Historical and generational patterns of youth movements: A global perspective. Comparative Social Research 7: 3–62.

Becker

(1968) Crime and punishment: An economic approach. Journal of Political Economy 76(2): 169–217.

Benyishay

Pearlman

(2014) Crime and microenterprise growth: Evidence from Mexico. World Development 56: 139–152.

Berman

Callen

Felter

Shapiro

(2011) Why do working men rebel? Insurgency and unemployment in Afghanistan, Iraq, and the Philippines. Journal of Conflict Resolution 55(4): 496–528.

Birdsall

Sinding

(2001) How and why population matters: New findings, new issue. In Birdsall

Kelley

Sinding

(eds) Population Matters: Demographic Change, Economic Growth, and Poverty in the Developing World. New York: Oxford University Press, pp. 3–23.

Blattman

Annan

(2016) Can employment reduce lawlessness and rebellion? A field experiment with high-risk men in a fragile state. American Political Science Review 110(1): 1–17.

10.

Brett

Specht

(2004) Young Soldiers: Why They Choose to Fight. Boulder, CO: Lynne Rienner.

11.

Buhaug

Rød

(2006) Local determinants of African civil wars, 1970–2001. Political Geography 25(3): 315–335.

12.

Cincotta

Engelman

Anastasion

(2003) The Security Demographic: Population and Civil Conflict After the Cold War.Washington, DC: Population Action International.

13.

Choucri

(1974) Population Dynamics and International Violence: Propositions, Insights and Evidence. Lexington, MA: Lexington.

14.

Cole

Gramajo

(2009) Homicide rates in a cross-section of countries: Evidence and interpretation. Population and Development Review 35(4): 749–776.

15.

Collier

(2000) Doing well out of war: An economic perspective. In Berdal

Malone

(eds) Greed & Grievance: Economic Agendas in Civil Wars. Boulder, CO: Lynne Rienner, pp. 91–111.

16.

Collier

Hoeffler

(2004) Greed and grievance in civil war. Oxford Economic Papers 56: 563–595.

17.

Collier

Elliott

Hegre

Hoeffler

Reynal-Querol

Sambanis

(2003) Breaking the conflict trap: Civil war and development policy. World Bank Policy Research Report. Oxford: Oxford University Press and Washington, DC: The World Bank.

18.

de Hoyos

Rogers

Székely

(2016) Out of school and out of work: Risk and opportunities for Latin America’s Ninis. Washington, DC: World Bank. Available from: https://openknowledge.worldbank.org/handle/10986/22349.

19.

Dell

Feigenberg

Teshima

(2019) The violent consequences of trade-induced worker displacement in Mexico. AER: Insights 1(1): 43–58.

20.

Demombynes

Ozler

(2005) Crime and local inequality in South Africa. Journal of Development Economics 76(2): 265–292.

21.

de Soysa

Vadlamannati

(2012) Do pro-market economic reforms drive human rights violations—An empirical assessment. Public Choice 150: 1–25.

22.

Dreher

Gassebner

Siemers

LHR

(2012) Globalization, economic freedom, and human rights. Journal of Conflict Resolution 56(3): 516–546.

23.

Dreze

Khera

(2000) Crime, gender, and society in India: Insights from homicide data. Population and Development Review 26(2): 335–352.

24.

Easterlin

(1987) Easterlin hypothesis. In Eatwell

Millgate

Newman

(eds) New Palgrave: A Dictionary of Economics, Vol. 2. New York: Stockton, pp. 1–4.

25.

Elbadawi

Sambanis

(2000) Why are there so many civil wars in Africa? Understanding and preventing violent conflict. Journal of African Economies 9(3): 244–269.

26.

Enamorado

Lopéz-Calva

Castelán

Winkler

(2016) Income inequality and violent crime: Evidence from Mexico’s drug war. Journal of Development Economics 120: 128–143.

27.

Engle

Granger

CWJ

(1987) Co-integration and error correction: Representation, estimation, and testing. Econometrica 55(2): 251–276.

28.

Fajnzylber

Lederman

Loayza

(2002) Inequality and violent crime. Journal of Law and Economics 45(1): 1–39.

29.

Fearon

Laitin

(2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 75–90.

30.

Fox

Hoelscher

(2012) Political order, development and social violence. Journal of Peace Research 49(3): 431–444.

31.

Gartner

Parker

(1990) Cross-national evidence on homicide and the age structure of the populatio. Social Forces 69(2): 351–371.

32.

Granger

CWJ

(1969) Testing for causality and feedback. Econometrica 37(3): 424–438.

33.

Greene

(2012) Econometric Analysis, 7th edn. Upper Saddle River, NJ: Prentice Hall.

34.

Goldstone

(1991) Revolution and Rebellion in the Early Modern World. Berkely, CA: University of California Press.

35.

Goldstone

(2001) Demography, environment, and security. In Diehl

Gleditsch

(eds) Environmental Conflict. Boulder, CO: Westview, pp. 84–108.

36.

Goldstone

Kaufmann

Toft

(eds) (2012) Political Demography: How Population Changes Are Reshaping International Security and National Politics. Oxford: Oxford University Press.

37.

Hansmann

Quigley

(1982) Population heterogeneity and the sociogenesis of homicide. Social Forces 61(1): 206–224.

38.

Hashimoto

(1987) The minimum wage law and youth crimes: Time-series evidence. Journal of Law and Economics 30(2): 443–464.

39.

Hausman

(1978) Specification tests in econometric. Econometrica 46(6): 1251–1271.

40.

Hirschi

Gottfredson

(1983) Age and the explanation of crime. American Journal of Sociology 89(3): 552–584.

41.

Hoelscher

(2015) Politics and social violence in developing democracies: Theory and evidence from Brazil. Political Geography 44(1): 29–39.

42.

Hoelscher

Miklian

Vadlamannati

(2012) Hearts and mines: A district-level analysis of the Maoist Conflict in India. International Area Studies Review 15(2): 141–160.

43.

INEGI (1990) 1990 Censo General de Población y Vivienda. [1990 General Census on Population and Housing.] Aguascalientes, Mexico. Available from: www.inegi.org.mx.

44.

INEGI (1995) 1995 Conteo de Población y Vivienda. [1995 Population and Housing Count.] Aguascalientes, Mexico. Available from: www.inegi.org.mx.

45.

INEGI (2000) 2000 Censo General de Población y Vivienda. [2000 General Census on Population and Housing.] Aguascalientes, Mexico. Available from: www.inegi.org.mx.

46.

INEGI (2005) 2005 Conteo de Población y Vivienda. [2005 Population and Housing Count.] Aguascalientes, Mexico. Available from: www.inegi.org.mx.

47.

INEGI (2010) 2010 Censo General de Población y Vivienda. [2010 General Census on Population and Housing.] Aguascalientes, Mexico. Available from: www.inegi.org.mx.

48.

INEGI (2012) Sistema de Cuentas Nacionales. [National Accounting System. Economic Data Bank]. Aguascalientes, Mexico. Available from: www.inegi.org.mx.

49.

Instituto de Mercadotecnia y Opinión (2012) Institute of Marketing and Opinion. Elecciones Estatales. [State Governor Elections.] Jalisco, Mexico. Available from: http://imocorp.com.mx/.

50.

Instituto Mexicano de la Juventud y Secretaría de Educación Pública, México (2008) Perspectiva de la Juventud en México. [Youth Perspective in Mexico]. Ministry of Public Education. Mexico.

51.

Kaufmann

Kraay

Mastruzzi

(2005) Governance matters IV: Governance indicators for 1996–2004. Policy Research Working Paper 3237. Washington, DC: World Bank.

52.

Korenman

Neumark

(1997) Cohort crowding and youth labor markets: A cross-national analysis. Working Paper 6031. Cambridge, MA: National Bureau of Economic Research.

53.

Machunovich

(2000) Relative cohort size: Source of a unifying theory of global fertility transition. Population and Development Review 26(2): 235–261.

54.

Mesquida

Wiener

(1996) Human collective aggression: A behavioral ecology perspective. Ethology and Sociobiology 17: 247–262.

55.

Miron

(2001) Violence, guns, and drugs: A cross-country analysis. Journal of Law and Economics 44(2): 615–633.

56.

Moller

(1968) Youth as a force in the modern world. Comparative Studies in Society and History 10: 238–260.

57.

Moura

Neto

(2016) Individual and contextual determinants of victimisation in Brazilian urban centres: A multilevel approach. Urban Studies 53(8): 1559–1573.

58.

Neapolitan

(1997) Cross-National Crime—A Research Review and Sourcebook. Westport, CT: Greenwood Press.

59.

Neumayer

(2003) Good policy can lower violent crime: Evidence from a cross-national panel of homicide rates, 1980–9. Journal of Peace Research 40(6): 619–640.

60.

Neumayer

(2005) Inequality and violent crime: Evidence from data on robbery and violent theft. Journal of Peace Research 42(1): 101–112.

61.

Nivette

(2011) Cross-national predictors of crime: A meta-analysis. Homicide Studies 15(2): 103–131.

62.

Østby

Urdal

Tadjoeddin

Murshed

Strand

(2011) Population pressure, horizontal inequality and political violence: A disaggregated study of Indonesian provinces, 1990–2003. Journal of Development Studies 47(3): 377–398.

63.

Pampel

Gartner

(1995) Age structure, socio-political institutions, and national homicide rate. European Sociological Review 11(3): 243–260.

64.

Pridemore

(2011) Poverty matters: A reassessment of the inequality–homicide relationship in cross-national studies. British Journal of Criminology 51: 739–772.

65.

Robbins

Pettinicchio

(2012) Social capital, economic development, and homicide: A cross-national investigation. Social Indicators Research 105(3): 519–540.

66.

Rogers

Pridemore

(2016) The (null) effects of percent young on 15 to 24 age-specific and male- and female-specific cross-national homicide rates. Homicide Studies 20(3): 257–292.

67.

Sambanis

(2002) A review of recent advances and future directions in the quantitative literature on civil war. Defence and Peace Economics 13(3): 215–243.

68.

Schneider

(2013) Globalizing electoral politics: Political competence and distributional bargaining in the European Union. World Politics 65(3): 452–490.

69.

Schrodt

(2014) Seven deadly sins of contemporary quantitative political analysis. Journal of Peace Research 51(2): 287–300.

70.

Urdal

(2006) A clash of generations? Youth bulges and political violence. International Studies Quarterly 50(3): 607–630.

71.

Urdal

(2008) Population, resources and violent conflict: A sub-national study of India, 1956–2002. Journal of Conflict Resolution 52(4): 590–617.

72.

Urdal

Hoelscher

(2012) Explaining urban social disorder and violence: An empirical study of event data from Asian and Sub-Saharan African cities. International Interactions 38(4): 512–528.

73.

Vadlamannati

(2011) Why Indian men rebel? Explaining armed rebellion in north-eastern states of India, 1970–2007. Journal of Peace Research 48(5): 605–619.

74.

Vadlamannati

(2015) Fighting corruption or elections? The politics of anti-corruption policies in India: A subnational study. Journal of Comparative Economics 43(4): 1035–1052.

75.

Widner

Reyes-Loya

Enomoto

(2011) Crimes and violence in Mexico: Evidence from panel data. Social Science Journal 48(4): 604–611.

76.

World Bank (2011) World Development Report 2011: Conflict, Security, and Development. Washington, DC: The World Bank.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.27 MB