temporal and geographic trends in homicide and suicide rates in mexico

Available online 22 October 2014

Keywords:HomicidesSuicidesViolenceMexicoGeneralized linear mixed model

temporal and geographic trends in the ofcially registered violence-related deaths

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699

. . . . . . . . . 706

. . . . . . . . . 707

Aggression and Violent Behavior 19 (2014) 699707

Contents lists available at ScienceDirect

Aggression and V1. Introduction4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7002.1. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7002.2. Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700

3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7013.1. Homicides. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7013.2. Suicides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704 The authors gratefully acknowledge the assistanInstitute of Geography of the National Autonomous Univethe maps in Fig. 3.

Corresponding author.E-mail addresses: [email protected] (I. Leene

(A. Cervantes-Trejo).

http://dx.doi.org/10.1016/j.avb.2014.09.0041359-1789/ 2014 Elsevier Ltd. All rights reserved.Contentseffects logistic regression models were tted, separately for men and women and different age groups, where(a) the evolution of the log-odds of the homicide and suicide rates over time was assumed to follow a piecewiselinear function and (b) the geographic variation in the latter function was accounted for by random effects asso-ciated with the 2,456 municipalities in the country. The homicide analyses show that, although the absolutehomicide rates strongly differ betweenmen and women (with a factor of about 10), the overall temporal trendsforwomen between 16 and 49 years andmen between 16 and 60 years of age are remarkably similar, in that therates decrease until 2007 and strongly increase afterwards. In absolute terms,men between 20 and 49 years formthe most vulnerable group with an averaged homicide rate in 2012 of over 40 per 100,000. Geographically, homi-cide rates in 2012 are found to be the highest in the states known for the drugs-related violence, which are also thestates where the increase between 2007 and 2010 was the strongest. As to the suicide rates, a steady increase wasfound over the 15-year study period in females between 12 and 39 years and in men between 12 and 49 years ofage, while in the other age groups, the rates remained relatively constant. On average, a completed suicide is abouteight timesmore likely in adultmen than in adultwomen (although important interactionswith age should be con-sidered); men between 20 and 49 years of age, together with those over 75, are most vulnerable, with an averagesuicide rate of over 10 per 100,000 in 2012. In order to decrease the levels of violence inMexico,moving beyond thepolitical rhetoric and the implementation of evidence-based prevention programs are imperative.

2014 Elsevier Ltd. All rights reserved.Received 21 August 2014Accepted 27 September 2014

(216,462 homicides and 77,334 suicides) that occurred between 1998 and 2012 on Mexican territory. Mixed-

Article history: We analyze and discuss thea b s t r a c ta r t i c l e i n f oTemporal and geographic trends in homicide and suicide rates inMexico,from 1998 through 2012

Iwin Leenen a,, Arturo Cervantes-Trejo b,c

a Faculty of Medicine, National Autonomous University of Mexico, Mexicob Carlos Peralta Chair of Public Health, Anahuac University, Mexicoc National Institute for the Evaluation of Education (INEE), Mexicoce of Luis Chias Becerril of thersity of Mexico in constructing

n), [email protected] BehaviorIn 1996, the World Health Organization dened violence as theintentional use of physical force or power, threatened or actual, againstoneself, another person, or against a group or community, which eitherresults in or has a high likelihood of resulting in injury, death,

Experts agree that deaths, injuries and disabilities due to violence areexpected to increase in the world, especially in low- and middle-

700 I. Leenen, A. Cervantes-Trejo / Aggression anincome countries (United Nations Ofce on Drugs & Crime [UNODC],2014; VPA, 2012). Recent data show that Central America, togetherwith Southern Africa, has the highest homicide rates in the world,reaching above 24 victims per 100,000 population in 2012, whereasthe global, worldwide homicide rate amounted to 6.2 per 100,000;when considering the ve global regions dened by the United Nations,homicide rates are highest in the Americas with 16.3 homicides per100,000 as compared to 2.9 in Asia, 3.0 in Oceania and Europe, and12.5 in Africa (UNODC, 2014). Conversely, suicide rates in theAmericaswith 7.9 per 100,000 aretogetherwith Africa and theEasternMediterranean regionconsiderably lower as compared to Europe,South-East Asia (which includes India), and the Western Pacic region(including China and Japan), where about 14 suicides occur per100,000 population (Vrnik, 2012).

In the last decade, mortality from violence, and particularly homi-cide, has strongly increased in Mexico, which moved the country intothe group of Latin American countries with the highest rates ofviolence-related deaths (Gawryszewski, Sanhueza, Martinez-Piedra,Escamilla, & Marinho de Souza, 2012; UNODC, 2014). Although thisincrease has been amply documented and discussed in the popular aswell as the academic press (Dube, Dube, & Garca-Ponce, 2013;Gonzlez-Prez et al., 2012; Hernndez-Bringas & Flores-Arenales,2011; Salama, 2013; Vilalta, 2013, 2014), the evolution over time hasbeen studied either globally or at the level of distressed areas or states.The present study examines the overall temporal pattern of change inhomicides and suicides since 1998 in Mexico, segregated by genderand age groups, and how this temporal evolution spatially distributesat the aggregate level of municipalities. The ndings are discussed inrelation to existing literature on the psychological and sociologicalimpact of living and growing up in a violence-stricken neighborhood.We conclude with some implications for policy-making.

2. Method

2.1. Data

The following two sources of data were combined in the presentproject. Both sources are made available by Mexico's National Instituteof Statistics, Geography and Information (Instituto Nacional deEstadstica y Geografa [INEGI], n.d.). On the one hand, we downloadedpopulation counts from the Mexican censuses in the years 2000, 2005,and 2010, separated for each of the 2,456 municipalities in the countryand segregated by sex and for each year of age. Subsequently, the countsby year of age were aggregated in ten age groups (011, 1215, 1619,2023, 2429, 3039, 4049, 5059, 6074, and 75 or older), so that thenumber of individuals in each of the 20 combinations of sex and agegroups (further called sexage groups) were obtained for each munici-psychological harm, maldevelopment or deprivation (as cited inDahlberg and Krug (2002), p. 5). Since the Forty-Ninth World HealthAssembly [WHA] (1996) declared violence a major and growing globalpublic health problem, many international efforts, such as the GlobalBurden of Disease study (Lozano et al., 2012; Murray et al., 2012), theWorld Report on Violence and Health (Krug, Dahlberg, Mercy, Zwi, &Lozano, 2002) and the associated Global Campaign on Violence Pre-vention (Violence Prevention Alliance [VPA], 2012), have broughtevidence-based violence prevention onto the public health agenda(see also, Rutherford, Zwi, Grove, and Butchart (2007a, 2007b)).

Every year in the world, an estimated 1.34 million violent deathsoccur from intentional injury, of which around 880,000 are self-inicted (i.e., suicides) and 460,000 are other-inicted (i.e., homicides;Lozano et al., 2012). Worldwide, both homicide and suicide rankamong the leading causes of death among those aged 1544 years.pality in each of the three census years. By linear interpolation,population counts were estimated for the other years covered by the15 year study period.

On the other hand, we consulted the ofcial mortality data sets ofMexico, inwhich, among others, details on the cause of death (classiedaccording to the international classication of diseases ICD-10, WorldHealth Organization [WHO], 2011), exact time of death, municipalityand date of registration, and sex and age of the deceased are recorded.From this data base, we extracted all deaths classied either as homi-cides (i.e., categories X85Y09, Y35, Y36, Y87.1 and Y89.9 in the ICD-10 classication) or as suicides (i.e., ICD-10 categories X60X84 andY87.0). The cases for which themunicipality of registrationwasmissingor for which the registration occurred more than a year after the actualdecease (0.9% for homicides and 0.5% for suicides) were removed. If sexand/or age of the victim was missing (3.1% of the homicides and 0.7%of the suicides), the observation was randomly redistributed takinginto account the actual distribution of sex and age in the population ofhomicides/suicides across the country.

Combining both sources of information resulted into a data set withthe number of residents as well as the number of homicides andsuicides registered in each municipality, by sexage group, and foreach year included in the study.

2.2. Statistical analysis

These datawere analyzed, separately for homicides and suicides andfor each sexage group, by means of multilevel logistic regressionmodels (which belong to the family of hierarchical or mixed-effectsgeneralized linear models; see Snijders & Bosker, 2012, chap. 17;Wong & Mason, 1985). In view of the small numbers of suicides in thelowest age group (an average of 1.0 per 1,000,000 per year, boys andgirls taken together) and some methodological issues in classifyingchild mortality as suicides (Crepeau-Hobson, 2010), we excludedchildren younger than 12 years of age from the analysis.

At the lowest level of themodel, the number Yit of homicides or sui-cides in the sexage group under study that occurred in municipality i(i=1,, 2456) during year t (t= 1998, ,2012) is assumed to followa binomial distribution with parameters nit (i.e., the number of inhabi-tants belonging to the given sexage group in municipality i in year t)and it (i.e., the probability that an individual from this subpopulationbecomes the victim of homicide or commits suicide). It is furtherassumed that, within municipality i, the log-odds of the latter probabil-ities follow a (continuous) piecewise linear function of time, parame-trized as follows:

logit

1it

0i mj1 ji max t; j

j1 h i

I tb j1

; 1

where m denotes the number of linear pieces and 0, , m thebreakpoints in descending order (i.e., the years where the respectivelinear pieces begin and end, with 0 2012 and m 1998), whichare specied as constants in the analysis (see below). The indicatorfunction I(expression) equals 1 if expression is true, and 0 otherwise. Asillustrated in Fig. 1, the parametrization in Eq. (1) renders 0i the inter-pretation of the (log-odds of the) homicide or suicide rate inmunicipal-ity i in the year 2012, while ji = (j= 1,,m) indicates how this ratehas changed (per year) between j and j 1.

Whereas at the rst level the temporal evolution within municipal-ities is modeled, the second level species parameters which allowstudying the variation amongmunicipalities with regard to this tempo-ral evolution. In particular, the between-municipality variation in eachof the parameters ji at the right-hand side of Eq. (1) is modeled asfollows:

ji 0 j S jXSi M jXMi L jXLi B jXBi uji; 2

d Violent Behavior 19 (2014) 699707

for j=0,,m. The XSi, XMi, XLi, and XBi in the latter equation are dummyvariables which indicate whether municipality i is small (i.e., the totalnumber of inhabitants across all age and sex groups, averaged overthe 15 years between 1998 and 2012, is lower than 10,000), medium-sized (between 10,000 and 100,000 inhabitants), large (between100,000 and 500,000 inhabitants), or big (over 500,000 inhabitants).

3. Results

Table 1 presents a summary of the raw data, with nationwide popu-lation sizes, and number of homicides and suicides per year and sex. Inthe 15-year period covered by the study, a total of 216,462 homicides(192,259 males and 24,203 females) and 77,334 suicides (53,131

Fig. 1. Elaboration and graphical representation of the piecewise linear function in Eq. (1) for the case that m= 3 and 0 = 2012, 1 = 2010, 2 = 2007, and 3 = 1998. This exampleillustrates the interpretation of the model parameters, with 0i being the intercept at 2012, and 1i, 2i, and 3i, the increase/decrease between successive breakpoints, 20102012,20072010, and 19982007, respectively.

00,0

701I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 699707As a result, the associated parameters express the differences amongthe four categories of municipalities with respect to the intercepts andslopes of the piecewise linear function in Eq. (1). In order to resolvethe unidentiability of the model (due to the interdependence of thedummy variables), we restricted Sj + Mj + Lj + Bj = 0 for all j, sothat 0j can be interpreted as an overall mean intercept/slope acrossmunicipalities. The random effects (u0i, u1i,, umi) account for the re-sidual variance amongmunicipalities and are assumed to be identicallyand independently drawn (iid) from amultivariate normal distribution,with means equal to 0 and a positive-denite covariance matrix.

For each sexage group, 379 variants of the described model weretted to the data, obtained by varying the value of m from 0 to 4 andby considering all possible combinations for the associated breakpoints1,, m 1 given the value for m (which resulted in 1, 1, 13, 78, and286 models for m equal to 0, 1, 2, 3, and 4, respectively). Among these379 variants, the model with the lowest value for the BIC statistic(Schwarz, 1978; Zucchini, 2000) was selected and will be discussedbelow.

All models were tted bymeans of the PROC GLIMMIX procedure inSAS Version 9.2, using maximum likelihood estimation based on theLaplace approximation (SAS Institute, 2013).

Table 1Nationwide population sizes, number of homicides and suicides, and associated rates per 1

HomicidesMales Females

Year Population Count (rate) Population Count (rate)

1998 47,201,666 11,940 (25.3) 49,412,151 1529 (3.1)1999 47,457,624 10,694 (22.5) 49,714,147 1407 (2.8)2000 47,713,583 9309 (19.5) 50,016,144 1291 (2.6)2001 48,225,500 8876 (18.4) 50,620,136 1300 (2.6)2002 48,737,418 8752 (18.0) 51,224,129 1283 (2.5)2003 49,249,335 8680 (17.6) 51,828,122 1322 (2.6)2004 49,761,253 8073 (16.2) 52,432,114 1208 (2.3)2005 50,273,170 8600 (17.1) 53,036,107 1294 (2.4)2006 51,189,582 9079 (17.7) 53,925,147 1295 (2.4)2007 52,105,994 7711 (14.8) 54,814,187 1080 (2.0)2008 53,022,407 12,726 (24.0) 55,703,227 1434 (2.6)2009 53,938,819 17,817 (33.0) 56,592,267 1943 (3.4)2010 54,855,231 23,452 (42.8) 57,481,307 2429 (4.2)2011 55,313,437 24,432 (44.2) 57,925,827 2732 (4.7)2012 55,771,643 22,118 (39.7) 58,370,347 2656 (4.6)

Note. Infants up to 11 years are excluded from the suicide data, which explains the different pomales and 24,203 females of 12 year or older) were registered acrossall municipalities. The increase in both homicides and suicides asdiscussed in Section 1 is apparent from Table 1. In the remainder ofthis section we present the results on how these temporal trends arebest summarized in each sexage group, and how they differ amongmunicipalities, for homicides and suicides, respectively.

3.1. Homicides

The main results of the analyses of the homicide data based on themultilevel model described in the previous section are presented inTable 2 (females) and Table 3 (males). In the rst place, from thecolumn labeled Parameter it can be derived which model (i.e., withhow many linear pieces and the associated breakpoints) best ttedthe data for each sexage group. For example, the evolution of thefemale infanticide rate (i.e., girls from 0 to 11 years) tted a linearmodellinear in the log-oddsthat shows a nonsignicant increasebetween 1998 and 2012 (note that the value for the slope equals0.014 units per year and that the associated 95%-condence intervalincludes 0), leading to a level of11.76 in 2012. As explained in thepre-vious section, the parameters are expressed on the log-odds scale,

00 inhabitants for males and females in each year between 1998 and 2012.

SuicidesMales Females

Population Count (rate) Population Count (rate)

33,467,473 2769 (8.3) 36,077,769 517 (1.4)33,777,314 2802 (8.3) 36,436,040 493 (1.4)34,087,155 2911 (8.5) 36,794,311 538 (1.5)34,706,838 3081 (8.9) 37,510,854 665 (1.8)35,326,520 3183 (9.0) 38,227,396 651 (1.7)35,946,202 3385 (9.4) 38,943,938 687 (1.8)36,565,884 3393 (9.3) 39,660,481 667 (1.7)37,185,566 3553 (9.6) 40,377,023 730 (1.8)38,045,581 3522 (9.3) 41,207,151 705 (1.7)38,905,595 3610 (9.3) 42,037,279 768 (1.8)39,765,609 3798 (9.6) 42,867,408 852 (2.0)40,625,623 4144 (10.2) 43,697,536 985 (2.3)41,485,637 4065 (9.8) 44,527,664 913 (2.1)41,915,645 4583 (10.9) 44,942,728 1080 (2.4)42,345,652 4332 (10.2) 45,357,792 1043 (2.3)

pulation sizes for homicides and suicides.

1) a

S

S

702 I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 699707Table 2Estimates of the parameters in the selected multilevel logistic regression model (see Eqs. (

Overall

Age group Parameter Estimate 95%-CI

011 Intercept at 2012 11.76 [11.94,11.58]Slope 19982012 0.014 [0.008, +0.035]

1215 Intercept at 2012 13.85 [15.20,12.50]Slope 20112012 2.334 [3.717,0.951]Slope 19982011 0.016 [0.021, +0.054]

1619 Intercept at 2012 10.22 [10.48,9.949]Slope 20102012 0.119 [0.066, +0.305]Slope 20102012 0.164 [+0.052, +0.275]Slope 19982007 0.033 [0.072, +0.005]

2023 Intercept at 2012 10.04 [10.29,9.801]Slope 20102012 0.050 [0.116, +0.215]Slope 20072010 0.250 [+0.144, +0.356]Slope 19982007 0.075 [0.108,0.042]


3039 Intercept at 2012 10.00 [10.17,9.832]Slope 20102012 0.168 [+0.049, +0.288]Slope 20072010 0.093 [+0.019, +0.168]Slope 19982007 0.018 [0.042, +0.006]




75+ Intercept at 2012 14.34 [15.64,13.04]which may be converted to a homicide rate (e.g., the number of homi-cides per 100,000 inhabitants) by the logistic transformation (seeEq. (1)). From Fig. 2, which depicts the evolution after transformingthe original parameters to homicide rates, it can be read that the inter-cept at 2012 of11.76 for the female infanticides corresponds with ahomicide rate of 0.78 per 100,000 girls of between 0 and 11 years old.

It has to be noted that the estimates for girls between 12 and15 years and for women over 75 years turn out to be rather unstable,due to the very low number of homicides in these subpopulations(less than 100 per year across all municipalities), which is reected inthe low values for the intercept parameters, the wide condence inter-vals and large residual variances. Therefore, the results for these agegroups are difcult to interpret and will not be discussed further (notethat the estimated functions for these subpopulations have also beenomitted in Fig. 2).

Interestingly, the evolution of the homicide rate in the extreme agegroups can be described by simpler functions compared to the otherage groups. In particular, the homicide rate for girls up to 12 years oldand for women in the age groups 5059 and 6074 is virtually constant(note that the apparent increase for these age groups shown in the leftpanel in Fig. 2 is nonsignicant, see Table 2). On the other hand, theextreme age groups in men (viz., boys up to 12 years old and men ofover 75) show a somewhat different pattern as compared to women,in that the homicide rate decreases in the rst years of the study periodand increases in the last years (with 2003 being the turning point forboys and 2006 for men over 75).

In the ve age groups for women between 16 and 49, the samemodel was selected as the best t to the data (viz., with twobreakpoints, at 2007 and 2010, respectively), andmoreover, the obtain-ed parameter estimates are remarkably similar: From 1998 to 2007, the

Slope 20112012 4.111 [5.436,2.787]Slope 19982011 0.033 [0.065,0.000]

Notes. For each row in the table, the four parameter estimates for municipality sizes are centereAn asterisk indicates that the estimate for the variance parameter is signicantly larger than 0nd (2)) in each sexage group, applied to the homicide data (females).

ize of municipality

mall Medium Large Big Residual variance

0.02a 0.18a 0.00a 0.19a 0.7430.040a 0.018a 0.021a 0.001a 0.0080.10a 0.20a 0.68a 0.78a 16.8900.258a 0.160a 0.758a 0.660a 18.8240.005a 0.003a 0.011a 0.003a 0.0230.30a 0.01a 0.02a 0.26a 1.332

0.028a 0.079a 0.061a 0.047a 0.3960.120a 0.009a 0.063a 0.065a 0.1530.026a 0.007a 0.016a 0.003a 0.0110.27ab 0.26a 0.25b 0.28b 0.9050.158a 0.059a 0.160a 0.061a 0.1560.094a 0.111a 0.016a 0.033a 0.1470.034a 0.013a 0.013a 0.034a 0.0030.38a 0.09a 0.03a 0.44a 1.1150.180a 0.178a 0.009a 0.010a 0.1880.048ab 0.129a 0.042b 0.135b 0.0880.000a 0.006a 0.001a 0.006a 0.0050.31a 0.01a 0.01a 0.31a 0.6270.041a 0.084a 0.015a 0.028a 0.1490.104ab 0.063a 0.061b 0.106b 0.1030.006a 0.006a 0.005a 0.004a 0.0060.32a 0.06a 0.01a 0.37a 1.1260.333a 0.136a 0.136a 0.060a 0.142

0.051ab 0.148a 0.026b 0.071b 0.1110.013a 0.004a 0.023a 0.006a 0.0090.02a 0.02a 0.19a 0.16a 1.2440.010a 0.005a 0.003a 0.018a 0.0190.06a 0.09a 0.02a 0.01a 0.2440.022a 0.003a 0.007a 0.013a 0.019

0.60a 0.03a 0.05a 0.57a 30.973a a a ahomicide rate decreases, although only in the age group 2023 yearsthe decrease is statistically signicant. Between 2007 and 2010, a strongincrease is observed in all age groups, which in the nal two years di-minishes and becomes nonsignicant, except in the group between3039 years. In the latter group, it seems that the increase intensiesbetween 2010 and 2012 (although the difference between the slopeparameters for 20072010 and 20102012 is nonsignicant). Theevolution in boys between 12 and 15 years of age is virtually identical,with the exception that the rst breakpoint comes at 2006 ratherthan 2007.

The temporal trends in the homicide rates for men in the age groupsabove 15 are shown in the right panel in Fig. 2. The most salient differ-ence with the panel on the left is the change of the scale on the verticalaxis, which illustrates the evidence for the large difference in homiciderates between women and children, on the one hand, and adult men,on the other. However, apart from the absolute differences in homiciderate, the pattern in the temporal evolution is similar to the onedescribed for women in the previous paragraph. For men between 15and 74, a statistically signicant decrease is observed between 1998and 2007, followed by a strong increase which diminishes or levelsoff in the last two years. As follows from Table 3, though, the best-tting model in male age-groups between 16 and 59 years is one withthree (rather than two) breakpoints. In the groups of 1619 and5059 years, the additional breakpoint basically takes account of a turn-ing point in the last year of the study, where the increase in the homi-cide rate actually becomes zero or negative. In the other age groups(20 to 49 years), the additional break point is at 2008 and splits theperiod between 2007 and 2010. Besides, it may seem strange that abreakpoint is added while the slopes that describe the increase in theadjacent periods are quite similar (as in most age groups, the slopes

0.692 0.041 0.001 0.649 31.8110.009a 0.015a 0.005a 0.002a 0.014

d at 0 and estimates sharing a letter in their superscript do not differ signicantly (p b .05).(p b .05).

Table 3Estimates of the parameters in the selected multilevel logistic regression model (see Eqs. (1) and (2)) in each sexage group, applied to the homicide data (males).

Overall Size of municipality

Age group Parameter Estimate 95%-CI Small Medium Large Big Residual variance

011 Intercept at 2012 11.52 [11.71,11.33] 0.11a 0.18a 0.03a 0.26a 1.095Slope 20032012 0.034 [+0.002, +0.066] 0.028a 0.006a 0.0.002a 0.035a 0.026Slope 19982003 0.095 [0.150,0.041] 0.063a 0.011a 0.052a 0.001a 0.041

1215 Intercept at 2012 10.47 [10.76,10.17] 0.45a 0.05a 0.11a 0.39a 2.176Slope 20102012 0.014 [0.173, +0.201] 0.264a 0.210a 0.079a 0.024a 0.507Slope 20062010 0.112 [+0.026, +0.197] 0.037a 0.103a 0.078a 0.062a 0.217Slope 19982006 0.085 [0.122,0.048] 0.042a 0.001a 0.042a 0.000a 0.023

1619 Intercept at 2012 8.45 [8.603,8.306] 0.16ab 0.29a 0.06bc 0.40c 1.385Slope 20112012 0.018 [0.134, +0.170] 0.380a 0.008ab 0.056bc 0.332c 0.323Slope 20102011 0.177 [+0.025, +0.328] 0.301a 0.036a 0.049a 0.216a 0.511Slope 20062010 0.159 [+0.116, +0.202] 0.058ab 0.063a 0.042b 0.079b 0.096Slope 19982006 0.092 [0.108,0.075] 0.019a 0.010a 0.008a 0.001a 0.006

2023 Intercept at 2012 7.85 [7.985,7.716] 0.33a 0.25a 0.18b 0.40b 1.4270.0 a a a a 0.00.10.00.20.00.0

703I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 699707Slope 20102012 0.103 [+0.036, +0.170]Slope 20082010 0.244 [+0.166, +0.321]Slope 20072008 0.234 [+0.097, +0.370]Slope 19982007 0.072 [0.085,0.058]

2429 Intercept at 2012 7.66 [7.786,7.541]Slope 20102012 0.050 [0.004, +0.105]Slope 20082010 0.270 [+0.209, +0.332]for 20072008 and 20082010 are not signicantly different). After all,if the overall slopes do not differ, why does a model with an additionalbreakpoint better t the data?

The answer lies in the geographical differences with respect to theslopes. The last ve columns in Tables 2 and 3 describe deviations

Slope 20072008 0.210 [+0.100, +0.320] 0.1Slope 19982007 0.057 [0.068,0.046] 0.0

3039 Intercept at 2012 7.64 [7.748,7.527] 0.0Slope 20102012 0.061 [+0.016, +0.106] 0.0Slope 20082010 0.207 [+0.157, +0.257] 0.0Slope 20072008 0.335 [+0.246, +0.425] 0.0Slope 19982007 0.051 [0.061,0.040] 0.0

4049 Intercept at 2012 7.83 [7.929,7.721] 0.0Slope 20102012 0.071 [+0.022, +0.119] 0.0Slope 20082010 0.169 [+0.117, +0.221] 0.0Slope 20072008 0.191 [+0.094, +0.288] 0.1Slope 19982007 0.039 [0.049,0.029] 0.0

5059 Intercept at 2012 8.19 [8.312,8.075] 0.1Slope 20112012 0.059 [0.174, +0.055] 0.0Slope 20092011 0.128 [+0.064, +0.192] 0.0Slope 20072009 0.166 [+0.107, +0.225] 0.1Slope 19982007 0.057 [0.069,0.045] 0.0

6074 Intercept at 2012 8.39 [8.498,8.276] 0.1Slope 20092012 0.086 [+0.038, +0.134] 0.0Slope 20072009 0.128 [+0.063, +0.193] 0.0Slope 19982007 0.057 [0.071,0.043] 0.0

75+ Intercept at 2012 8.69 [8.850,8.535] 0.1Slope 20062012 0.050 [+0.013, +0.087] 0.0Slope 19982006 0.074 [0.100,0.048] 0.0

Notes. For each row in the table, the four parameter estimates for municipality sizes are centereAn asterisk indicates that the estimate for the variance parameter is signicantly larger than 0

Year2000 2005 2010

Hom

icid

e rat

e (pe

r 100

,000)

1

2

3

4

5

Females 011

Females 1619

Females 2023Females 2429Females 3039

Females 4049

Females 5059Females 6074

Males 011

Males 1215

Fig. 2.Graphical representation of the overall evolution over time of the homicide rate (per 100the parameter estimates in Tables 2 and 3. The order of the labels on the right follows the sam22 0.017 0.034 0.028 0.13340ab 0.091a 0.108b 0.024ab 0.16750a 0.084a 0.027a 0.261a 0.41626a 0.003a 0.003a 0.020a 0.0045a 0.17a 0.10b 0.33b 1.30812a 0.031a 0.015a 0.034a 0.10547a 0.063a 0.023a 0.086a 0.130

ab a b b from the overall temporal trend described in the previous paragraphsand relates them to the municipalities through, on the one hand, thexed effects parameters associated with the variable Municipality sizeand, on the other hand, the random effects that take account of theresidual variance among municipalities. With respect to Municipality

42 0.136 0.093 0.185 0.39721a 0.005a 0.005a 0.011a 0.005

9ab 0.18a 0.00ab 0.27b 1.17328a 0.001a 0.018a 0.011a 0.09872a 0.042a 0.052a 0.062a 0.11218a 0.098a 0.008a 0.072a 0.33338a 0.009a 0.012c 0.035c 0.0064a 0.08a 0.08a 0.12a 0.91301a 0.014a 0.021a 0.033a 0.07847ab 0.097a 0.038b 0.106b 0.06653a 0.082a 0.005a 0.076a 0.31104a 0.014a 0.004a 0.014a 0.0037a 0.06a 0.13a 0.03a 0.84346a 0.073a 0.080a 0.107a 0.13587ab 0.086c 0.081ac 0.080b 0.09100a 0.019a 0.049a 0.071a 0.09708a 0.005a 0.002a 0.011a 0.0032a 0.10a 0.02a 0.03a 0.65720ab 0.057a 0.063b 0.026ab 0.04599a 0.019a 0.048a 0.166b 0.10323a 0.010a 0.002a 0.015a 0.0032a 0.02a 0.13a 0.27a 0.78548a 0.000a 0.028a 0.076a 0.03524a 0.009a 0.016a 0.030a 0.010

d at 0 and estimates sharing a letter in their superscript do not differ signicantly (p b .05).(p b .05).

Year2000 2005 2010

Hom

icid

e rat

e (pe

r 100

,000)

5101520253035404550

Males 1619

Males 2023

Males 2429Males 3039

Males 4049

Males 5059Males 6074

Males 75+

,000 inhabitants) in different sexage groups based upon the selected statistical model ande order as the homicide rates in 2012.

size, a systematic effect is found on the 2012 intercept parameter for themale homicides in the age groups between 16 and 39 years and forfemale homicides only in the age group 2023 years, which showsthat in highly populated municipalities the homicide rate in 2012 issignicantly larger compared to small municipalities. Additionally, inseveral age groups (females between 24 and 49 years and malesbetween 20 and 29 and between 40 and 49 years) the slope parameter(or one of the slope parameters) associated with the general increase inthe period 20072010 differentiates betweenmunicipalities of differentsizes, with larger municipalities undergoing a stronger increase. Fur-thermore, the signicant residual variance in each of the interceptsand slope parameters is noteworthy, which implies that, within thegroup of municipalities of the same category for Size, signicant dif-ferences remain with respect to the temporal evolution of the homiciderates.

Reconsidering the homicide rates in males between 20 and 49 yearsold, and particularly their evolution between 2007 and 2010, weobserve that the variance among municipalities is much larger for theslope between 2007 and 2008 as compared to the slope for 2008 to2010. An additional analysis showed that the larger variance in theformer slopes is due to the distribution being skewed to the right,which means that, at the onset, the increase between 2007 and 2010is more pronounced in a relatively small number of municipalities,while, a year later the increase has become more widespread andturned out to be more homogeneous among municipalities.

In order to further study the variation among municipalities, theempirical Bayes estimates for intercepts and slopes were obtained foreach municipality in each sexage group. A principal component analy-sis on the 2012-intercepts in the 20 sexage groups further showed that

more than 54% of the variance in these variables was accounted for by asingle factor (with all the variables having a positive loading). A similaranalysis showed that a single factor accounted for 52% of the variance inthe slopes related to the increase comprised between the 20062010period (17 variables in total, e.g., the 20072010 slopes in femalesbetween 16 and 49 years, and the 20072008 and 20082010 slopesin males between 20 and 49). Subsequently, a territorial map ofMexico was constructed where gray levels distinguish the municipali-ties with respect to these factors (see the upper left and upper rightgraph, respectively, in Fig. 3). Not surprisingly, these graphs show verysimilar patterns in that the municipalities where the homicide ratesincreased most strongly between 2007 and 2010 are those with thehighest rates in 2012; they aremostly located in the states of Chihuahua,Durango, Sinaloa, Nuevo Len, Tamaulipas, and Guerrero. Strikingly,in the 5% of the municipalities with the highest rates in 2012 (i.e., thetwo highest categories in the upper left map of Fig. 3), 210 per100,000 males between 20 and 49 were murdered in 2012.

3.2. Suicides

Themain results from the suicides analyses are presented in Table 4,which shows the best-ttingmodel in each sexage group and the esti-mates of the most important parameters in this model. Contrary to thehomicide data, where the best-tting model was rather complex withup to three breakpoints for the piecewise linear function, the modelsfor the suicide data are very simple. In the younger age groups (up to39 years in women and up to 49 years in men), the evolution in thesuicide rates is parsimoniously described by a steady increase over thefull study period (the slope parameter is signicantly larger than 0 in

icids in

704 I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 699707< Perc. 50Perc. 5075Perc. 7590Perc. 9095Perc. 9599> Perc. 99

< Perc. 50Perc. 5075Perc. 7590Perc. 9095Perc. 9599> Perc. 99

Fig. 3. Geographical distribution of the 2012 homicide rate (upper left), the change in homchange in suicide rate between 1998 and 2012 (bottom right) in the 2456 municipalitie

distributions.< Perc. 50Perc. 5075Perc. 7590Perc. 9095Perc. 9599> Perc. 99

< Perc. 50Perc. 5075Perc. 7590Perc. 9095Perc. 9599> Perc. 99

e rate in the 20072010 period (upper right), the 2012 suicide rate (bottom left), and theMexico. The gray levels are based on particular ranges of percentiles of the respective

Table 4Estimates of the parameters in the selected multilevel logistic regression model (see Eqs. (1) and (2)) in each sexage group, applied to the suicide data.

Overall Size of municipality

Age group Parameter Estimate 95%-CI Small Medium Large Big Residual variance

Females1215 Intercept at 2012 10.50 [10.66,10.33] 0.05a 0.00a 0.17a 0.12a 0.359

Slope 19982012 0.073 [+0.050, +0.096] 0.008a 0.010a 0.002a 0.016a 0.0061619 Intercept at 2012 10.23 [10.37,10.09] 0.14a 0.13a 0.03a 0.04a 0.339

Slope 19982012 0.033 [+0.016, +0.051] 0.029a 0.009a 0.007a 0.013a 0.0032023 Intercept at 2012 10.47 [10.63,10.31] 0.23ab 0.16a 0.09bc 0.31c 0.216

Slope 19982012 0.040 [+0.018, +0.061] 0.024a 0.007a 0.009a 0.022a 0.0022429 Intercept at 2012 10.52 [10.66,10.38] 0.15a 0.15a 0.14a 0.14a 0.146

Slope 19982012 0.062 [+0.042, +0.082] 0.030a 0.016a 0.016a 0.003a 0.0013039 Intercept at 2012 11.01 [11.17,10.85] 0.31a 0.06a 0.04ab 0.32b 0.380

Slope 19982012 0.036 [+0.016, +0.055] 0.035a 0.00a 0.012a 0.021a 0.001

tered

705I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 6997074049 Overall intercept 11.26 [11.37,11.15]5059 Overall intercept 11.49 [11.64,11.33]6074 Overall intercept 11.59 [11.74,11.45]75+ Overall intercept 12.33 [13.04,11.61]

Males1215 Intercept at 2012 10.29 [10.42,10.16]

Slope 19982012 0.047 [+0.029, +0.064]1619 Intercept at 2012 9.25 [9.349,9.160]




Slope 19982012 0.032 [+0.025, +0.040]4049 Overall intercept 9.20 [9.274,9.120]

Slope 19982012 0.031 [+0.023, +0.040]5059 Overall intercept 9.53 [9.604,9.461]6074 Overall intercept 9.43 [9.500,9.357]75+ Overall intercept 9.14 [9.239,9.036]

Notes. For each row in the table, the four parameter estimates for municipality sizes are cenall these subpopulations). This evolution is graphically represented inFig. 4. Apart from the gender difference (adult males are on average,about 8 times more likely to commit suicide than adult females), thestrong increase in the youngest age group is striking: Suicide rates ingirls of 12 to 15 years old almost triplicated in the period under study,while for boys in the same age group, the suicide rate nearly duplicated.In none of the other age groups, the relative change has been so large.

In the other age groups, an even simpler model, which only includesan intercept,was found to have the bestt to the data,which implies thatin the older subpopulations (women of 40 years and older, and men of50 and older) the suicide rate is approximately constant. From the esti-mates for the overall intercept in Table 4, the estimated suicide rate canbe calculated for each of these sexage groups: These rates are 1.3, 1.0,0.9, and 0.4 per 100,000 women in the age groups 4049, 5059,6074, and 75+, respectively; and 7.3, 8.0, and 10.7 per 100,000 menin the age groups 5059, 6074, and 75+, respectively. Remarkably,the suicide rates decline over the women's life span including at the

An asterisk indicates that the estimate for the variance parameter is signicantly larger than 0

Year2000 2005 2010

Suic

ide r

ate (

per 1

00,00

0)

1

2

3

4

Females1215

Females 1619

Females 2023Females 2429

Females 3039

Males 1215

Fig. 4.Graphical representation of the overall evolution over time of the suicide rate (per 100,000parameter estimates in Table 4. The order of the labels on the right follows the same order as t0.28a 0.04ab 0.10bc 0.22c 0.2090.27a 0.09a 0.03a 0.33b 0.3390.18ab 0.03a 0.05a 0.26b 0.1290.10a 0.22a 0.09a 0.24a 1.533

0.07a 0.03a 0.01a 0.03a 0.1490.020a 0.000a 0.001a 0.020a b0.0010.15a 0.14a 0.05a 0.04a 0.3940.027a 0.004a 0.008a 0.015a 0.0020.10a 0.01a 0.07a 0.01a 0.3200.014a 0.006a 0.008a 0.013a 0.0020.07a 0.05a 0.00a 0.02a 0.2960.006a 0.007a 0.002a 0.011a 0.0010.23a 0.14b 0.06b 0.04b 0.2680.027a 0.001b 0.005bc 0.021c 0.0010.32a 0.16b 0.14b 0.02b 0.2790.021a 0.004a 0.014a 0.003a 0.0010.09a 0.01a 0.03a 0.06a 0.328

0.07a 0.07a 0.07a 0.07a 0.3330.11ab 0.27a 0.11bc 0.27c 0.607

at 0 and estimates sharing a letter in their superscript do not differ signicantly (p b .05).later stages, while in men the suicide rate started to increase after theage of 50.

The estimates of the parameters associated with Municipality sizeshow that the women suicide rate tends to be higher in the larger mu-nicipalities; in absolute terms, the effect is the largest in the age groupof 20 to 23 years, where the suicide rate in big municipalities is 3.9per 100,000 as compared to 2.3 in small municipalities. The effect ofMunicipality size on suicides in men is less clear, where, at least in theage groups up to 49 years, there is some tendency that the suiciderate in small municipalities up to 10,000 inhabitants is larger than inothermunicipalities. Similar to the homicide analyses, principal compo-nents were retained to describe the variation among municipalitieswith respect to (a) the 2012-intercept in all age groups and (b) theslope in the period 1998 and 2012 in the younger age groups. Themap at the bottom left in Fig. 3 shows that the 2012 suicide rates arethe highest in the north (Baja California, Baja California Sur, Chihuahua,and Sonora) and in the southeast (Campeche, Quintana Roo, and

(p b .05).

Year2000 2005 2010

Suic

ide r

ate (

per 1

00,00

0)

123456789

10111213

Males 1619

Males 2023Males 2429Males 3039Males 4049

inhabitants) in different sexage groups based upon the selected statisticalmodel and thehe suicide rates in 2012.

of the suicide rates, accurately described by a linear function of timeover the full study period, is found in women up to 39 years and men

706 I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 699707up to 49 years of age. Among these, the alarming increase in youngadolescents (both males and females, between 12 and 15 years ofage) warrants immediate attention. On the other hand, our results indi-cate that in the older age groups, the suicide rates have remained rela-tively constant. A study of Shah (2012) comparing the temporaltrends in suicide rates across countries found that, although suiciderates worldwide increased more in older age groups, the pattern wasopposite in developing countries, particularly with high income in-equality and low per capita expenditure in health care. Furthermore,in line with a general difference in gender ratio (see, Schrijvers, Bollen,Tabasco) of the country, while the bottom right map reveals that thedifferences in increase are not particularly linked to certain regions.The latter nding must be seen in relation to, on the one hand, thesmall effect of Municipality size on the slopes (nonsignicant in all butone sexage group) and, on the other hand, the small residual variancefor the slope parameters, which for practical purposes is negligible(even though restricting this variance component to zero results in asignicant deterioration of model t). This means that the increasingtendency of suicides in the younger age groups is a homogeneousphenomenon across the country.

4. Discussion

The avalanche of violence that has characterized Mexico in the lastdecade has been amply discussed by the national as well as the interna-tional scientic community (Krug et al., 2002; Lozano et al., 2012). Inline with other studies (Gawryszewski et al., 2012; Gonzlez-Prez,Vega-Lpez, Cabrera-Pivaral, Vega-Lpez, et al., 2012), we found anoverall gradual decrease in the yearly homicide rates from the start ofour study period in 1998 to 2007, followed by a strong increase leadingto a triplication of the homicide rate in the period 20072012. Ourdifferentiated approach highlighted that this overall evolution is notuniform across all sex and age groups. Indeed, the temporal evolutionturned out to be fundamentally different for the younger and theolder age groups as compared to the middle age groups. The obtainedresults suggest that the apparent general increase of violence in Mexicoprimordially strikes women between 16 and 50 years and men be-tween 16 and 60 years of age, while for young children and womenabove 50, the homicide rate has remained relatively constant over thestudy period. (For males between 12 and 15 years, and above 60, theevolutionary trend is somewhat mixed in that the increase since2006/2007 is smaller than in the other age groups for men). In spite ofthe evolution being similar for women and men in the middle agegroups, homicide rates are considerably higher in men from the age of16 onwards.

Homicide rates and their evolution also strongly differ geographical-ly and are found to be the highest in areas traditionally related to drugtrafc and organized crime (in particular the border with the UnitedStates). Moreover, our results suggest that in urban, densely populatedareas homicide rates are higher than in rural, sparsely populatedareas, which was also a signicant factor in explaining the differencesin homicide rates among Latin-American countries and cities in thestudy by Briceo-Len, Villaveces, and Concha-Eastman (2008). How-ever, as pointed out by the latter authors, the degree of urbanization islikely to be entangled with other factors, including poverty and socialinequality, which may be more strongly related to interpersonal vio-lence. It would be interesting for a follow-up study to relate the homi-cide rates in Mexican municipalities with the well-known Gini indexfor income inequality (Ceriani & Verme, 2012) and other indices sum-marizing the municipalities' level of social welfare.

Our suicide results largely coincide with and extend those obtainedby Hernndez-Bringas and Flores-Arenales (2011). A steady increase& Sabbe, 2012), completed suicide was found to be more likely inmales than in females (while the reverse is true for nonfatal suicideattempts), which is a gap that tends to increase further with age.

This violence epidemic entails adverse effects in a wide range ofthe country's developmental parameters. First, the direct impact onphysical health can be quantied not only in the number of deaths,but also in terms of life years lost, which according to estimates by theInstitute for Health Metrics & Evaluation [IHME] (2013) amounts to1.27 million yearly due to the combination of self-inicted and other-inicted violence. In this respect, Gonzlez-Prez, Vega-Lpez andCabrera-Pivaral (2012) argued that the increase in the rate of homicidalviolence, especially among young people, explains why in Mexico,contrary to a worldwide trend, life expectancy in males fails to showan increase.

Second, Krug et al. (2002) amply discussed the negative impacts of aviolent context not only on physical but also onmental health. In partic-ular, violence often leads to depressions, anxiety disorders, post-traumatic stress disorders, not only in the victims and their families,but also in the perpetrators. In this respect, it is interesting to see asignicant positive correlation (r=.24, p b .01) between the suiciderate in a municipality and its increase in homicide rate between 2007and 2010; this in spite of a negative correlation between homicideand suicide in the Americas found by Bills and Li (2005) prior to theviolence increase in Mexico. Moreover, in many cases these mentalhealth problems translate to a wide range of problematic behaviors,including but not limited to drugs and alcohol abuse. In addition, thelevels of stress experienced by communities in Mexico with higherlevels of violence have been linked to lower academic performanceamong students (Magaloni, 2012). In the same vein, other studieshave estimated the acute effect of exposure to local homicides on thegeneral cognitive development of children suggesting the need for abroader recognition of the negative impact of extreme violent acts onchildren (even regardless of whether the violence is witnessed directly,Sharkey, 2010).

Third, there is a huge economic cost associated with violence. TheWorld Bank (2012) estimated the costs as a consequence of violencebetween 8% and 15% of the gross domestic product (GDP) of Mexico(depending on whether direct, indirect, and/or intangible costs are in-cluded in the assessment). In addition, the cost of violence containmentspent inMexico is currently estimated at around 6.8% of theGDP, that is,around 126 billion US$ annually (Institute for Economics and Peace,2014).

The World Health Organization [WHO] (2012) recently endorsed aset of policy recommendations in order to reduce the prevalence andinverse impact of violence. These recommendations vary accordingto the type of violence and the sector of the population affected(e.g., child abuse, intimate partner violence, youth violence, violenceagainst the elderly, and self-inicted violence) and include (a) thedevelopment of primary intervention programs for young childrenand their caregivers aimed at fostering strong, stable, and stimulatingrelationships (which have been shown to be more cost-effective,Institute of Medicine, 2013), (b) programs for enhancing the cognitive,emotional, interpersonal, and social life-skills in children and adoles-cents, (c) the reduction of the availability and consumption of alcohol,which is considered a risk factor for all types of violent behavior, (d) arestricted accessibility to lethal means, including hand guns, blades,and poisons, and (e) an improved attention and support for the victims.

Given the extent and the growth of the violence problem in Mexicoduring the recent years, it is mandatory to translate the available scien-tic evidence into effective actions, like the suggestions by the WHO(2012). Therefore, Mexico must move beyond political discourse andensure that the nite resources available for violence prevention areused in a scientically sound way. The evidence further indicates that,to this end, a multisectoral approach is most successful, that breaksdown silos and in which the efforts of policy makers, the state healthdepartment, the educational system, the judicial system, and the civil

society members are orchestrated.

References

Bills, C.B., & Li, G. (2005). Correlating homicide and suicide. International Journal ofEpidemiology, 34, 837845, http://dx.doi.org/10.1093/ije/dyi079.

Briceo-Len, R., Villaveces, A., & Concha-Eastman, A. (2008). Understanding the unevendistribution of the incidence of homicide in Latin America. International Journal ofEpidemiology, 37, 751757, http://dx.doi.org/10.1093/ije/dyn153.

Ceriani, L., & Verme, P. (2012). The origins of the Gini index: Extracts from Variabilit eMutabilit (1912) by Corrado Gini. Journal of Economic Inequality, 10, 421443,http://dx.doi.org/10.1007/s10888-011-9188-x.

Crepeau-Hobson, F. (2010). The psychological autopsy and determination of childsuicides: A survey of medical examiners. Archives of Suicide Research, 14, 2434,http://dx.doi.org/10.1080/13811110903479011.

Dahlberg, L.L., & Krug, E.G. (2002). Violence A global public health problem. In E.G.Krug, L.L. Dahlberg, J.A. Mercy, A.B. Zwi, & R. Lozano (Eds.), World report on violenceand health (pp. 121). Geneva, Switzerland: World Health Organization.

Dube, A., Dube, O., & Garca-Ponce, O. (2013). Cross-border spillover: U.S. gun laws andviolence in Mexico. The American Political Science Review, 107, 397417, http://dx.doi.org/10.1017/S0003055413000178.

Gawryszewski, V.P., Sanhueza, A., Martinez-Piedra, R., Escamilla, J.A., & Marinho de Souza,M.F. (2012). Homicdios na regio das Amricas: Magnitude, distribuio etendncias, 19992009 [Homicides in the Americas region: Magnitude, distribution,and trends]. Cincia & Sade Coletiva, 17, 31713182, http://dx.doi.org/10.1590/S1413-81232012001200003.

Gonzlez-Prez, G.J., Vega-Lpez, M.G., & Cabrera-Pivaral, C.E. (2012). Impacto de laviolencia homicida en la esperanza de vidamasculina de Mxico [Impact of homicideon male life expectancy in Mexico]. Revista Panamericana de Salud Pblica, 32,335342, http://dx.doi.org/10.1590/S1020-49892012001100003.

Gonzlez-Prez, G.J., Vega-Lpez, M.G., Cabrera-Pivaral, C.E., Vega-Lpez, A., & Muoz dela Torre, A. (2012). Mortalidad por homicidios en Mxico: tendencias, variacionessocio-geogrcas y factores asociados [Deaths by homicides in Mexico: Trends,

Murray, C.J.L., Vos, T., Lozano, R., Naghavi, M., Flaxman, A.D., Michaud, C., et al. (2012).Disability-adjusted life years (DALYs) for 291 diseases and injuries in 21 regions,19902010: A systematic analysis for the Global Burden of Disease Study 2010. TheLancet, 380, 21972223, http://dx.doi.org/10.1016/S0140-6736(12)61689-4.

Rutherford, A., Zwi, A.B., Grove, N.J., & Butchart, A. (2007a). Violence: A glossary. Journal ofEpidemiology & Community Health, 61, 676680, http://dx.doi.org/10.1136/jech.2005.043711.

Rutherford, A., Zwi, A.B., Grove, N.J., & Butchart, A. (2007b). Violence: A priority for publichealth. Journal of Epidemiology & Community Health, 61, 764770, http://dx.doi.org/10.1136/jech.2006.049072.

Salama, P. (2013). Homicidios, es ineluctable la violencia en Amrica Latina? [Homicides,is violence ineluctable in Latin America?] Frontera Norte, 25, 727.

SAS Institute (2013). SAS/STAT 13.1 user's guide. Cary, NC: Author.Schrijvers, D.L., Bollen, J., & Sabbe, B.G.C. (2012). The gender paradox in suicidal behavior

and its impact on the suicidal process. Journal of Affective Disorders, 138, 1926, http://dx.doi.org/10.1016/j.jad.2011.03.050.

Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics, 6, 461464,http://dx.doi.org/10.1214/aos/1176344136.

Shah, A. (2012). Suicide rates: Age-associated trends and their correlates. Journal of Injuryand Violence Research, 4, 7986, http://dx.doi.org/10.5249/jivr.v4i2.101.

Sharkey, P. (2010). The acute effect of local homicides on children's cognitive perfor-mance. Proceedings of the National Academy of Sciences of the United States ofAmerica, 107, 1173311738, http://dx.doi.org/10.1073/pnas.1000690107.

Snijders, T.A.B., & Bosker, R.J. (2012). Multilevel analysis: An introduction to basic and ad-vanced multilevel modeling (2 ed.). London: Sage.

United Nations Ofce on Drugs and Crime (2014). Global study on homicide 2013: Trends,contexts, data. Vienna, Austria: United Nations publication (Sales No 14.IV.1).

Vrnik, P. (2012). Suicide in the world. International Journal of Environmental Research andPublic Health, 9, 760771, http://dx.doi.org/10.3390/ijerph9030760.

Vilalta, C.J. (2013). Anomia institucional, espacialidad y temporalidad en las muertesasociadas a la lucha contra la delincuencia organizada en Mxico. Mexican Studies/Estudios Mexicanos, 29, 280319, http://dx.doi.org/10.1525/msem.2013.29.1.280.

Vilalta, C.J. (2014). How did things get so bad so quickly? An assessment of the initial con-

707I. Leenen, A. Cervantes-Trejo / Aggression and Violent Behavior 19 (2014) 699707Hernndez-Bringas, H.H., & Flores-Arenales, R. (2011). El suicidio en Mxico [Suicide inMexico]. Papeles de Poblacin, 17, 67101.

Institute for Economics and Peace (2014). The economic cost of violence containment: Acomprehensive assessment of the global cost of violence Retrieved from www.visionofhumanity.org.

Institute for Health Metrics and Evaluation (2013). GBD Compare [Data le].Institute of Medicine (2013). The evidence for violence prevention across the lifespan and

around the world: Workshop summary. Washington, DC: The National AcademiesPress.

Instituto Nacional de Estadstica y Geografa. (n.d.). Consulta interactiva de datos [Interac-tive data query] [Data les].

Krug, E.G., Dahlberg, L.L., Mercy, J.A., Zwi, A.B., & Lozano, R. (Eds.). (2002).World report onviolence and health. Geneva, Switzerland: World Health Organization.

Lozano, R., Naghavi, M., Foreman, K., Lim, S., Shibuya, K., Aboyans, V., et al. (2012). Globaland regional mortality from 235 causes of death for 20 age groups in 1990 and 2010:A systematic analysis for the Global Burden of Disease Study 2010. The Lancet, 380,20952128, http://dx.doi.org/10.1016/S0140-6736(12)61728-0.

Magaloni, B. (2012). The impact of violence on human capital: Findings from schools inMexico. Retrieved from. iis-db.stanford.edu/res/2977/EduPolicyBrief.pdfditions of the war against organized crime in Mexico. European Journal on CriminalPolicy and Research, 20, 137161, http://dx.doi.org/10.1007/s10610-013-9218-2.

Violence Prevention Alliance (2012, May 21). Global campaign for violence prevention: Planof action for 20122020. Geneva: Switzerland.

Wong, G.Y., & Mason, W.M. (1985). The hierarchical logistic regression model for multi-level analysis. Journal of the American Statistical Association, 80, 513524, http://dx.doi.org/10.2307/2288464.

World Bank (2012, June). La violencia juvenil en Mxico [Youth violence in Mexico] (Techni-cal Report).

World Health Assembly (1996, May). Prevention of violence: A public health priority (Res.WHA49.25). Geneva, Switzerland: World Health Organization.

World Health Organization (2011). International statistical classication of diseases andrelated health problems (10th revision), ICD-10 (2010 ed.). Geneva, Switzerland:Author.

World Health Organization (2012). Violence prevention: The evidence. Geneva, Switzerland:Author.

Zucchini, W. (2000). An introduction to model selection. Journal of MathematicalPsychology, 44, 4161, http://dx.doi.org/10.1006/jmps.1999.1276.socio-geographical variations and associated factors]. Cincia & Sade Coletiva, 17,31953208, http://dx.doi.org/10.1590/S1413-81232012001200005.

Temporal and geographic trends in homicide and suicide rates in Mexico, from 1998 through 20121. Introduction2. Method2.1. Data2.2. Statistical analysis

3. Results3.1. Homicides3.2. Suicides

4. DiscussionReferences

temporal and geographic trends in homicide and suicide rates in mexico

Documents