Cole and Andrs Marroqun Gramajo Source: Population and Development Review, Vol. 35, No. 4 (Dec., 2009), pp. 749-776 Published by: Population Council Stable URL: http://www.jstor.org/stable/25593685 . Accessed: 06/01/2014 10:20
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Population Council is collaborating with JSTOR to digitize, preserve and extend access to Population and Development Review.
http://www.jstor.org
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Homicide
Rates
in
Although death?and
violent
death
has been
declared
lem worldwide
a great deal from one examines variation in the country cross-country study a rates for a large sample of countries. We start by providing homicide summary of the data for 2002 (the most recent year forwhich descriptive data of wide coverage are available), the differences stressing particularly to another. This between to which
as fa (2001), we define homicides Following Reza, Mercy, and Krug tal injuries inflicted by another person with intent to injure or kill, by any means. category, "deaths caused Specifically, our study focuses on theW-158 in WHO the tabulation.1 This by intentional injuries (violence)/' (2004b) measure excludes both ofwhich
seven major regions of the world. We then investigate the extent can be explained in terms of this pattern of regional variation socioeconomic variables. underlying
self-inflicted deaths and deaths resulting from civil wars, are treated as separate categories in theWHO cause-of-death tables. (We purposely refrain from using the word "murders" in this study, since the data we use report only the cause of death, and there is no way of classification scheme, what proportion of knowing, from this epidemiological "violent deaths" actually involves criminal intent. It seems reasonable to as sume, on the other hand, that inmost countries most reported violent deaths are in fact crime-related, even though many homicides are not themselves cause The WHO rates tabulation of "Death of death" for 2002 crimes.2) by covers 187 countries, with a mid-2002 of 6.179 billion (i.e., 99.2 population of the estimated world coverage complete total). Thus, this source provides of cause-of-death data around the world.3 essentially
percent
POPULATION
AND
DEVELOPMENT
REVIEW
35(4):
749-776
(DECEMBER
2009)
749
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
750
Homicide
Rates
Around
the World
Homicide
Some
rates around
statistics
theworld
descriptive
Homicide
rates vary, often by a wide margin, across time. As Gartner in developed democracies: 92) writes in her study of homicides Homicide
hundred
(1990:
rates inwestern societies appear to have declined over the last several
years, but more recently, they experienced sharp, short-term upsurges
in the early 19th century and in the last two decades (Gurr 1981). Even over relatively brief periods, the risk of violent death can vary greatly: homicide rates in developed democracies averaged 60 per cent higher in the late 1970s than in the late 1950s. At any point in time the homicide rate also varies greatly across coun tries. Moreover, the distribution of countries by homicide rates is highly skewed: a few countries have very high homicide most while countries rates, relatively low rates. Figure to their homicide 1 shows the frequency distribution of coun rates (per 100,000 population) in 2002.
have
tries according
FIGURE 1 Frequency distribution of 187 countries their homicide rates (per 100,000) in 2002
60-<-1
according
to
50"
Observations
187
Mean
Median Maximum Minimum
9.38
5.7 72.4 0.2
Std.Dev.
10.6
3?- n 8 }->
<v s
20
10-
r-|
50
60
in
70
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
MarroquIn
Gramajo
751
9.38 per 100,000. an similar number: 9.08 per yields essentially 100,000. For comparison, Reza, Mercy, and Krug (2001) found that in 1990 rate was 10.5 per 100,000. As noted above, most the world average homicide have low homicide is boosted rates?half the world
tion of 970.8 million) with rates above 20 per 100,000 (i.e., over twice the rate The homicide for these 25 countries average). (weighted) average was 30.3 per 100,000, over three times the world average.5 world Regional There variation
have under 5.7 per 100,000?but a average upward by relatively few countries with very in the world (52.6 high rates.4 Indeed, in 2002 just over half of all homicides for by the 25 countries percent) were accounted (comprising a total popula
countries
is a very marked degree of regional variation in worldwide homicide rates. This can be seen by comparing averages for countries grouped along and/or socioeconomic lines. For this purpose, we follow the geographic regional groupings used by Reza, Mercy, countries into six categories: 1. EME and Krug (2001), who classified
= established market economies = 2. FSE formerly socialist economies = Middle 3. MEC East crescent 4. OAI = other Asia and islands6 5. SSA = sub-Saharan Africa 6. LAC = Latin America and the Caribbean.
This is a useful classification, although we found it convenient to disaggregate the LAC countries into two distinct categories: ? LA = Spanish- and Portuguese-speaking countries of theWestern and Hemisphere; ? CAR = English-, French-, and Dutch-speaking countries of the Caribbean. The reason for this separation is that these two groups of countries, though 1 lists close, have markedly different homicide rates. Appendix in each of the seven regional groups and their homicide rates.
1 reports both weighted and unweighted (by total population) rates and other summary statistics for each regional group average homicide ing. For most regions the population-weighted averages and the simple, un are similar. The averages weighted closely exception is the formerly socialist economies in which the (FSE) group, average is affected by the weighted Russian Federation, which has both a large population and an exceptionally high homicide rate.
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
752
Homicide
Rates
Around
the World
TABLE
1 Worldwide
Average_ Weighted EME 2.44 18.34 FSE 3.93 MEC 5.65 OAI 21.01 SSA LA 24.98 8.38 CAR
homicide
Unweighted 1.34 7.48 3.81 6.84 16.48 18.42 9.09
World
9.08
9.38
5.7
10.61
72.4
0.2
187
NOTE: For fuller designation of regions, see list in text above. For the full listof countries in each regional cat egory, see Appendix 1.
and
sub-Saharan
African
countries of the
average. As a group, the FSE region is above the world average, though, as noted, this is largely the effect of a single large country. If rate for that region is slightly below Russia is excluded, the average homicide lower than the world average. these results is to compute a regression of Another way to summarize rates on regional dummies homicide (Table 2, Panel A). In this regression, OAI, SSA, LA, and CAR are dummy variables EME, FSE, MEC, (= 1 if the = 0 to the so, country belongs otherwise), region, by construc corresponding tion, the estimated regression coefficients equal the (unweighted) averages are well Africa and Latin America for each region. As noted, sub-Saharan and Mid the established market economies above the world average, while dle East crescent are well below the world average. The FSE, OAI, and CAR the world
in the world (listed in endnote rates are two Homicide in from the these 5) regions. developed countries (the East crescent, on the other hand, are much EME group) and in the Middle
Note
that most
regions are not significantly different from one another.7 variable Given the highly skewed nature of the dependent (homicide are determined by a few ex these results rate), it is useful to check whether treme values repeated variable. The
rates for each region. the logarithm of the geometric average of the homicide on the FSE dummy is 1.496, and elA96 = 4.46, the coefficient (For example, rate for that group of countries, is the geometric average homicide which
cide rate shows a fairly symmetric and roughly bell-shaped distribution.) is analogous results are shown in Table 2, Panel B. The interpretation to the previous regression, although now each estimated coefficient equals
in some of the regional groupings. The regression was therefore rate as the dependent using the natural logarithm of the homicide of for the visual the logarithm of the homi histogram (A inspection
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
MarroquIn
Gramajo
753
TABLE
regional Regional
2 Regressions
dummy dummies
rates in 2002 on
error t-statistic Prob.
variables,
Panel A
(dependent
variable
rate)
1.778
2.039 1.596 1.547
0.754
3.670 2.389 4.422
0.452
0.000 0.018 0.000
1.310 2.039
2.375
12.573 9.031
3.828
0.000 0.000
0.000
Panel B (dependent
variable
is natural
0.150 EME 1.496 FSE
log of homicide
0.170 0.195
rate)
0.882 7.667 0.379 0.000
NOTE: For fuller designation of regions, see list in text above. For the full listof countries in each regional cat egory, see Appendix 1.
Geometric and similarly for each of the other regional dummies.) averages are much less sensitive to extreme values, so any observed differences in this regression are much less likely to reflect the influence of a few outliers in some regions. Mutatis mutandis, the results for Panel B are in fact quali regions have lower tatively similar to those for Panel A: the EME and MEC than average homicide the SSA and LA regions are well above rates, while the world average. The FSE, OAI, and CAR regions are not significantly dif ferent from one another.8 Of course, there effects this way is no numerical
the difference between measuring as in and the averages regional computing regional directly Table 1. The point to note about these regressions, however, is that they "ex plain" about 32 percent of the cross-country variation in homicide rates?and about
variation
50 percent of the variation in the logarithm of the homicide rate?us ing only the regional dummies. One major purpose of this study, then, is to examine the extent towhich this "region effect" can be explained in terms of in underlying socioeconomic variables effect persists after controlling for other, more (i.e., how much of the region fundamental variables).
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
754
Homicide
Rates
Around
the World
Empirical
Literature The social
correlates of violence:
review science literature on factors affecting levels of lethal violence variables have been found to be correlated with homicide is at
on the empirical literature in order to identify the socioeco focus mainly rates in nomic variables that are most commonly associated with homicide finds support in at least some cross-national studies. Each of these variables studies, but there is far from a uniform consensus. structure known 15 and 29, are in the age range between and than females to be engaged in crime in particular (Wolfgang 1968; Hirschi and
Many of these variables are the subject of ongoing debates in of field the (For reviews of the theoretical literature on macro criminology. see Neuman and Berger 1988; Land, McCall, level predictors of homicide 2003: 620-623; and Cohen 1990; Pridemore 2002; Neumayer and Pratt and Cullen 2005). We are not interested in taking sides in these debates, so we
Age
that young males, more likely than other age groups in general, and in violent activities Gottfredson 1983; Wilson
It iswell
1985: 126-147). The victims of ho micide are also predominantly young male adults (Fingerhut and Kleinman 1990; Reza, Mercy, and Krug 2001). Therefore an increase in the proportion in the population is expected, other things equal, to elevate of young males and Herrnstein the homicide
rate. Plausible as this notion may seem, it finds little support studies. Gartner and Parker (1990) and Pampel and Gart in cross-national ner (1995) argue that this effect holds in some countries, such as the United
across all countries, since it States, but that it cannot easily be generalized the presence or absence of certain types of institutional is context-specific: in any arrangements will mitigate or strengthen the age effect on homicides given country. Differences in cultural norms regarding alcohol consumption and firearms possession (see below) are almost surely a factor as well.
Urbanization Urbanization has also been linked with criminality (Wilson and Herrnstein and Loayza 2002: 26), 1985: 411, 430-431, 444-446; Fajnzylber, Lederman, on homicide per a effect strong although it is not clear that itnecessarily has se (with the exception, perhaps, of homicides that are incidental to property
which goes hand in crimes). It is expected that a high level of urbanization, hand with rural-to-urban migration, especially in developing countries, leads to a more intense competition for resources in overcrowded cities, character and urban poverty. Rising criminality within ized by high unemployment the context of the greater anonymity provided by urban settings might result
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
MarroquIn
Gramajo
755
from and Herrnstein put it: "Amigrant from such social changes. As Wilson to commit crime will find the countryside with any preexisting tendency the tendency strengthened when the risk of recognition is slight, and where (1985: 445). These by people he does not know" the active factor that suggest might not be the level our in but rising urbanization. Thus, regression analysis in urbanization as possible
of urbanization, below we consider both the level and the change explanatory variables. Income inequality and poverty
equality, as measured by the Gini coefficient, and violent crime rates (Krohn and Gartrell 1986; Bourguignon 1976; Krahn, Hartnagel, 2001; Fajnzylber, a strong cor not and Loayza others do find Lederman, 2002), although note that the effect of and Herrnstein relation (Neumayer 2005). Wilson "Wealth tends to be accumu inequality may be enhanced by urbanization: lated unequally, and to those not possessing it, itmay seem ... inequitable as well as unequal. The contrast between the haves and have-nots becomes not separated ies" (1985:446). The
Income inequality has received much attention in the literature on crime Some studies find a positive correlation between and homicide. income in
more,
the theory says, as affluence grows, if people are less, palpable, if they live side by side in cit by wider gaps in wealth, especially
that it is not inequality as such (relative deprivation), but poverty (absolute is the that factor. Pridemore deprivation) determining (2008), for instance, notes that most studies based on United States data find a significant rela
connection between inequality and violence has often been ex in terms the relative deprivation of so-called plained theory of homicide, according towhich "aggression is held to be spurred by a sense of frustration 2003: 498). Other studies argue, however, and relative poverty" (MacKellar
studies rates, but that cross-national tionship between poverty and homicide on consider this instead is This variable, rarely focusing inequality. largely because we lack internationally He proposes comparable poverty measures. rates as a proxy for poverty, and finds that once the use of infant mortality is controlled for, the effect of inequality disappears. His study, how poverty
in this ever, is based on a sample of only 46 countries. One of our objectives to a is this for much of countries. study replicate analysis larger sample density
Population
the relationship between population rates Although density and homicide a has never been the main focus of cross-national empirical study, this vari able is often included as a control variable in studies focused on other ex planatory variables (see, for instance, Hansmann and Quigley 1982; Avison
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
756
Homicide
Rates
Around
the World
It has been and Loring 1986; Neumayer 2003). can highlight inequalities and can provide more
lis 1974). Itmay make distinctions between so generate conflict. Itmight also increase contact between individuals and thus increase the likelihood of interpersonal violence" (Krahn, Hartnagel, and Gartrell 1986: 287). This seems plausible, although Neuman and Berger
for crime (Gil opportunities rich and poor more visible and
suggested
(1988: 294) note that "[m]ost theorists expect high density to be associated with high crime rates, but low crime rates are found in both high and low at best. density societies." Thus the effect of population density is ambiguous We nonetheless include this variable in order to see whether on homicides in our sample of countries.
ithas an effect
Ethno-linguistic
and religious
heterogeneity
rate (Hansmann and Quigley 1982; groups tends to increase the homicide Avison and Loring 1986). (Often it is the converse argument that is proposed, societies tend to be less conflict-prone.) namely that relatively homogeneous It has also been
It is frequently argued that a high degree of heterogeneity within a society to conflict and that the interaction of heterogeneous social is conducive
some groups at the expense of others (Easterly and Levine recent cross-country studies of the effects of ethnic diversity 1997). Most fractionalization have been based on the so-called ethno-linguistic (ELF) in policies benefit dex, which was compiled 1980s (see Roeder 2001). randomly selected
suggested that ethnic diversity can bring conflict because of since in a fragmented society it might be particularly political competition, difficult to agree on the amount and kind of public goods the government should provide, and that ethnic differences can lead to conflict when public
groups. One problem with this index is that it conflates two ethno-linguistic of types heterogeneity, linguistic and ethnic, that might not induce conflict are to the same degree. In fact, ithas been suggested that as far as homicides even and have effects concerned Quigley (Hansmann opposite they might et al. (2003) compiled a 1982: 215). To overcome these difficulties, Alesina for a large number of countries along of heterogeneity database ofmeasures three dimensions: to examine ethnic, linguistic, and religious. We use this data source rates the effect of these three types of heterogeneity on homicide in our sample of countries.
by Russian scholars in the 1960s and updated in the This index is interpreted as the probability that two in any given country will belong to different individuals
Education As in the case of population density, education has also been used mainly as can focus of analysis. Education a control variable, rather than as the main in several ways. Low education influence rates of homicide among poorer
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
MarroquIn
Gramajo
757
sectors of society might lead to high unemployment and to poverty-related and crimes and homicides. Fajnzylber, Lederman, Loayza (2002) found that educational attainment, as measured by the Barro and Lee (2001) dataset, has a negative and significant effect on homicides. They also used an index of educational
inequality and found that itdoes not affect homicides, although it an effect on robberies. Pridemore has (2008) used the education component Index as a control variable to explain variation of the Human Development and Gartner rates, and Macmillan (1999) used an education In their meta index to explain spousal violence against women. to 1960 research from 1999, Pratt and analysis of quantitative criminological on crime, and found the effect of education Cullen (2005) also examined in homicide attainment that this effect was Given rather weak. the well-known
rates, it is perhaps gender patterns in homicide never considered the possibility that male surprising that past studies have is to and female education might differ in their effect. One of our objectives examine whether there is a gender effect in the relationship between educa tion and homicide Governance rates.
indicators
institutions should therefore have The quality of a country's government some relevance for crime rates. Over the past several years, a major research an internationally project at the World Bank has been developing compa rable "World Governance 2008
In a well-functioning social system, efficient institutional mechanisms will both help to prevent crimes and prosecute them effectively when they occur.
(which includes effective criminal prosecution and certainty in punishment) might provide a deterrent effect and therefore contribute to lower rates of in general. In addition, an effective court system provides an criminality alternative for conflict resolution, and ithas been suggested that this explains the long-run historical decline of homicide rates in European countries. LaFree and Johnson Monkkonen 193, (2005: citing 1996), writes, "as disputes were increasingly resolved in courts rather than on street cor in bars, violent to violence
Indicator" (see Kaufmann, Kraay, and Mastruzzi for a description of this project and the results achieved so far). Several as the "rule of law," "political stability," components of this indicator?such and "control of corruption"?might be correlated with the level of crime and For example, one would homicide. expect that an effective judicial system
crime rates declined." Countries with weak justice systems, in contrast, would be expected to have higher levels of violence. Other factors often mentioned in the empirical literature are the preva lence of firearms possession and the consumption of alcohol (Killias 1993; 1998; Lester 1995; Parker and Cartmill 1998; Krug, Powell, and Dahlberg Rossow and 2001). Divorce has also been linked to homicides. Williams Flewelling (1988), for instance, found that cities with high divorce rates have
ners and
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
758
Homicide
Rates
Around
the World
significantly higher rates of both family and non-family homicides, while Gartner (1990) found that rates of divorce are correlated with homicide rates, but only for a certain age range of victims. Itwould be difficult to incorporate such factors in this study, however, given the lack of reliable and comparable data for a large enough sample of countries, although we revisit some of these factors when we interpret the empirical results obtained.
Regression
analysis
were
vari Regressions of the logarithm of the homicide rate on the socioeconomic are 3. in Table ables described above The specific regression models reported are numbered 1 to 7. The analysis is based on 91 countries forwhich data 1 (see Appendix 2 for data for all of the variables in Regression list of countries). Thus, because of data limitations our original sample of 187 countries is essentially cut in half. On the other hand, the 91 of 5.2 billion in countries in the reduced sample had a combined population available sources and rates per 100,000 (83 percent of the world's total), and their homicide as a as rates to for the similar world indicated whole, by the fol fairly
2002 were
lowing comparison:
Average_ Weighted Unweighted Standard deviation
Median
Max
Min
8.12 9.08
9.89 9.38
6.2 5.7
12.28 10.61
72.4 72.4
0.5 0.2
The summary statistics for the reduced sample are quite similar to those for the full 187 countries, so we feel confident that a regression analysis based on bias. this cross-section of countries will not suffer from sample-selection Regression 1 includes the following aged = Gini 15-29 explanatory variables: Young men = Males as a percent (2002, of total population or nearest available
(2002).
Income distribution coefficient year), measured est inequality). on a scale from 0 (lowest inequality) to 100 (high under one
= Number of deaths among Infant mortality 1000 live of births, 2000.9 year age per Urbanization tion (2002). level = Urban population
children
as a percent
of total popula
= in the percent of in urbanization Change (1980-2002) Change 1980 and 2002. between urban population = Number of inhabitants per square km (2002). density Population
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
Marroquin
Gramajo
759
a number ranging ELF = Index of ethno-linguistic fractionalization, to 1 from 0 (totally homogeneous population) (totally heterogene ous population), for 1985.10 Male aged = Average years of schooling schooling 15 and over, in 2000.n for themale population
= Average years of schooling schooling lation aged 15 and over, in 2000. = World Governance Indicator, a number to +2.5 (worst possible quality of governance) in 2002.12
for the female popu ranging (best possible from -2.5 govern
in influencing homicide Interestingly, most of the "usual suspects" of young males, poverty (as proxied by the infant mortal rates?proportion not level and/or ity rate), change in urbanization, population density?are main is Gini The the which index, sug statistically significant.13 exception gests
that, in the inequality versus poverty debate, (pace Pridemore 2008) inequality is the driving factor. It should be noted, however, that the effect of
the Gini
take a 10-point increase in the Gini index (say, from 30 to 40, a to increase the homicide rate by about 24 percent. This very large change) is not a large effect. Itwould On the other hand, two variables not often stressed literature, ethno-linguistic are both highly significant and with schooling vari ables are also significant, and the results suggest that the higher the average rate. A seemingly aber the lower the homicide males, years of schooling for fractionalization and the quality the expected sign. The in the empirical of governance,
index, although statistically significant by conventional standards, on a is in fact quite small. Recall that in our data the Gini index ismeasured scale from 0 to 100. Thus, according to the regression estimate, a one-point increase in inequality would increase the homicide rate by about 2.4 percent.
rant finding, however, is the direction of the effect for female schooling: the these data would higher the average years of schooling forwomen, suggest, the higher the rate of homicides per 100,000 inhabitants. the ethno-linguistic frac Regression 2 experiments with disaggregating tionalization measure, using data from the Alesina et al. (2003) database: = Index of ethnic heterogeneity. = Index of Linguistic linguistic heterogeneity. = Index of Religious religious heterogeneity. Ethnic These three heterogeneity indexes are similar to the ELF index: they are as the in any interpreted probability that two randomly selected individuals to different ethnic, linguistic, or religious groups, given country will belong ranging from 0 (totally homogeneous population).
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
index 1.160** 1.261** 1.018** 1.139* 1.403** ELF [1.405] [1.265] [1.353] [1.397] [1.069] [4.209] [3.630] Constant 1.600 1.493 1.590 1.603 1.380 1.464** 2.267**
[-2.340] [-2.597] [-3.037] [-2.248] [-2.580] [-2.589] [-1.118] [-LIU] [-1.158] [-1.060] [-0.076] [0.204] [-0.406] [-0.662] [-0.436] [-0.380] [-0.407] [-0.828] [-0.877] [-0.657] [-0.836] [2.087] [2.446] [1.988] [-0.508] [-0.800] [-0.804] [-0.821] [-0.806] [0.054] [0.126] [2.644]
[2.565]
(1980-2002) [1.398] [1.382] [1.391] [1.364] [1.140] -0.0001 -0.0001 -0.0001 -0.0001 -0.006 -0.002 -0.005 -0.004 -0.0001 [0.547]
Male schooling (years) -0.301* -0.283* -0.311* -0.301* -0.260* -0.294** 0.001 [-0.416]
Regression number
0.029*
0.024*
Independent variable_[1]_[2]_[3]_[4]_[5]_[6]_[7]_
Homicide 3 TABLE 2002: Regression rates, results
heterogeneity Ethnic 1.039* 0.995* 0.032 Linguistic Religious heterogeneity -0.163 0.237 heterogeneity
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
[-2.398] [2.053]
[1.693]
[0.310] [-0.874]
[2.580]
[2.531]
[2.849]
793
** -0.792 -0.742
**
-0.697
**
-0.738
**
-0.711
**
NOTE: All of the regressions were estimated by brackets in Numbers OLS. are t-values of the estimated coefficien
(%) literacy Female *rate 0.023 literacy Male rate (%) -0.036 *
-0.436 SSA 0.182 dummy 0.233 OAI dummy dummy 0.088 FSE MEC dummy
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
762
Homicide
Rates
Around
the World
Three
observations
for El Salvador, Haiti, and Rwanda are quite for the other variables
the explanatory power the proaches margin of 5 percent significance. If the linguistic and religious indexes are dropped index is heterogeneity (Regression 3), the ethnic to ELF its is and coefficient the index that of highly significant, comparable 1. This seems to suggest that ethnic diversity, not linguistic is the the explanatory power of Regres diversity, driving factor. However, sion 3 is somewhat lower than forRegression 1, so ELF must add something that is not reflected in the ethnic variable. This can be seen clearly in Re in Regression gression includes both variables, ELF index apparently the Thus, significant. in the ethnic variable, plus some contained ethnic and linguistic heterogeneity that are 4, which of which captures additional
values for the linguistic variable are missing. Although the coefficients similar to the results for Regression 1, is somewhat lower, and only the ethnic index ap
only the ELF index is all of the information interactions between in the ethnic vari
not reflected
Itmay well be that the ethno-linguistic conflates ethnic and linguistic differences, for the other variables 1. in homicide
fractionalization index, is on the right track after are practically identical to those
1 it adds the dummy variables Regression is the established market ble 2 (the reference point for the regional dummies has the economies which lowest homicide average rate). With regard region, to the non-dummy 1 main the difference between regressors, Regressions and 5 is that income distribution is no longer significant. None of the regional dummies variation is individually significant, which suggests thatmost of the regional 1.However, the addition of the re is in fact reflected in Regression
5 is to see whether
1 adequately Regression rates. To the list of regressors in for six of the regions defined in Ta
6 drops all non-significant regressors, but retains the Latin Regression In variables are statistically sig all of the America this dummy. regression nificant at the 1 percent level of confidence. The regression indicates that fractionalization four basic variables?ethno-linguistic (ELF), average years Index and female), and the World Governance of schooling (both male one almost variable (Latin America) explain regional dummy (WGI)?plus rate in the sam 70 percent of the variation in the logarithm of the homicide
in Regression gional dummies significantly increases the adjusted R-square for 5, indicating that some residual regional variation remains unaccounted on that this indicate 1. The dummies coefficients the regional by Regression effect is likely to come from Latin America.
ple of 91 countries. The following results stand out: a) The Latin America dummy remains statistically significant, with a positive effect on the dependent variable. This means that, other things to rates that are, on have in America homicide Latin tend countries equal,
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
Marroquin
Gramajo
763
average, roughly twice as high as those in similar countries located outside 1980 on the inter and Palmquist this region (e0JU = 2.04) (see Halvorsen in variables of regressions). Recall from semi-logarithmic pretation dummy Table 1 that average homicide rates in Latin America are in fact roughly twice the world average. An increase in the ELF index from 0 to 1 (i.e., the difference be b) and a totally heterogeneous tween a totally homogeneous one) population a rate (el-261 implies roughly 250 percent increase in the average homicide
3.53).
male
things equal, an increase of one year in the average level of an rate by about 29 percent, whereas schooling reduces the homicide increase of one year in the level of female schooling increases the homicide c) Other
rate by about 26 percent. This result, if correct, should be kept in proper is larger, perspective. Note that the negative coefficient on male schooling in absolute value, than the positive coefficient on female schooling. Thus, if increase by about the same amount, then both male and female schooling rate. An increase in homi in the homicide the net effect is a slight decrease cide rates associated with more female education would women's show up if
that a social transition of this denly caught up. It seems intuitively plausible sortmight create significant maladjustments, at least in some cultures. We return to this point in the following section.) We should also mention that
increased at a significantly greater rate than men's schooling schooling. be the case in a country where attainment for educational (This would females traditionally lagged behind male educational attainment and sud
the opposite direction of the effects of male and female education is quite on not measure since direction of effect does the robust, depend particular of education that is used. Indeed, we find the same effect in Regression 7, which uses literacy rates as the measure of education. Data on male and measured female literacy rates are from the World Development Indicators (World Bank), as percentages. Years of schooling is a better measure of educa
tional attainment, since literacy rates only indicate the proportion that has achieved a given minimum level of education. population
of the
In any case, both female literacy rates and female years of schooling are positively and significantly correlated with homicide rates. A in increase the World Governance Index reduces the d) one-point
rate by about 74 percent. The WGI is defined over a range average homicide from -2.5 to +2.5, with countries with "average" governance having a rating of 0. Therefore, other things equal, countries at the extreme low end of this range (worst possible governance) would be expected to have a homicide rate (~2-5) about 6 times higher than a country with "average" governance (r?-738x = a 6.33), while country at the high end of the range (best possible govern rate about 6 times lower than a country with ance) would have a homicide 5= 1/6.33 = 0.16). average governance (r0738x2
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
764
Homicide
Rates
Around
the World
Summary
Relative Most
unimportance
commonly thought to influence country-level homicide rates according to the social science literature (income inequality, proportion and population of young males, urbanization, density) are not statistically our in that studies might This suggests past cross-country significant study. of the variables omitted
have
The present analysis variables. sug important explanatory include ethno-linguistic the quality gests that such variables heterogeneity, of governance, the separate effects ofmale and female education, and some cultural or institutional factors that are related to one main region (Latin
that the interaction of heterogeneous there is a growing consensus Although social groups within countries often tends to promote conflict and associated on how population there is no consensus social pathologies, heterogeneity should be measured. been used
for a large nic, linguistic, and religious diversity drawn from a new database other things, we find that religious diversity sample of countries. Among no on to have homicide effect rates, but ethno-linguistic diversity appears rate. In addition, we find that the original tends to increase the homicide ELF index is a much linguistic measures Importance better predictor of homicide considered separately. rates than the ethnic and
index (ELF) has The ethno-linguistic fractionalization it has been criticized on the grounds for this purpose, although In this that its conflation of ethnic and linguistic diversity is unwarranted. measures to of eth we ELF it index and compared tested the separate study
of institutions
institutions mainly In this study we conceptualize by the degree of good as index. This variable the Bank's World defined by governance governance rates. Quality of was found to be a highly significant predictor of homicide and political law enforcement, stability appear government, in homicide the observed variation factors in understanding the world. results for female Unexpected Possible interpretations At first glance, Nevertheless, this result.14 education: to be critical rates around
result for female education is counterintuitive. the unexpected that can explain it is possible to propose plausible hypotheses
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
Marroquin
Gramajo
765
become more educated, (1) One might expect that as women larger as a consequence, for home devote less of work them leave and, proportions time to early childrearing. This early neglect might have adverse effects on
on homicide offspring later in life. The positive effect of female education rates might also be traced through a different route: higher female educa in the labor market, tion leads to more women participating leaving more to violence at home. Gartner (1990: 101) remarks that children vulnerable appear
the "strongest effects of the shift away from nuclear family-based activities in the labor force to among children. The greater the ratio of women a rates the the of child result also found by households, homicide, greater Fiala and LaFree (1988)." in the job market, one con become more competitive (2) As women and/or lower real sequence might be higher levels of male unemployment
is a significant determinant of homicides in Australia their (see especially on More Hansen ris found that also p. 230). Figure 2, (2006: 1-2) recently, is positively associated with crime in England and ing female employment Wales, and argues that this is because (i) increased female employment lowers wages as a result of the increased supply of labor, (ii) low wages and crime are known to be related, and (iii) "because women tend to have less labour market experience than men, or because they are discriminated against, they tend to enter the job market downward wage pressure to be on the margins of crime." that a higher level of female education delays the (3) It iswell-known average age at which women marry. Accordingly, this also delays the average
at which men marry. Thus, as female education increases, more young
in turn might wages (particularly among less skilled male workers), which lead tomore crime in general. One study that supports this view isKapuscin and Chapman ski, Braithewaite, (1998), who found that female employment
on males
lower down on the earnings distribution, putting in lower skilled jobs who are more likely
age
men
domesticates males and makes them marriage of this domestication, males tend to have more other
single for a longer period of time (and precisely at the ages at are most prone to engage in criminal activities). Hence one would which they expect that, keeping other factors constant, a higher proportion of relatively increases the likelihood of occurrence of illegal activities young single males that might lead to crime and homicides. one could say that Colloquially, remain less violent. In the absence risky lifestyles, and, among are more social pathologies, homicides likely to occur. (Akerlof 1998 a detailed discussion of behavioral differences between married and provides men.) In (4) Some men tend to be intimidated by highly educated women. such cases, an increase in the general level of female education might increase the general level ofmale insecurity in society, which might tend to increase violence. Macmillan and Gartner noting that women who work (1999) propose a slightly different are at less risk of spousal violence
unmarried
male
hypothesis,
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
766
Homicide
Rates
Around
the World
when
to reclaim their status and power and reaffirm their role as theirmasculinity, breadwinner and provider. This was once achieved in the workplace, but is now challenged by the rising number of females entering employment." are not mutually These four hypotheses exclusive, and they might all be part of the explanation for our finding that female education increases the rate. (On the other hand, they might all be wrong as well.) Note homicide must be consistent with the also that any explanation for this phenomenon are stylized facts regarding gender patterns in homicides. Most homicides committed by men, and most victims of homicide also are men, so any tenable explanation must be in terms of the effect that female education has upon violence. Hypotheses
partners are also employed, but that they are at substan risk when the male partner is unemployed. Hansen tially greater (2006: 23) similarly claims that "increasing male crime is an attempt by men to reassert their male
male-on-male
requirement. The first hypothesis, in addition, why a lesser degree ofmaternal investment affect male children more than female children. Homicide There in Latin America
(2), (3), and (4) are consistent with this on the other hand, would have to explain, in childrearing could
rates in this region of the is clearly something special about homicide in world. One element that probably contributes to high levels of violence is the region with the is the "alcohol culture/' Latin America Latin America second highest alcohol consumption Granted, it is per capita (WHO 2004a). even higher in Europe, but good governance probably mitigates the effect of there. high alcohol consumption rates generally tend to increase in coun civil war. Collier and Hoeffler (2004: of experienced periods a to of factors: a large stock of guns this due combination that is 11) argue ofmany in the hands of the civilian population, the sudden demobilization across the violent men into civil society, and "a legacy of scores to be settled There is evidence that homicide tries that have society, and reduced inhibitions about settling them through violence." Hom in some Latin American countries might therefore be partly icidal violence of the experience long-lasting and intense civil wars during by explained the second half of the twentieth Violencia'm
(1981-88), in society accept past has also created a "culture of violence/' where people and approve of violence to defend themselves and to resolve disputes.16 have not improved since 2002. Circumstances Indeed, violence has increased in some Central American countries, with the emergence of large street gangs known as maras (for an analysis of this phenomenon organized see Pinheiro 2007). Nor do our data reflect the recent upsurge in drug-related
(since La example, Colombia century?for Guatemala Nicaragua (1968-96), (1979-92), 1948), El Salvador of civil wars in the recent and Peru (1981-95).15 The prevalence
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
MarroquIn
Gramajo
767
violence war
inMexico,
which
by some accounts
has
reached
(Gonzalez
2009).
Conclusions
Our initial goal was to explain the cross-country, cross-sectional variation rates in a large sample of countries, using data for the most re cent year forwhich suitable data were available. We showed that regional in homicide
dummies became
can explain a large proportion of this variation, and our challenge to include socioeconomic in the variables that have been proposed for the regional effects. We found that socio explain most of the regional effects, with the
exception of the Latin America dummy. One major finding was that countries with high levels of cultural and tend to have higher homicide rates. On the other hand, ethnic heterogeneity countries with a high ranking on theWorld Bank's quality of governance indicator tend to have lower rates. In short, "culture matters," but so do in stitutions.
Education was also found to be significantly associated with homicide this is the first study that separates rates, and to the best of our knowledge as explanatory male and female education variables. Our most novel (and unexpected) finding was that female education cide rate. This finding, on which sociological, psychological hypotheses may appears to increase the homi economic, shed some light, deserves and demographic, further scrutiny.
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
APPENDIX inhabitants
Countries
by regional
group,
with
100,000
(2002) homicides
Mid-year (millions)
group economies
(per population and country Number 100,000) Andorra 0.9 1.5 0.9 1.7 173 Belgium 1.5 Canada Denmark France 1.0 56 0.7 407 Finland 166 3.2 10.3 45631.4 5.4 5.2 59.6 130 11.0 2 39 628 57.2 785127.4 0.3 1 0.1 285 75 8.1
19.7
Germany 0.7 594 82.5 1.2 Greece Iceland Ireland 0.6 1.0 1.1 Italy 0.6 Japan 1.0 0 0.03
3.9
1.7
0.4 8
Monaco
Norway Portugal
1.1 1.4
50
1.0 Spain 1.0 Sweden 0.9 67 7.3 1.1 States 59.3 631 5.4 Subtotal2.44 Kingdom
92
8.9
Formerly
socialist
economies Albania 6.0 184 3.1 1,291 9.9 235 86 210 1.4 2952.3 61 4964.0 68338.2 78821.8 145.3 7.5 5.4 16 7,448 60,736 48.2 331.1 2.0 2.0 7.9 4.4 13.0 Belarus
and Herzegovina
2.0
3.7 76 3.0 1.9 1.3 135 10.2 Estonia 15.4 Hungary 2.4 243 10.2 Latvia12.6 Lithuania 370 10.7 3.0 Moldova 12.3 Poland 1.8 Romania3.6 3.5
Republic
Macedonia
Russian
Federation
Slovenia
15.5 Ukraine Subtotal 18.34 Middle East crescent 4.0 12.0 Algeria 3.7 Armenia
Afghanistan
12.1
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
homicides
(per population and country Number 100,000) 2.9 1.1 Cyprus 0.2 2 175 0.7 79669.0 4.6 2,522 47 6.6 148 5.0 2,921 2.3 3.9 142 5.6 6 0.4 322 29.2 51 2.4 5,330 8 0.7 46917.4 404 1.9 3.4 Turkey 189 2,382 69.6 6.3 14.9 66.0 706 24.4 Egypt1.2 3.8 Georgia Iran 3.8 Israel 2.9 Iraq 0.7 240 7
Jordan 2.9 Kazakhstan 19.7 Kuwait Kyrgyz Republic 8.9 442 5.0 Lebanon 2.6 99 2.5 Libya Malta 1.5 Morocco 1.1 Oman 2.1 Pakistan3.7 Qatar Saudi Arabia 3.0 641 21.6 2.7 Syria Tajikistan Tunisia Turkmenistan United Arab Emirates Uzbekistan Yemen, Rep. 10.2 1.0 35 3.6 3.7 2.1 414 19.3 Subtotal 3.93 Other Asia and islands 7.4 Bangladesh Bhutan 26 4.4 Brunei Darussalam 1.4 5 0.3 Cambodia 17.1 China 3.0 Fiji 0.9 7 0.8 India 5.5 Indonesia 9.4 Kiribati Korea, Korea, South North 1.8 880 47.6 Laos 5.7 Maldives Marshall Mauritius Micronesia Mongolia Islands 1.8 1 0.1 2.6 1.5 3.5 87 2.4 Myanmar16.7 14.8 Nepal Palau Papua New Guinea Philippines 1.0 0 0.02 15.6 885 5.7 21.1 8.7 Malaysia 7.1 6.5 19.5 6 6.4 1.1 1.4 32
144.9
9.8
24,295
10,751 0.6 2,266 38,184 57,390 20,001 0.1 23.3 4,541 309 5.4 2,109 20 31 2 7,799 3,784 1.2 0.1 46.8 25.5 24.3 13.3
144.9
0.3
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Appendix
Regional
(per
and
country Samoa
Number
Seychelles Singapore
3.7
3 0.1 35 Lanka 7.9 1,504 5,795 1 3 3,350 176,692 79.7 3,129.4 19.0 61.7 0.8 Tonga 0.7 Vanuatu 1.3 4.2 Vietnam Subtotal 5.65 0.1 0.2 4.2
Solomon Islands
2.1
9
Sri 9.4 Thailand
0.4
Timor-Leste
17.6
146
Sub-Saharan
Africa Angola39.6 10.0 Benin 13.1 Burkina Faso 5,831 774 1,663 12.7 14.7 7.7
1.8 1,270
1,792 16.6
18.0
7.0 0.5
4.0
10.8 Cameroon
10
23.5 940
Central
African
9.1 0.6
3.4
Congo, Congo,
Djibouti
Equatorial
3.5
13.0
27
59
0.8
0.5
306
14,294 69.6
4.0 1.2
1.5 1,942 21.1 2,030 4,846 4,907 17.7 32.9 8.5 1.5
114 9.3
148
1.9 3.2
17.1
1,697
Mauritania
12.8
348 Namibia
2.7 1.9
27.0 525
4.6
Rwanda
20.0
1,750
1,194 10.9
8.8 0.1
4.9
Somalia
South
33.1
43.2 Africa 30.4 Sudan
2,483
19,538 10,559 45.2 34.8
7.5
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
homicides
Rate (per population and country Number 100,000) 6.0 Tanzania 24.0 10.5 Togo Uganda20.8 Zambia3.7 Zimbabwe 11.3 Subtotal 21.01 66 8,543 6015.7 5,467 26.3 398 10.9 1,449 12.9 148,711
707.7
countries 8.8 Argentina 4.0 346 8.7 Brazil 32.6 Chile5.5 Colombia 72.4 3,299 58,491 43.0 31,133 264 596 11.2 2,915 2,453 4,370 10,159 598 939 5.3 2973.1 5.6 941 26.3 184 3.3 8,889 128,531 25.2 514.6 12.6 6.4 11.8 100.0 37.6 179.2
Bolivia
87015.8 4.1
Costa
Dominican
Ecuador 23.1 Salvador 38.4 37.1 Guatemala Honduras 13.4 Mexico 10.2 Nicaragua Paraguay 11.4 Panama 9.7 16.9 Peru 3.6 Uruguay5.6 Venezuela35.2 Subtotal 24.98 English-, Antigua
8666.4
of the Caribbean
and Barbuda
French-,
and
Dutch-speaking 10.2
66 25
Dominica Grenada
9.9 Guyana Haiti 10.2 Jamaica St. Kitts St. Lucia St. Vincent Trinidad and Grenadines 8.8 13.0 15 0.1 Suriname and Nevis 11.9 0.056 7.6 4.2 0.5
and Tobago
World
total
(187 countries)
9.08561,265
SOURCES: Homicide rates (2002): World Health Organization (2004b), Table 3;Mid-year population (2002): World Bank, World Development Indicators (online version). The number of homicides in 2002 is calculated by ap plying the reported homicide rates for2002 to the reported mid-year population for 2002.
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Homicide
Rates
Around
the World
from data Proportion of males (15-29): Computed "Table 094. Midyear Population, by Age and Sex, 2002" Gini Population index of income distribution, density, Literacy rates: World fractionalization: Infant mortality, Roeder
in Census
Bureau,
International
Data
Base,
and Female
org/governance/wgi/index.asp) Countries included in the regression Finland Algeria France Argentina Gambia Austria Bangladesh Germany Ghana Belgium Greece Benin BoliviaGuatemala Botswana Guinea-Bissau Brazil Guyana Haiti Bulgaria Cameroon Honduras Central African Republic Hungary India Chile China Indonesia Iran Colombia Costa Czech Republic Republic Rica Ireland Israel Croatia Italy Jamaica Denmark Japan Jordan Ecuador Kenya Egypt Korea, South Papua
analysis: Lesotho Senegal Malawi Malaysia Mali Slovakia Sierra Singapore Leone
Mexico Slovenia Mozambique NepalSpain Netherlands New Zealand Sweden South Africa Sri Lanka Swaziland
Nicaragua
Switzerland Niger Thailand Norway Pakistan Panama Tunisia New Guinea Turkey Kingdom States ParaguayUganda Peru United United Philippines PolandUruguay Venezuela Portugal Romania Zambia Russian Rwanda Federation Zimbabwe Trinidad and Tobago
Dominican
El Salvador
Notes
The authors for help timated cause, W158, Dec. thank Olufunmilayo in data processing. basic dataset deaths 100,000 State, injuries Odushola self-defense, as well forces in the course 3, "Es by line as persons killed by police of law enforcement.
1 The
is from Table
is therefore not an 3 Although coverage is quite another matter, issue here, accuracy are often expressed about the and concerns of under-reporting degree is especially This statistics. in cause-of-death in problematic since the degree to from country is an extreme
2004
justifiable
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
Marroquin
Gramajo
773
situation than
and
could well
be better
other
causes
of death,
to think that registration in countries with well-developed plete even if it is statistical systems. On the other hand, to assume reasonable that the quality of offi cial statistics level improves with a country's overall then it of socioeconomic development, is somewhat encouraging (though only in this
8 Again, the equality of the coefficients and CAR was for FSE, OAI, tested using a Wald was test. The hypothesis not rejected. source for this variable (WorldDevel 9 The of equal coefficients
opment Indicators, online version) reports many for 2002. Country coverage for values missing in contrast, is fairly complete. 2000, 10 The 2001) source for this for 1960 male for and index and reports values 11 The ing (Barro source and Lee (Roeder 1985 only. school
countries do sense) to find that less developed not generally show very low levels of reported homicide countries. conventional our conclusions, quality rates, which were so, we Even would be if under-reporting caveat the norm expected in such
female
are obliged to state the the validity of regarding are only as good as the which data. distribution of homi
at 2001) reports values intervals, up to 2000. Figures on male for 2000 were derived from data on and female schooling.
of the underlying
12 The
quality, and since there is nothing special about the nation-state as a geographic unit, it is not surprising to find of homi that the same skewed distribution in any up within countries as well: some regions are much more given country than others. This same pat homicide-prone cides shows tern recurs to the city level?see, right down on the New for instance, the interactive map York Times website ?http://projects.nytimes. com/crime/homicides/map?ref=nyregion?. the 25 most homicide-prone rates per 100,000 their homicide (and were: Colombia (72.4), Sierra Le population) one (50.3), South Africa (43.2), Angola (39.6), El Salvador (38.4), Guatemala (37.1), Venezu countries ela (35.3), Somalia (33.1), Russia (32.9), Liberia (32.8), Brazil (32.6), Sudan (30.4), Ivory Coast (27.4), Namibia (27.0), Tanzania (24.0), Guinea 5 In 2002
actually dif indicators, measuring of quality of governance: effectiveness, rule of law, regulatory voice and
of corruption,
political of violence/terrorism.
stability and ab For our purposes, and calculated each country's as the simple average of the
"total WGI
six component
indicators.
13 We with the experimented using logarithm of population density, but this did not change also ex any of the results. We to with different time horizons perimented measure in urbanization the change (e.g., 10 but none year, 30-year, and 40-year changes), was of these alternative measures significant es either, and the resulting regressions were sentially the same. 14 Several gested Cole, Ghersi. 15 Note that three of these countries were were of these hypotheses sug by our colleagues, including Joseph Marco Antonio del Rio, and Enrique
(23.8), Central African Republic (23.5), Ecua dor (23.1), Nigeria (23.1), Congo (Dem.) (21.3), Bahamas (21.2), Philippines (21.1), Uganda and Rwanda (20.8), Ethiopia (20.5), (20.0). these countries account for over half Although of all homicides in the world, they contain only 16 percent of the world's population. 6 This demographic almost merit their own India regional group China includes the two and India, which as separate regions in a study of homicides in (2000).
the top six countries with the highest among rates in 2002, and Colombia homicide had by far the world's highest rate. (Colombia, where civil conflict began American countries, earlier was than in other Latin the world's already top homicide-rate country as far back as 1960, with 34 per 100,000; Wolfgang 1968: 490.) 16 There might be some path-dependent in homicide rates in certain highly trajectories violent This might be due to "ha countries. or "desensitization": bituation" societies might to high homicide accustomed simply become rates (see Bandura 1973).
giants, treatment
for FSE, equality of the coefficients CAR was test. tested using a Wald of equal coefficients was not
hypothesis
rejected.
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
774
Homicide
Rates
Around
the World
References
Akerlof, Alesina, George Alberto, A. 1998. "Men without Devleeschauwer, children," William Economic Journal 108: 287-309. and Roman Wac Sergio Kurlat, Easterly, Journal ofEconomic Growth 8: 155-194. "Fractionalization," ziarg. 2003. R. and Pamela L. Loring. 1986. "Population ho Avison, William diversity and cross-national micide: Bandura, Barro, Robert Updates The Albert. and heterogeneity," effects of inequality 1973. Aggression: A Social Learning Analysis. J. and Jong-Wha Lee. 2001. "International Criminology Englewood data on 24: 733-749. Cliffs, NJ: Prentice-Hall. educational attainment: Arnaud
implications," Oxford Economic Papers 53: 541-563. F. 2001. "Crime as a social cost of poverty and inequality: A review focusing on Bourguignon, in Shahid Yusuf, Simon Evenett, Wu and Weiping countries," (eds.), Facets developing International and Local Dimensions Bank Discussion of Globalization: of Development, World and Collier, Paper No. 415. Washington, Paul and Anke Hoeffler. of homicide Series, Dreze, Easterly, Jean micide sions," No. and civil war," DC: World 2004. Center "Murder pp. 171-191. Socio-economic by numbers: for the Study of African Economies Bank, in India: Policies determinants Working Insights and Paper from ho divi
210. Department of Economics, Oxford University. 2000. and Reetika Khera. "Crime, gender, and society and Development Review 26: 335-352. data," Population and Ross Levine. 1997. "Africa's 112: Quarterly Journal ofEconomics and Norman Pablo, Daniel Lederman, of Law and Economics 45: 1-40. "Cross-national determinants growth 1203-1250. Loayza. tragedy:
William
ethnic
2002.
"Inequality
and violent
crime," American
1990. "International of homi and interstate comparisons 263: 3292-3295. Journal of theAmerican Medical Association young males," A temporal and cross-national 1990. "The victims of homicide: Gartner, Rosemary. compari son," American Sociological Review 55: 92-106. cide among on homicide and 1990. "Cross-national evidence and Robert Nash Parker. Rosemary Social Forces 69: 351-371. the age structure of the population," Social Forces 53: 306-314. 1974. "Population Gillis, A. Ronald. density and social pathology," E. 2009. "Mexico's Francisco Gonzalez, drug wars get brutal," Current History 108: 72-76. Gartner, Gurr, Ted Robert. Halvorsen, 1981. "Historical trends in violent crime: A critical review of the evidence," Crime and Justice 3: 295-353. in 1980. "The interpretation of dummy variables Robert and Raymond Palmquist. American Economic Review 70: 474-475. equations," semilogarithmic for Longitudinal Center 2006. "Male crime and rising female employment," Kirstine. Hansen, London. At Studies <<http://www.learningobservatory.com/resource/ (unpublished). male-crime-and-rising-female-employment? Henry B. and John M. Quigley. Gottfredson. (accessed 1982. "Population 3 July 2009). heterogeneity and the socio
1988. and Gary LaFree. Sociological Review 53: 432-445. L. A. and J. C. Kleinman. Fingerhut,
of child homicide,"
Hansmann,
Social Forces 61: 206-224. 1983. "Age and the explanation of crime," American
1996. The Civilization of Crime: Violence in Town (eds.). of Illinois Press. Middle Ages. Urbana: and Country since the University and 1998. "Unemployment and Bruce Chapman. Cezary A., John Braithwaite, Kapuscinski, crime: Toward Kaufmann, gregate Working Killias, Martin. and Daniel, and 14: 215-243. Journal of Quantitative Criminology the paradox," resolving matters VII: Ag 2008. "Governance Mastruzzi. Aart Kraay, and Massimo Bank Policy Research World individual 1996-2007," indicators, governance gun ownership 148: 1721-1725. and rates of homicide
DC. Paper No. 4654. Washington, between 1993. "International correlations Canadian Medical Association Journal
suicide,"
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Julio
H.
Cole
/ Andres
Gramajo
775
Krahn,
and data
John W. and
Gartrell.
1986.
"Income
criminological
theories,"
Criminology
inequality 24:
D. 1976. "Inequality, and crime: A cross-national Marvin analysis," unemployment The Sociological Quarterly 17: 303-313. 1998. "Firearm-related in the United and L. L. Dahlberg. States deaths Krug, E., K. E. Powell, International Journal ofEpide and 35 other high- and upper-middle-income countries," Krohn, miology LaFree, Land, Gary. trends, Kenneth Journal Lester, David. rates: A 27: 214-221. 2005. 1956 C, "Evidence to 2000," Patricia for elite convergence Quarterly Lawrence in cross-national 46: across 191-211. 1990. "Structural social space?," covariates American time and E. Cohen. homicide victimization The Sociological L. McCall, and there any
of homicides
rates: Are
invariances
and suicide and homicide between alcohol consumption 30: 465-468. Alcohol and Alcoholism study of 13 nations," F. Landis. and suicide," in Paul Demeny and Geoffrey McNicoll 2003. "Homicide MacKellar, 1. New vol. York: Thomson Gale, pp. 496-499. (eds.), Encyclopedia of Population, 1999. "When Ross and Rosemary the bacon: Labor-force Gartner. she brings home Macmillan, and participation 61:947-958. Family Neuman, Neumayer, -. W. Lawrence Eric. 2003. crime: An risk of spousal and Ronald violence against women," Journal of Marriage and the
evaluation
on cross-national J. Berger. 1988. "Competing perspectives of theory and evidence," The Sociological Quarterly 29: 281-313. crime: Evidence "Good policy can lower violent from a cross-national theft,"
Journal of Peace Research 40: 619-640. rates, 1980-97," panel of homicide 2005. and violent crime: Evidence from data on robbery and violent "Inequality Journal of Peace Research 42: 101-112. Gartner. Fred C. and Rosemary Pampel, national homicide rates," European S. Cartmill. Parker, Robert N. and Randi 1934-1995," Pinheiro, Paulo Pratt, Travis Pridemore, Journal 1995.
and institutions, "Age structure, socio-political Sociological Review 11: 243-260. 1998. "Alcohol and homicide in the United States,
of Criminal Law and Criminology 88: 1369-1398. 2007. and democracy," Current History 106: 64-69. "Youth, violence, Sergio. C. and Francis T. Cullen. 2005. and theories of "Assessing macro-level predictors structure and homicide: A review
crime: A meta-analysis," Crime and Justice 32: 373-450. "What we know about social William Alex. 2002. and
of the theoretical -.
literature," Violence and Victims 17: 127-156. empirical to the cross-national 2008. "A methodological addition literature on social empirical structure A first test of the poverty-homicide and homicide: thesis," Criminology 46: 133-154. 2001. "Epidemiology of violent deaths in the world," Injury
Reza,
fractionalization Philip G. 2001. "Ethnolinguistic (ELF) indices, 1961 and 1985," Febru 16. At ?http//:weber.ucsd.edu\~proeder\elf.htm? on 10 August (accessed 2008). 2001. A cross-cultural "Alcohol and homicide: of the relation Ingeborg. comparison matrix estimator and a direct test
Addiction 96: 77-92. countries," ship in 14 European H. 1980. "A heteroskedasticity-consistent covariance for heteroskedasticity," 48: 817-838. Econometrica Kirk R. and Robert A comparative study Review 53: 421-431. James Q. & Schuster. and Richard E. 1968. L. Flewelling. of disaggregated J. Herrnstein. 1988. "The rates 1985.
Williams,
Wilson,
York:
"Crime:
Encyclopedia
York: Macmillan,
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions
Homicide
Rates
Around
the World
Organization.
2004. At
-.
int/substance_abuse/publications/global_status_report_2004_overview.pdf? on 9 February 2009). 2004b. (Dec xls? (accessed Estimates Member States of Disease forWHO and Burden Mortality At 2004). <<http://www.who.int/healthinfo/statistics/bodgbddeathdalyestimates. on 15 August 2008).
This content downloaded from 200.0.176.5 on Mon, 6 Jan 2014 10:20:28 AM All use subject to JSTOR Terms and Conditions