Global Crime

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/fglc20

The crime decline in cross-national context: a

panel analysis of homicide rates within latent
trajectory groups

James Tuttle, Patricia McCall & Kenneth Land

To cite this article: James Tuttle, Patricia McCall & Kenneth Land (2021): The crime decline in
cross-national context: a panel analysis of homicide rates within latent trajectory groups, Global
Crime, DOI: 10.1080/17440572.2021.1920931

To link to this article: https://doi.org/10.1080/17440572.2021.1920931

Published online: 04 May 2021.

Submit your article to this journal

Article views: 103

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=fglc20
GLOBAL CRIME
https://doi.org/10.1080/17440572.2021.1920931

ARTICLE

The crime decline in cross-national context: a panel analysis

of homicide rates within latent trajectory groups
a
James Tuttle , Patricia McCallb and Kenneth Landc
a
Department of Sociology, University of Montana, Social Sciences, Missoula, United States; bDepartment of
Sociology and Anthropology, North Carolina State University, Raleigh, United States; cDepartment of
Sociology, Duke University, Durham, United States

ABSTRACT KEYWORDS
During the 1990s, the United States and other wealthy democracies crime decline; homicide;
experienced a decline in homicide rates. However, not all nations cross-national; latent
shared this trend. Despite disparate homicide patterns, researchers trajectories; crime trends
usually examine the average effect of correlates on homicide,
potentially obscuring the impact of heterogeneity within large
samples. The current study addresses this implicit homogeneity
assumption by identifying three distinct latent trajectory groups
of homicide trends among 77 nations from 1989 to 2010. To exam­
ine differences in the correlates of homicide trends, we analyse the
impact of demographic and economic influences on homicide rates
in separate fixed-effects panel regression analyses for each trajec­
tory group as well as for the overall sample. We find that demo­
graphic and economic forces impact homicide rates differently
across subsets of nations. Our findings suggest that universal expla­
nations of 1990s cross-national homicide trends are misleading, as
the same set of factors influence homicide rates differently across
national contexts.

Introduction
Most research examining 1990s cross-national crime trends has highlighted the simila­
rities in homicide patterns between the United States and other wealthy democracies.
During the 1990s, the homicide rate fell by nearly 50% in the United States (Zimring,
2006). This dramatic trend garnered considerable attention (e.g., Blumstein & Wallman,
2006; Levitt, 2004) and it soon became apparent that the seemingly ‘All-American’ crime
decline was mirrored by similar trends in other wealthy democracies, such as in Australia,
Canada, Japan and several nations in Western Europe (Eisner, 2008; Farrell, 2013; Zimring,
2006). While the 1990s crime decline in the United States was more dramatic than crime
trends in other nations, homicide rates appear to be converging among wealthy countries
(LaFree, 2005). The 1990s homicide decline among wealthy nations has been attributed to
shared trends in demographics, economics, and criminal justice practices (Baumer &
Wolff, 2014; Buonanno, Drago, Galbiati, Zanella, 2011; Farrell, Tseloni, Mailley, & Tilley,
2011; Rosenfeld, 2014; Rosenfeld & Messner, 2009).

CONTACT James Tuttle [email protected] University of Montana, Social Sciences 259B Missoula,MT
59812
© 2021 Informa UK Limited, trading as Taylor & Francis Group
2 J. TUTTLE ET AL.

However, neither 1990s homicide trends nor the correlates of homicide rates are uni­
versal. Some nations, most notably in Latin America, have suffered increases in homicide
rates during recent decades (Baumer & Wolff, 2014; Rogers & Pridemore, 2017; LaFree, Curtis,
McDowall, 2015; Tuttle, McCall, Land, 2018). This divergent homicide pattern occurred
despite similar economic and demographic trends occurring within nations that did and
did not experience a decline in homicide during the 1990s. In an analysis of Latin America,
Bergman (2018) links improving economic conditions to higher rates of crime due to an
increase in attractive targets for theft and a growing underground economy. This conclusion
is the opposite of research findings in wealthy nations (e.g., Rosenfeld & Messner, 2009) and
larger cross-national samples (Baumer & Wolff, 2014), which attribute the crime decline to an
improving economy. Additionally, recent research suggests that certain demographic fac­
tors, such as age structure, impact homicide rates differently across world regions (Santos,
Testa, Porter, Lynch, 2019). Overall, 1990s homicide trends diverged dramatically, possibly
reflecting different correlates of homicide trends across subsets of nations.
The typical methodological convention within cross-national homicide research is to
group all nations together within a single sample to estimate the average impact of
different factors on homicide. This methodological convention implicitly assumes that the
same demographic and economic trends are expected to produce similar effects on
homicide rates across all nations. While the technique of grouping nations together and
examining the overall average effect of covariates has value to understanding cross-
national variation in homicide, this common practice can also obscure the impact of
relevant factors within subsets of nations. While research on small homogeneous cross-
national samples does not present this problem (e.g., Buonanno, Drago, Galbiati, Zanella,
2011; Farrell et al., 2011; Rosenfeld, 2014; Rosenfeld & Messner, 2009), analyses on
homicide trends in larger samples (e.g., Baumer & Wolff, 2014) potentially contributes
to misleading conclusions. Given the divergent patterns of homicide and dissimilar
correlates of homicide across subsets of nations, the possibility remains that the correlates
of 1990s cross-national homicide trends are not invariant.
This paper addresses hidden heterogeneity within cross-national homicide trends and
its impact on the temporal correlates of homicide during the 1990s. To examine the
possibility that the same structural correlates of crime could produce disparate impacts,
we apply an empirically-driven approach by identifying disparate patterns of homicide
using latent class-trajectory analysis (see Tuttle, McCall, Land, 2018; Weiss, Santos, Testa,
Kumar, 2016). After identifying three different patterns of homicide victimisation from
1989 through 2010 using latent-class trajectory analysis, we conduct four separate panel
regression analyses (the overall sample and each trajectory group) examining the impact
of several commonly-assessed demographic and economic correlates of cross-national
homicide rates. These analyses reveal different predictors of homicide trends across latent
trajectory groups. The implications of these differences across trajectory groups are
examined below. But first, we contextualise the current study by briefly examining
previous research on 1990s homicide trends.

Homicide trends during the 1990s

The rapid decrease of aggregate crime rates in the United States during the 1990s
garnered considerable attention, with a variety of explanations proffered initially focusing
 GLOBAL CRIME 3

on the American context (e.g., Blumstein & Rosenfeld, 1998; Cook & Laub, 1998; Zimring,
2006). While some of these explanations may have more merit than others (see Baumer,
2008; Levitt, 2004), one crucial test of different perspectives is their cross-national validity
(Farrell, 2013). As scholars began to recognise that other wealthy democracies experi­
enced a similar decline in crime and homicide during the 1990s (e.g., Baumer & Wolff,
2014; Eisner, 2008; LaFree, Curtis, McDowall, 2015; Messner, Pearson-Nelson, Raffalovich,
Miner, 2011; Van Dijk, Tseloni, Farrell, 2012), the limitations of the United States-specific
nature of many of these homicide-decline explanations became apparent. Crucially,
several nations experienced a homicide decline while not sharing societal trends, such
as the waning crack-cocaine epidemic, that some scholars attributed as the cause of the
American homicide decline (e.g., Blumstein & Rosenfeld, 1998; Cook & Laub, 1998).
Among the consistencies across wealthy Western-style nations, there were several
shared economic, demographic, and criminal justice trends that contributed to lower
rates of crime. For example, improving economic conditions in the United States (McCall,
Parker, McDonald, 2008) and in other countries (e.g., Rosenfeld, 2014; Rosenfeld &
Messner, 2009) have been linked to declining crime rates during the 1990s. These
economic improvements contributing to lower rates of crime and homicide include
reductions in poverty (Baumer & Wolff, 2014; Tuttle, 2020), reduced inflation (Rosenfeld,
2014; Tuttle, 2020), and improving consumer sentiment (Rosenfeld & Messner, 2009).
Additionally, an ageing population has been linked to crime declines in both the United
States (Fox, 2006; Fox & Piquero, 2003) and in cross-national research (Baumer & Wolff,
2014; Santos, Testa, Porter, Lynch, 2019). Finally, there is conflicting evidence concerning
the rate of imprisonment and crime rates. Some research shows a significant negative
association between imprisonment and crime in smaller, more homogeneous samples
consisting primarily of wealthy democracies (Buonanno, Drago, Galbiati, Zanella, 2011;
Rosenfeld & Messner, 2009; Tuttle, 2020), but other studies using more heterogenous
samples fail to document this finding (Baumer & Wolff, 2014).
However, there is evidence to suggest that instead of a single cross-national homicide
trend, there are multiple 1990s homicide trends. In an examination of an overall suprana­
tional homicide trend, Rogers and Pridemore (2018) find there is a relatively weak global
trend from 1979 to 2013. This relatively weak global trend implies considerable variability
in both regional and national homicide trends during this era. In large, diverse samples,
researchers have documented dissimilar regional homicide patterns during the 1990s
(Baumer & Wolff, 2014; Rogers & Pridemore, 2017; Tuttle, McCall, Land, 2018). For exam­
ple, in Eastern Europe there was a decline in homicide during the 1990s, but this decline
was somewhat later than the decline than in Western Europe (LaFree, Curtis, McDowall,
2015). The most divergent trends have been noted in Central and South America, where
some nations have actually experienced an increase in homicide rates during recent years
(Bergman, 2018; LaFree, Curtis, McDowall, 2015; Peres & Nivette, 2017).
Additionally, these divergent trends were associated with a different set of factors than
the homicide decline in the United States and Western Europe. Notably, homicide rates in
Eastern Europe (Stamatel, 2009) and Latin America (Santos, Testa, Porter, Lynch, 2019)
were negatively associated with the percentage of the population aged 15 to 24 during
the 1990s. This finding is inconsistent with other research in larger samples suggesting
that an ageing population contributed to the homicide decline (Baumer & Wolff, 2014),
although it should be noted that inconsistent findings concerning the relationship
4 J. TUTTLE ET AL.

between age structure and homicide within the broader homicide literature are common
(Rogers, 2014; Rogers & Pridemore, 2016). Additionally, some have attributed increases in
Latin American crime rates to improving economic conditions (Bergman, 2018), which is
one of the purported causes of the decrease in crime among wealthy Western nations
(Baumer & Wolff, 2014; Buonanno, Drago, Galbiati, Zanella, 2011; Rosenfeld, 2014;
Rosenfeld & Messner, 2009). These discrepancies raise questions about the tenability of
universalistic explanations of 1990s homicide trends.
Previous research examining different homicide trajectories suggest that stable struc­
tural characteristics differ across subsets of nations with differing homicide trends. Tuttle
et al. (2018) identify distinct latent class trajectory homicide patterns among 82 nations
from 1980 to 2010. After highlighting three trajectory groups, the authors perform binary
and ordered logistic regression analyses to determine which factors predict the selection
of countries into each of the three groups. Overall, they find that nations located in Latin
America, with less economic development, a larger population size, and more divorced
males are more likely to select into the highest homicide rate group, which exhibited an
increase in homicide rates during the 1980s and no significant decline in homicide during
the 1990s. However, these analyses only indicate stable differences between subsets of
nations with distinctive homicide patterns and not what is associated with temporal
variation in homicide within these nations over time. It is yet unclear if the temporal
correlates of homicide differ across subsets of nations during the 1990s.

Theoretical considerations of homicide across subsets of nations

While the literature on 1990s cross-national homicide trends has not always prioritised
theory testing, there are theoretical perspectives and grand metanarratives that inform
homicide trend research. For example, the ‘modernisation perspective,’ originating with
the work of Durkheim (1984), suggests that the homicide rate increases during periods of
rapid societal transition in which older values, norms, and practices become obsolete and,
therefore, are no longer able to regulate social behaviour. In previous research, rapid
societal transition has been operationalised as rapid industrialisation and urbanisation,
hypothesised to break down community social bonds (Shelley, 1981). This would imply
that the correlates of homicide rates would differ across subsets of nations with differing
levels of economic development. In subsets of less economically-developed/urbanised
nations, homicide rates would be influenced deleteriously by increasing urbanisation
whereas economically-developed/urbanised nations would not necessarily be impacted
by greater rates of urbanisation having already endured transitions in industrialisation
and urbanisation.
Additionally, the ‘conflict perspective,’ inspired by Marx’s (1977) economic theory,
predicts increases in homicide over time within capitalist nations. Conflict theorists
posit an increase in homicide due to the negative consequences wrought by capitalist
economies, such as inequality and impoverishment of the reserve army of labour (see
LaFree & Tseloni, 2006). Additionally, the world systems perspective (see Wallerstein,
1974) implicates potential differences across subsets of nations based on their classifica­
tion as core, semi-periphery and periphery within the capitalist world order. In accordance
with this classification system, core nations represent the wealthiest nations, semi-per­
iphery could be described as ‘developing’ nations that still lack the resources of a core
 GLOBAL CRIME 5

nation, and periphery nations are the least economically developed and are even more
dependent than the semi-periphery on aid from core nations (Wallerstein, 1974). While it
is unclear what the precise prediction of crime trends according to world systems theory
(Rogers & Pridemore, 2018), there are potential differences in trends and the correlates of
homicide trends across these classifications of nations. For example, some have sug­
gested that there may be homicide declines in the ‘core’ capitalist nations while nations in
the periphery of the world system may experience high and rising rates of homicide due
the a greater concentration of poverty, unemployment, and lack of social services in
periphery nations (LaFree, 2005).
A closer examination of the implications of prominent criminological theories in
broader comparative context also suggests potential differences across subsets of
nations. For example, Institutional Anomie Theory (i.e., Messner & Rosenfeld, 1994) is
designed to explain the disparity in violent crime between the United States and other
wealthy democracies. This theory focuses on the institutional imbalance between eco­
nomic and non-economic institutions, arguing that this institutional imbalance causes
decreased social control and higher rates of violent crime within American society.
However, the theoretical tradition of American criminology often uses observations of
the social dynamics of the wealthy ‘West’ and attempts to generalise and apply these
observations to other contexts, with varying degrees of success (see Messner, 2015). While
it is often implicitly assumed that theories designed to explain variation in crime are
widely applicable, the possibility remains that the causes of crime trends differ across
social contexts. Among wealthy nations, certain conditions that suppress crime are
implicitly assumed, such as political legitimacy (e.g., Nivette & Eisner, 2013), a relatively
large welfare state (e.g., Messner & Rosenfeld, 1997), effective criminal justice institutions
(e.g., Pare, 2014), and a relatively high level of economic affluence (e.g., Pridemore, 2008;
2011; Nivette, 2011). Therefore, the institutional imbalance may or may not have the same
impact in nations where these conditions, common to wealthy democracies, are not
present.
Additionally, some theoretical concepts may manifest themselves differently across
nations. For example, the umbrella concept of social support (Cullen, 1994) is often
operationalised as some form of de-commodification (e.g., Messner & Rosenfeld, 1997;
McCall & Brauer, 2014) or social welfare expenditures (e.g., Pratt & Godsey, 2003) in
relatively small samples of wealthy nations. However, social support exists in some form
without government-sponsored social welfare expenditures (Cullen, 1994). Therefore, it
may be difficult to detect social support in nations with relatively weak social welfare
systems or few economic resources using de-commodification and social welfare expen­
ditures as proxy measures. The application of the social disorganisation perspective may
also differ across national contexts. The impact of urbanisation, one component of social
disorganisation (e.g., Sampson & Groves, 1989; Wirth, 1938), may impact crime rates in
relatively urban and rural nations differently. Overall, broad theoretical concepts may
manifest themselves differently in across subsets of nations.
There is also the distinct possibility that that economic and demographic trends
actually produce different impacts on homicide rates across subsets of nations. For
example, in nations where homicide rates are relatively low, a comparatively small
impact exerted by demographic and economic factors will be more readily detected.
Therefore, the same factors may still impact homicide across nations, but the impact of
6 J. TUTTLE ET AL.

economic and demographic factors may not be perceivable in nations with social
dynamics contributing to high homicide rates (Santos, Testa, Porter, Lynch, 2019).
Additionally, while this seems less likely, Bergman (2018) suggests that economic
and demographic trends have produced unanticipated effects in Latin American
nations during recent decades. As economic conditions have improved in the ‘legit­
imate’ economy, they have helped spur an increase in the demand for consumer
goods and an underground economy to fulfill this increased demand. Organised crime
has flourished under these conditions, facilitating an increase in homicide rates.
Therefore, the same economic conditions that facilitated a 1990s decline in crime
across wealthy democracies (Baumer & Wolff, 2014; Rosenfeld, 2014; Rosenfeld &
Messner, 2009) may have actually caused homicide rates to increase in Latin
America and other nations with relatively large underground economies (Bergman,
2018).
Despite the possibility of divergent impacts of demographic and economic trends,
previous research on 1990s international crime trends has (implicitly) assumed that the
correlates of homicide trends are invariant across nations. For example, Baumer and
Wolff's (2014) analyses of (up to) 65 nations is one of the largest samples addressing
the covariates of 1990s homicide trends. Overall, the authors find that declining poverty,
percent urban, and increased youth oversight (population aged 45–64 divided by popula­
tion aged 15–24) are significantly associated with the homicide decline. However, by
including a diverse set of nations within a single analysis, Baumer and Wolff's (2014)
research may obscure the impact of heterogeneity within their relatively large sample. It is
unclear, whether these factors, which are associated with shifts in the homicide rate in the
overall sample, exert the same impact on nations that experienced divergent crime trends
(i.e., increases vs. decreases in homicide). Accordingly, the findings of the Baumer and
Wolff (2014) study raise the question: Are the predictors of temporal variation in homicide
invariant across nations?

The current study

To address this question, we employ latent trajectory analysis to uncover hidden
heterogeneity within the sample by grouping subsets of countries with similar trends.
Importantly for this analysis of 1990s homicide trends, we identify both the direction of
homicide trends (decline, stability, or increase) as well as the overall magnitude of
homicide rates across nations. We begin our analysis of hidden heterogeneity by
identifying distinctive trajectories among a sample of 77 nations from 1989 through
2010 using latent class trajectory analysis techniques. After identifying three latent
trajectory groups, we explore the economic and demographic correlates of crime across
these subsets of nations. We assess the impact of these economic and demographic
trends on longitudinal variation in homicide rates in separate panel analyses (using
fixed-effects regression techniques) for the overall sample as well as for the three
trajectory groups identified in latent trajectory analyses. By assessing the correlates of
the crime decline within subsets of nations grouped by homicide trend direction and
magnitude, we can determine whether the crime decline was driven by the same
universal factors and if these influences vary by the amount of violent crime or the
direction of homicide trends within a nation.
 GLOBAL CRIME 7

Sample and methods

In the first stage of this analysis, annual data for total homicides are utilised for the latent
class trajectory procedure. World Health Organisation (WHO) mortality data are used as
the standard for this analysis1. These measures are derived from the nation-reported WHO
mortality counts, not the potentially problematic WHO estimates based upon statistical
modelling (see Kanis, Messner, Eisner, Heitmeyer, 2017). In total, 77 countries are included
in this sample based upon sufficient availability of data concerning homicide rates and
relevant explanatory variables for the 22-year period. In an effort to obtain an uninter­
rupted time series, we supplemented homicide data collected by the United Nations
Office on Drugs and Crime (UNODC) Annual Survey on Crime Trends and Operations of
Criminal Justice Systems2. Additional missing data points are supplemented with National
Research Institute of Legal Policy’s Comparative Homicide Time Series (NRILP-CHTS)3.
After these efforts to obtain a complete time series for each nation (1,694 country-years
in total representing 77 nations over 22 years), 64 country-years (3.78%) remained miss­
ing. The final sample included 1,630 country-years, which is approximately 96.2% of a
possible 1,694 country-years. Despite this missing data, the trajectory procedure can still
be estimated effectively. We used a three-year-moving-average to smooth periods of
extremely high or low homicide rates and employed homicide data for the years 1988 and
2011 in calculating the moving average to retain cases. Due to the skewed distribution of
homicide rates in this sample, the three-year-moving average of homicide rates are
transformed using the natural logarithm, consistent with previous research.
It should be noted that supplementing the WHO dataset with additional homicide data
sources (UNODC, NRILP-CHTS) is not a common practice. We also use linear interpolation
for instances in which a single year is missing within a series rather than more conven­
tional multiple imputation techniques (see Allison, 2002). The main reason for this
divergence from common practice, consistent with some previous latent trajectories
research (i.e., Tuttle, McCall, Land, 2018), is that multiple imputation techniques use
data from across the dataset, producing estimates based upon all available cases with
complete data. Because this study emphasises hidden heterogeneity within distinctive
groups of countries, an imputation technique which imputes the ‘average’ expected value
would potentially bias our results towards finding greater homogeneity of homicide
patterns in latent trajectories models and greater similarities of effects in regression
models than actually exists. Because measurement is not as reliable/consistent within
the UNODC and NRILP-CHTS homicide datasets (Kanis, Messner, Eisner, Heitmeyer, 2017),
we also conducted the analyses using only WHO mortality count statistics to determine if
combining these datasets influenced the substantive results. Both the WHO and the
combined data produced substantively similar results, available upon request4.

Latent trajectory modelling

To estimate latent class trajectory models of the longitudinal homicide rate data from
1989 to 2010 described above, we use Stata’s procedure written by Jones and Nagin
(2013). Latent class trajectory analysis allows researchers to examine hidden heterogene­
ity in variation in longitudinal trends in homicide rates. Furthermore, we are able to
determine the number and nature of latent classes of countries that emerge from using
8 J. TUTTLE ET AL.

this procedure. While latent class trajectory modelling techniques were initially applied to
criminological research to study individual-level variation in offending (e.g., Nagin & Land,
1993), aggregate-level research has increasingly used this technique to identify unique
trajectories in homicide rates across cities (McCall, Land, Parker, 2011; Peres & Nivette,
2017), regions (Roldan, Fondevila, García-Tejeda, 2021), and nations (Tuttle, McCall, Land,
2018; Weiss, Santos, Testa, Kumar, 2016). The current study uses a similar procedure to
analyse latent trajectories within this diverse cross-national sample.5
Our latent class trajectory analysis model reveals three homicide patterns within the
sample. The parameter generated for annual national homicide rates (the three-year-
moving average transformed using the natural logarithm) estimates a three-trajectory
model. The model depicted in Table 1 was selected based on the statistical significance of
the highest order polynomial, the BIC statistic and how closely the predicted homicide
rates mirrored the observed data6. The three groups (henceforth labelled as ‘low rate,’
‘medium rate,’ and ‘high rate’ based on their position in Figure 1) present three homicide
patterns. The best fitting three-trajectory model with polynomials of the second and third
order (1 3 2) resulted in a linear term for the low rate group, a cubic term for both the
medium rate group and a quadratic term for the high rate group. The low rate group
includes nations located mostly in Western Europe, the Middle East, and Oceania. The
medium rate group largely consists of Eastern European nations, the United States, Asia
and Latin American countries. The high rate group includes mostly Central and South
American nations, as well as Caribbean Islands, a few Eastern European nations and
Russia. See Table 2 for the nations identified in the low, medium and high rate groups.
Figure 1 depicts the trajectories predicted (smoothed) and observed (dotted-lines) of
the natural logarithm of the three-year-moving average of homicide rates among nations
within each trajectory group. These groups vary dramatically in the absolute magnitude
of their homicide rates, but there are some similarities in homicide trends across groups.
In the low rate group, the homicide rate gradually declines from the early 1990s onward.
The medium rate group also displays a slight decline in homicide rates during the early to
mid-1990s, sharing a similar downward trend in homicide as the low rate group from 1992
to 2008. Yet, the medium rate group displays a distinct upward pattern in the most recent
years of this time period (post-2007). Finally, in the high rate group, there is no significant

Table 1. Time Predictors Associated with Latent Group Membership

with Low to High National Three-Year Moving Averaged Homicide
Rates from 1989 to 2010 with Parameter Estimates and T-Values (in
Parentheses), N = 77.
Low Medium High
Rate Group Rate Group Rate Group
N = 33 N = 20 N = 24
Intercept 0.54* 1.69* 2.76*
(23.20) (40.84) (91.06)
Time
Linear −.18* −.51* .18*
(−6.48) (−5.72) (5.48)
Quadratic –- −.17* −.22*
(−2.66) (−3.88)
Cubic –- .43* –-
(3.87)
Note: * p < .001. BIC = −1156.39 (N = 1630)
 GLOBAL CRIME 9

3
logged Homicide Rate, 3 yr ma
Log of 3 Year Average of Homicide

2
1
0

1989 1994 1999 2004 2009

Year

1 42.4% 2 26.5%
3 31.2%

Figure 1. Average Predicted and Observed Homicide Rates for Three-Group Models, 77 Nations, 1989–
2010.

homicide rate decline noted in the 1990s. In the final year of this analysis (2010), there is a
marked decline in homicide rates in the high rate group, but it is unclear if this is indicative
of a larger pattern7.

Descriptive statistics
After identifying three patterns of homicide in this sample, we analyse temporal variation
in homicide in separate panel regression analyses. As indicated in Table 2, the three
trajectory groups (low, medium, and high) have 33, 20, and 24 nations in each, respec­
tively. To minimise the impact of missing annual data, the data are organised into panels
representative of every third year of the time series. Therefore, there are eight waves in
total (circa 1989, 1992, 1995, 1998, 2001, 2004, 2007 and 2010), with 616 nation-years
(potentially included) in the total sample8. After including the explanatory variables (listed
below), there are 573 complete cases (93%) of 616 possible nation-years in the overall
sample of 77 nations. In the low rate group, there are 253 complete cases (96%) of 264
nation-years in 33 nations. The medium rate group has 149 complete cases (93%) of 160
nation-years in 20 nations. Finally, the high rate group has 171 complete cases (89%) of
192 possible nation-years in 24 nations.
Table 3 presents the descriptive statistics (means and standard deviations) for the
entire sample as well as those organised by latent trajectory group. In the sample of 77
nations, the overall mean of the homicide rate is 7.7 per 100,000 population, beginning
with 5.9 in 1989 and finishing with 7.5 in 2010. The mean homicide rate for the low rate
group is 1.4 per 100,000, 5.7 for the medium rate group and 18.1 for the high rate group.
Over time, we observe the patterns for these three samples reflected in Figure 1. In the
low rate group, the homicide rate decreases over time from 1.8 to 1. In the medium rate
10 J. TUTTLE ET AL.

Table 2. Country Composition of Groups Identified from Latent Class Trajectory Analysis, 1989–2010.
Low Rate Group (n=33) Medium Rate Group (n=20) High Rate Group (n=24)
Australia Albania [5] The Bahamas [2]
Austria Argentina Barbados [4]
Belgium Armenia Belarus
Canada Azerbaijan [1] Belize
Croatia Bulgaria Brazil
Czech Republic Chile Columbia
Denmark Costa Rica Ecuador
France Dominican Republic El Salvador [2]
Georgia [4] Finland Estonia
Germany Hungary Guatemala
Greece Kyrgyzstan Guyana
Iceland Mauritius Kazakhstan
Ireland Nicaragua Latvia
Israel Romania Lithuania
Italy Sri Lanka [16] Mexico
Japan Suriname [4] Moldova
Kuwait [4] Tajikistan Panama [8]
Luxembourg Thailand Paraguay [4]
Malta Uruguay Russian Federation
Netherlands USA Saint Lucia
New Zealand St. Vincent and the Grenadines [7
Norway Trinidad Tobago [1]
Peru Ukraine
Poland Venezuela
Portugal
Singapore
Slovakia [2]
Slovenia
Spain
Sweden
Switzerland
Turkey
United Kingdom
N = 77, number of missing annual country-year homicide data in brackets (out of 22 possible country-years).

group, there is a slight decline in the late 1990s and early 2000s and an overall decrease
between 1989 and 2010 from 6.1 to 4.5. Finally, the high rate group homicide rates
increased from the first year to the final year, ranging from 12.1 to 20.5.

Explanatory variables
Descriptive statistics for the explanatory variables analysed in this study are reported in
Table 3. These variables are analysed primarily because they are demographic and
economic variables that are commonly accounted for in the cross-national homicide
literature. Some of these variables also represent factors that are either implicitly or
explicitly expressed as potential causes of the 1990s crime decline. These include an
improving economy, changing demographics, and shifts in population urbanisation.
Beginning with economic conditions, declines in poverty (Baumer & Wolff, 2014) and
improving consumer sentiment (Rosenfeld & Messner, 2009) are associated with the crime
decline. To assess improving economic conditions, we analyse changes in GDP per capita
and infant mortality rates, hypothesised to indicate the level of poverty (e.g., Baumer &
Wolff, 2014; Pare & Felson, 2014; Pridemore, 2008; 2011). Other variables were considered
to capture economic conditions but were excluded because of missing data issues or
 GLOBAL CRIME 11

evidence of the partialling fallacy in analysis (see Gordon, 1968; Land, McCall, Cohen,
1990). Income inequality, which is consistently associated with homicide rate in cross-
sectional research (see Nivette, 2011; LaFree, 1999), had extensive missingness for many
nations. This missingness prevented an analysis of separate trajectory groups. In the
overall sample, income inequality was not significantly associated with homicide rates9.
Additionally, we initially considered the impact of consumer inflation as well (see
Rosenfeld, 2014), but it presented issues associated with the partialling fallacy when
included in the analysis. When analysed separately from variables contributing to this
collinearity issue, inflation was not a consistent statistically significant predictor of varia­
tion in homicide rates and was therefore not included in final analyses.
Additionally, prior research indicates that changing demographic structure is asso­
ciated with the cross-national crime decline (Baumer & Wolff, 2014; Santos, Testa, Porter,
Lynch, 2019), although this finding is not universal in homicide research (Rogers, 2014;
Rogers & Pridemore, 2016). Given the ‘age-crime curve’ (Hirschi & Gottfredson, 1983), a
greater number of young men in the population is expected to contribute to higher
homicide rates. As populations age, we hypothesise that fewer young males in the
population will contribute to lower homicide rates. We include an estimate of the
percentage of the male population aged 15 to 29 to capture this demographic correlate
of crime. Additionally, we account for sex ratio, which has also been included as a
demographic control in recent cross-national homicide research (e.g., Rogers &
Pridemore, 2018).
Finally, we accounted for shifts in population organisation and density hypothesised
to assist or inhibit social control. Urbanisation is hypothesised to reduce societal
capacity for social control according to the social disorganisation perspective
(Sampson & Groves, 1989; Wirth, 1938) and will potentially have a larger impact in
less economically-developed nations as part of the disruptive impact of modernisation
(Shelley, 1981). Additionally, declining urbanisation has been linked to the cross-

Table 3. Descriptive Statistics (Means and Standard Deviations) for 77 Nations and Three Trajectory
Groups 1989–2010.
Low High
Total Sample Rate Group Medium Rate Group Rate Group
(N = 77) (N = 33) (N = 20) (N = 24)
Homicide Rate 7.65 1.37 5.72 18.08
(10.92) (.95) (6.05) (13.72)
Homicide Rate 1989 5.87 1.81 6.08 12.31
(9.08) (1.39) (7.62) (13.27)
Homicide Rate 2010 7.54 .94 4.34 19.84
(11.13) (.47) (2.62) (13.89)
Population Size 24,081.48 22,294.82 25,256.31 25,548.96
(in thousands) (44,518.9) (28,847.07) (60,281.06) (46,700.59)
Percent Urban 65.60 75.65 58.85 57.41
(19.46) (13.69) (20.52) (19.08)
Sex Ratio (Males/Females) 96.70 97.48 97.41 95.02
(5.83) (6.98) (3.14) (5.46)
Percent of Males aged 24.64 22.54 25.68 26.65
15–29 (3.31) (2.88) (2.91) (2.41)
Infant Mortality 17.04 8.38 24.84 22.44
(14.63) (8.93) (18.47) (10.50)
GDP per capita 18,735.55 33,516.19 8,388.91 7,002.47
(19,467.64) (19,995.32) (11,558.57) (5,143.39)
12 J. TUTTLE ET AL.

national homicide decline (Baumer & Wolff, 2014). A variable also related to the social
disorganisation perspective is growth in population size, as this factor may also con­
tribute to reduced social control as rapidly growing populations may reduce social
integration. Due to a heavily skewed distribution and a non-linear relationship with
homicide rates, the natural logarithm of population size is used in the final models. See
Appendix A for detailed information on sources for all explanatory and outcome
variables. Notably, in relation to the emphasis of this study, all subsets of nations
share general trends in the explanatory variables, except for possibly the sex ratio
(see Appendix B). That is, despite differences noted in homicide patterns (see Figure
1), all three subsets of nations experienced the same general demographic and eco­
nomic trends.

Preliminary analyses
Before estimating fixed-effects pooled ordinary least squares (OLS) multiple regression
models, we conducted preliminary analyses on these panel data. Some common
potential problems in observational studies, including concerns with endogeneity
associated with possible omitted variable bias, do not have straight-forward or
satisfactory solutions (Baumer & Wolff, 2014). Other issues, such as non-stationarity
bias, are frequently assessed and accounted for in time-series analyses. Unfortunately,
due to the relatively short time series (eight panels total), we are unable to effectively
assess non-stationarity using a unit root test. However, scholars have argued that
short time series do not pose significant problems with non-stationarity and therefore
do not perform common transformations of the data, such as first-differencing (see
Baumer & Wolff, 2014; Messner, Raffalovich, Sutton, 2010). In addition to non-statio­
narity, scholars conducting aggregate-level homicide research are often concerned
with collinearity among explanatory variables (Land, McCall, Cohen, 1990). To account
for potential issues concerning multicollinearity and the partialling fallacy, bivariate
correlations between variables were analysed. Additionally, variables were system­
atically entered into each model to examine shifts in coefficients across nested
models. These preliminary analyses displayed no issues with collinearity.
To analyse the impact of demographic and economic forces on homicide rates over
time, the following analyses use fixed-effects pooled OLS regression. Fixed-effects regres­
sion analyses have the distinct advantage of holding all time-stable influences constant
across nations. After conducting a series of Hausman tests on the overall sample and each
trajectory group, it was determined that fixed-effects modelling is more efficient than
random effects estimation. Additionally, we attempt to reduce issues associated with
heteroscedasticity by using robust standard errors, clustered by nation, in estimated
models. Finally, we control for period effects using a dummy variable for each panel to
account for the impact of time on the outcome variable.
To address the concern that using multiple data sources for homicide data would
bias the final regression results, we conducted the final regression analyses both
including all three homicide data sources combined (WHO, UNODC and NRILP-CHTS)
as well as with only WHO data. The final analyses for the full sample and each trajectory
group were substantively identical between the combined and WHO-only homicide
 GLOBAL CRIME 13

data (results available upon request), which is indicative of the robustness of these
results.

Results
Main effects models
Four models are estimated in total, including a model representing nations from the full
sample, the low rate group, the medium rate group and the high rate group. The results
for the full sample are displayed in Model 1 located in Table 4. In the full sample, the
natural logarithm of population size and infant mortality rates are positively associated
with the natural logarithm of the three-year average of homicide rates, while GDP per
capita is negatively associated with homicide. That is, the increase in population size
contributed to higher rates of homicide during this period of overall homicide decline.
Conversely, the decrease in infant mortality and increase in GDP per capita (see Appendix
B) contributed to homicide rate decline in the overall sample. The other demographic
variables in the model (percent urban, sex ratio and males aged 15–29 years) fail to
achieve statistical significance. The explanatory variables in the model (as well as controls
for panel effects) explain 21% of the variation over time in the outcome variable (homi­
cide) within the total sample of 77 nations.
Model 2 in Table 4 depicts the results of the same model specifications as used in the
total sample for the low rate trajectory group. Within this subset of 33 nations, results
differed from the total sample of 77 nations. Unlike Model 1, GDP per capita and the
natural logarithm of population size are not significantly associated with homicide, while
percent urban and percentage of the male population aged 15 to 29 are both positively
associated with homicide. Similar to the full sample analysed in Model 1, the infant
mortality rate is significantly and positively associated with homicide rates in Model 2.
Substantively, this means that increases in urbanisation kept homicide rates higher, while
decreases in the percent males aged 15 to 29 and infant mortality contributed to the
homicide rate decline in the low rate trajectory group. Model 2 explains approximately
34% of the variation in homicide rates over time within these nations.
Model 3 (depicting the medium rate group) and Model 4 (high rate group) in Table 4
show similar results to Model 1 (full sample). The natural logarithm of population size and
the infant mortality rate are positively (and significantly) associated with homicide in
Models 1, 3, and 4 and GDP per capita is negatively associated with homicide in Models 1
and 3, but not Model 4 (the high rate group). Substantively, decreases in infant mortality
and increases in GDP per capita contributed to the homicide decline in the medium rate
group, while population growth prevented an even steeper homicide decline in the
medium rate group. In the high rate group, which exhibited no consistent decline and
even a slight increase in homicide, the significant impact of infant mortality during the
period failed to produce a homicide rate decline, possibly counterbalanced by the
increases in population size, which contributed to higher rates of homicide. The other
variables in the models (percent urban, sex ratio and percent of males aged 15–29 years)
fail to reach statistical significance for both the medium and high rate groups. Model 3
(medium rate group) explains approximately 48% of the variation in homicide while
Model 4 (high rate group) explains 42%. In each of the trajectory groups, the models
14 J. TUTTLE ET AL.

explained significantly more variation in the outcome variable (ranging from 34% to 48%)
than the model including all nations in the sample (21%)10.
There are two notable findings in these analyses. The first issue pertains to the
differences in the substantive findings across groups. In the low rate group models, the
percent males aged 15–29 and percent urban is significantly associated with the homicide
rate. Yet, in the full sample, medium rate group and high rate group, the percent males
aged 15–29 and percent urban are not significantly associated with homicide. In the total
sample, medium rate group and high rate group, population growth is significantly
associated with homicide, but not in the low rate group. Additionally, GDP per capita is
associated with homicide in the total sample and medium rate group, but not the low rate
or high rate group. While the infant mortality rate shares a fairly consistent association
with the homicide rate regardless of sample composition, the other correlates of homicide
are not consistent across different trajectory groups.
The second issue illuminated by these findings is the greater amount of variation
explained in homicide rates within the separate trajectory group models (ranging from
34% to 48%) compared to the variation explained in the overall sample (21%). While this
increase in explained variance might be due in part to a smaller sample size in the specific
trajectory groups models compared with the overall sample, it is also suggestive that
analysing all the nations in a relatively large sample in a single analysis may reduce
explained variance due to distinctive correlates of homicide across different trajectory
groups. Taken together, the results indicate that analysing all the nations together in this
sample obscures the distinct demographic and economic influences on the homicide
trends across subsets of nations11.

Table 4. Pooled Ordinary Least Squares Regression Fixed-Effects Coefficients on the Natural Logarithm
of Homicide Rates for 77 Nations and Three Trajectory Groups, 1989–2010.
Model 2 Model 4
Model 1 Low Model 3 High
Total Sample Rate Group Medium Rate Group Rate Group
77 Nations 33 Nations 20 Nations 24 Nations
N = 573 N = 253 N = 149 N = 171
Population (ln) 2.35** .68 3.50** 3.64**
(.82) (.86) (.78) (1.22)
Percent Urban .01 .04** −.01 −.05
(.02) (.01) (.02) (.03)
Sex Ratio .02 .05 −.003 −.04
(.04) (.04) (.02) (.07)
Percent Males Aged 15–29 .03 .05* .004 .004
(.03) (.02) (.03) (.06)
Infant Mortality .04** .02* .06** .03*
(.01) (.01) (.01) (.02)
GDP per Capita −.03** .002 −.06** −.02
(in thousands) (.01) (.01) (.02) (.05)
R2 -Within .21 .34 .48 .42
Note: Unstandardised regression coefficients with robust standard errors in parentheses. All models control for nation
fixed-effects and period effects. Robust standard errors are in parentheses.
* = p < .05, ** = p < .01 (one-tailed test)
 GLOBAL CRIME 15

Supplemental analyses
To examine the extent of the differences in findings across latent trajectory groups, we
conducted supplemental analyses testing equality of regression coefficients between
models. We employ techniques that simultaneously compare the regressors’ unstandar­
dised coefficients and standard errors across models (i.e., Paternoster, Brame, Mazerolle,
Piquero, 1998) to ascertain if there are statistically significant differences in predictors of
homicide rates across these models. The comparison of the equality of coefficients of
among these three independent groups (low, medium and high) is presented in Table 5.
As suggested by the differences in findings in Table 4 discussed above, the low rate
group appears to be distinct from the other two groups in this analysis. There are
statistically significant differences between the low rate group and medium rate group
in the coefficients for population size, percent urban, infant mortality and GDP per capita.
While the sex ratio and percent males aged 15–29 do not reach statistical significance, the
differences in coefficients and direction of the association in the regression analyses
located in Table 4 suggest that economic and demographic correlates of homicide differ
across these two groups as well. The low rate group and high rate group also significantly
differ on two regressors (population size and percent urban), but do not appear to diverge
as greatly from the low rate group as the medium rate group. The associations with
homicide rates in medium rate and high rate group are quite similar (see Table 4) and, not
surprisingly, do not share statistically significant differences in Table 5. Overall, it appears
that the subset of low rate group countries differs significantly from the rest of the sample.

Discussion
Overall, this study contributes to the literature on cross-national homicide trends during
the 1990s. We identify three distinctive trajectories in an examination of homicide rates in
77 nations from 1989 through 2010. In addition to identifying distinct homicide rate
trajectories, this study builds on previous research (i.e., Tuttle, McCall, Land, 2018; Weiss,
Santos, Testa, Kumar, 2016) by assessing the temporal correlates of homicide within
separate panel analyses. These panel analyses reveal that the factors contributing to
1990s crime trends are not invariant, as the temporal correlates of homicide rates vary
across subsets of nations. Even among subsets of nations that experienced a homicide
decline during the 1990s (low and medium rate groups), the correlates of homicide
differed (see Table 5). Given these differences, our results imply that examining temporal
variation in 1990s homicide rates within large, diverse samples of nations in a single
model may distort or obscure the importance of certain demographic and economic
forces in explaining disparate cross-national homicide trends. These results suggest that
future research on cross-national homicide trends in larger samples needs to exercise
caution in the analysis and interpretation of the temporal correlates of homicide.
While this study confirms some previous research on 1990s homicide trends, it also
contradicts other conclusions. Beginning with findings consistent with previous research,
we largely confirm the previous examination of homicide trajectories in both the basic
contours of homicide trends as well as the groupings of nations experiencing these trends
(i.e., Tuttle, McCall, Land, 2018). In terms of the correlates of homicide, a decline in the
percentage of the population comprised of young males is associated with a decline in
16 J. TUTTLE ET AL.

Table 5. Comparison of Coefficients across Models Tests (Z-Values).

Low Low
vs. Medium vs. High Medium vs. High
Population (ln) −2.53* −2.04* −0.10
Percent Urban 2.56* 2.42* 0.94
Sex Ratio 1.22 1.14 0.47
Percent Males Aged 15–29 1.31 0.84 0.001
Infant Mortality −2.63* −0.65 1.41
GDP per Capita 2.88* 0.43 −0.68
(in thousands)
Note: * indicates that the z-score exceeds 1.96, or the 95% confidence interval.

homicide rates in our low rate trajectory group, consistent with previous research (Baumer
& Wolff, 2014; Santos, Testa, Porter, Lynch, 2019). Additionally, the significant association
between infant mortality and homicide is consistent with research suggesting that the
crime decline was partially due to improving economic conditions (a reduction in poverty)
during the 1990s (Baumer & Wolff, 2014; Rosenfeld, 2014; Rosenfeld & Messner, 2009).
However, our other findings are inconsistent with previous research. For example,
Baumer and Wolff (2014) report that percent urban is positively associated with homicide
rates in their analysis of 65 nations. Accordingly, they suggest that ‘de-urbanisation’ may
have contributed to the 1990s homicide decline, as some of the nations in their sample
experienced a decline in urbanisation. Our findings concerning the low rate group mirror
their findings (a positive association with homicide rates), but our conclusions are vastly
different. Among low rate group nations, the percent living in urban areas increased
during this period (Appendix B), suggesting that 1990s homicide rate declines would have
been even steeper if urbanisation had not increased. We find no support for the de-
urbanisation hypothesis.
In relation to theoretical implications of these findings, the correlates of homicide
appear to differ the most between our subset of highly economically-developed low rate
homicide trajectory nations and the rest of the sample. For example, urbanisation exhib­
ited a significant and positive impact on homicide in the low rate trajectory group while
population growth was more influential on the homicide rate in the medium and high
rate trajectory nations. Both of these population variables are broadly conceptualised as
different elements of social disorganisation, hypothesised to reduce informal social con­
trol (e.g., Sampson & Groves, 1989; Wirth, 1938). In the more urbanised nations of the low
rate trajectory group, growth in urbanisation corresponds with higher rates of homicide,
suggesting that urbanisation may be a force of social disorganisation within these
nations. Conversely, social disorganisation may be expressed as population growth in
the less urban and economically-developed nations of the medium and high rate trajec­
tory groups.
Additionally, this research may have some implications for the grand metanarratives
of the modernisation and conflict perspective. Although these subsets of nations are
not perfect corollaries of the types of dynamics Durkheim (1984) and subsequent
authors (i.e., Shelley, 1981) describe in the modernisation perspective, the greater
impact of urbanisation on homicide rates in a subset of more economically-developed
nations contradicts this perspective. This finding suggests that the increases in urbani­
sation continue to exhibit an impact on homicide rates even in modern societies, for
which the transition to urbanisation and industrialisation have already occurred. In
 GLOBAL CRIME 17

reference to the conflict perspective, we find distinctive homicide trends that would be
consistent with some interpretations of world systems theory (i.e., LaFree, 2005). Low
and declining rates of homicide were evident in the wealthiest subset of nations while
high and stable rates of homicide were present in the least wealthy subset of nations
(high rate group). While these nations were not organised by a theoretical grouping
that strictly reflects world system theory (i.e., Rogers & Pridemore, 2018), the results do
suggest nations in the periphery of the world system have higher rates of homicide
than the core that are influenced by higher rates of poverty. This finding is consistent
with the idea that the negative impacts of the capitalist system are disproportionately
experienced by periphery and semi-periphery nations, resulting in higher rates of
homicide (i.e., LaFree, 2005).
The only consistency across trajectory groups is the association between infant mor­
tality (our proxy for poverty) and homicide, which contributed to the 1990s homicide rate
decline as infant mortality declined significantly during this time period (Appendix B). The
consistency of the infant mortality correlation across trajectory groups provides further
indications of the importance of poverty in shaping cross-national homicide variation (see
Pare & Felson, 2014; Pridemore, 2008; 2011). Yet, the high rate trajectory group did not
experience a decline in homicide during the 1990s despite falling rates of poverty
(measured as infant mortality). This suggests improving economic conditions may be
insufficient in nations with higher rates of homicide to reduce homicide, as there are likely
other criminogenic conditions at play. In terms of policy implications, anti-poverty pro­
grammes may assist in reducing homicide rates, but we cannot assume that this strategy
will be universally successful, as evidenced by the homicide trends of the high rate
trajectory group.
While we did not find any trajectory groups in which the same covariate produced the
opposite effect on homicide to a significant degree as suggested in previous research (i.e.,
Bergman, 2018), we find evidence that supports the idea that the effects of certain
economic and demographic factors are more pronounced in some contexts. For example,
homicide rates in the low rate trajectory group appear affected more by demographic
trends (a decline in males aged 15 to 29) than other trajectory groups. Confirming Santos
et al. (2019), there is a significant association between age structure and homicide in
nations with relatively low rates of homicide and no significant impact of age structure in
more homicide-prone nations. Additionally, there is some indication that the same
theoretical concept may be expressed differently across different subsets of nations12.

Limitations
This study of the impact of hidden heterogeneity on the correlates of 1990s homicide
trends has a few limitations. One limitation was the availability of data. There are some
factors implicated in the homicide decline of the 1990s that could not be accounted for
due to missing data for several countries. Notably, the impact of incarceration rates on
crime and homicide rates during the 1990s, which has been documented in smaller
samples of relatively wealthy nations (Buonanno, Drago, Galbiati, Zanella, 2011;
Rosenfeld & Messner, 2009; Tuttle, 2020), was not accounted for in this study.
Additionally, missing income inequality data did not allow for comparisons of the impact
of inequality on homicide across trajectory groups. Without inequality data, we are unable
18 J. TUTTLE ET AL.

to address the burgeoning poverty and inequality controversy within the cross-national
literature (Pare & Felson, 2014; Pridemore, 2008; 2011; Messner, Raffalovich, Sutton, 2010)
or the potential implications of various theories on cross-national variation in crime, such
as the conflict perspective. Finally, the time period under consideration in this study (1989
through 2010) was partially based upon data limitations of earlier periods. It is possible
that homicide patterns and the correlates of homicide differ depending upon the period
studied, but we were unable to examine those differences.
Additionally, we use an empirically-driven, rather than theoretically-driven, approach
to identify subsets of nations. This approach allows us to identify nations by their
respective homicide patterns to determine if these particular trends during the 1990s
were influenced by different factors. However, there are other relevant categories by
which comparisons across subsets of nations can be organised. Regional differences in
homicide rates (e.g., Nivette, 2011) and homicide trends (e.g., Rogers & Pridemore, 2017)
could serve as a precedent to examine distinctive correlates of homicide across regions.
Theoretically-driven groupings of nations are also conceivable, based upon level of
economic development, degree of decommodification (i.e., Messner & Rosenfeld, 1997),
or factors relevant to grand metanarrative on homicide trends such as modernisation and
conflict perspectives (i.e., Rogers & Pridemore, 2018). The current study suggests that
nations with relatively low and declining homicide rates differ in their correlates of
homicide in comparison with other nations, but future research is needed to address
whether the temporal correlates of homicide are related to theoretically-relevant features
of nations.

Conclusion
Our findings suggest that our analytical approach adds nuance to the analysis and
interpretation of cross-national crime trends, specifically, the interpretation of 1990s
homicide trends. Despite all three trajectory groups sharing similar economic and demo­
graphic trends (Appendix B), homicide rates declined in some nations (low rate and
medium rate groups) but not in others (high rate group). That is, all subsets of nations
experienced declines in infant mortality and percentage of the population aged 15 to 29
and increases in GDP per capita, urbanisation, and population, yet homicide trends
diverged between the high rate group and the low and medium rate groups (see
Figure 1). Additionally, even though the low and medium rate groups shared a period
of homicide decline during the 1990s, the correlates of homicide differed significantly
across these groups (see Table 5). These results imply that universalistic statements may
not accurately depict the reasons for the decline in homicide rates in some nations but
maintain stability or even increase in others. Instead, the correlates of 1990s homicide
patterns may be specific to different subsets of nations, rather than reflecting universal
correlates of homicide generalisable to all nations.

Notes
1. WHO data are available at: https://www.who.int/healthinfo/statistics/mortality_rawdata/en/.
2. UNODC data are available at:http://www.unodc.org/unodc/en/data-and-analysis/United-
Nations-Surveys-on-Crime-Trends-and-the-Operations-of-Criminal-Justice-Systems.html.
 GLOBAL CRIME 19

3. We used UNODC data for 66 missing data points, NRILP-CHTS for 12 years of data and
replaced the remaining missing cases using linear interpolation (25 cases) in instances for
which a single year is missing within a series.
4. WHO vital statistics data are considered to be the ‘gold standard’ homicide data, which are
based upon public health (mortality) data, while UNODC primarily base their information on
criminal justice surveys but supplement this data with other sources (Kanis, Messner, Eisner,
Heitmeyer, 2017). The UNODC criminal surveys may not be as reliable across national
contexts as cause of death data and are potentially prone to greater political influence. The
NRILP-CHTS also draws from multiple data sources including the WHO among others (see
Lappi-Seppälä & Lehti, 2014), combining estimates of homicide that are collected using
different definitions and methods of data collection.
5. Due to Colombia’s high rate of homicide during this period, the homicide rate for Colombia is
multiplied by .5 to bring it closer to other national homicide rates of other nations in the
sample to more effectively assess underlying trajectory group memberships. The results of
the trajectory analyses are substantively similar when Colombia is excluded from the sample
as compared with the results when it is included at .5 of its original homicide rate. To retain
cases, Colombia is included in the latent class trajectory analysis with the homicide rate
multiplied by .5. Note that this data transformation is only for the latent trajectory analysis,
not the panel analyses.
6. The best fitting model had the smallest BIC value (−1156.4), which is a better fit than the next
best model (1157.6) with an order of 112, suggesting the data favour our model (see Raferty,
1995: 129).
7. Relatedly, variation in homicide across situational circumstances also appears to change over
time (e.g., Skott, 2019), which may contribute to the different patterns in homicide we see
across subsets of nations.
8. To maximise complete cases, all available observations within the three-year panel period
were utilised if data were missing from the panel years. For example, if any of the three-year-
average of homicide rates were missing from panel year 1989, the observation three-year-
average for 1990 was substituted. This practice was also applied to missing data for expla­
natory variables within the three-year period for each panel as well, ensuring that the
explanatory variables were contemporaneous with or preceding the homicide data for
each panel. For example, for the panel year 1989, the data point for 1988 was employed
for a missing explanatory variable. Only five panel years include this data substitution
(Barbados, 1998; Kuwait, 1992; St. Vincent and the Grenadines, 1995; Suriname, 1992 & 1995).
9. Because of extensive missingness in income inequality data (derived from the United Nations
University – World Institute for Development Economic Research), we were unable to
estimate its impact on homicide in the subsamples of the medium and high rate nations.
For example, six of the 24 nations in the high rate group had one or fewer observations
during this period. Among the remaining 18 nations, missing data was extensive, limiting the
analysis to 93 country-years (64.6% of the possible 144 country-years). Due to the extremely
unbalanced nature of these panels, limited sample size and lack of association in the total
sample, income inequality is excluded from these analyses. Changes in substantive findings
were noted when income inequality was included in the total sample, but it was unclear if this
was due to the variable itself or the change in sample composition (fewer nations included
and more missing observations). Regression estimates for models including income inequal­
ity for the total sample are available upon request.
10. Because the trajectory groups are formed partially due to the pattern in homicide rates over
time, the control for period effects may explain the differences in variation explained
between the model for the overall group and the models for specific latent trajectory groups.
To determine if the higher percentage of the explained variation is due, in part, to the
controls for period effects, subsequent models are estimated removing the dummy variables
for each time period (not shown). Without the controls for period effects, the explained
variance for the low (28%), medium (47%), and high (36%) rate groups still exceeded that of
the model including all nations in the sample (20%).
20 J. TUTTLE ET AL.

11. Additional supplemental analyses were conducted to examine the robustness of these
findings. To examine the potential impact of influential outlier observations on parameter
estimates, Cook’s distance values were analysed (after analysing the data using OLS regres­
sion techniques) and outlier cases were removed to determine whether the models resulted
in substantive changes in the parameter estimates. In the low rate group, the percent males
aged 15 to 29 now approaches statistical significance (p < .06, one-tailed test), but is no
longer significant at p < .05. In the medium rate group, percent urban parameter changes
from not statistically significant to negative and significant. There are no substantive changes
in the high rate group model. Overall, while some findings were subject to change, the results
were fairly robust to the influence of potential outliers.
12. We do not address Bergman’s (2018) claims that nations with a large informal economy, such
as many in Latin America, will experience increases in crime due to economic growth. In
separate analyses of Latin American nations (results not shown), we do not find support for
this claim, as the positive association between GDP per capita and homicide did not achieve
statistical significance.

Acknowledgments
The authors would like to thank Eric P. Baumer for his helpful comments contributing this
manuscript.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Notes on contributor
James Tuttle is assistant professor of Sociology at the University of Montana in Missoula, MT. He has
previously published cross-national homicide research in Justice Quarterly, Homicide Studies, and the
International Journal of Comparative and Applied Criminal Justice.
Patricia McCall is professor emerita of Sociology at North Carolina State University. Her research
is embedded in the study of crime and social control and recent efforts involve cross-national
analyses of homicide trends.
Kenneth Land is the John Franklin Crowell professor of Sociology and Demography emeritus and
research professor in the Social Science Research Institute at Duke University. He is a fellow of the
American Society of Criminology and of the American Statistical Association. Known in criminology
for his work on unemployment and crime rate fluctuations, structural covariates of crime rates, and
finite mixture models of delinquent/criminal careers, he is the author or co-author of over 250
articles, chapters, and books.

ORCID
James Tuttle http://orcid.org/0000-0002-5197-4361

References
Allison, P. D. Missing Data. Thousand Oaks, CA: Sage, 2002.
Ayres, I., and J. J. Donohue III. “Shooting down the “More Guns, Less Crime” Hypothesis.” Stanford
Law Review 55, no. 4 (2002): 1193–1312.
 GLOBAL CRIME 21

Baumer, E. P. “An Empirical Assessment of the Contemporary Crime Trend Puzzle: A Modest Step to
A More Comprehensive Research Agenda.” In Understanding Crime Trends: Workshop Report,
edited by A. S. Goldberger and R. Rosenfeld, 127–176. Washington, D.C.: National Academy
Press, 2008.
Baumer, E. P., and K. T. Wolff. “The Breadth and Causes of Contemporary Cross-national Homicide
Trends.” Crime and Justice 43, no. 1 (2014): 231–287. doi:10.1086/677663.
Baumer, E. P., M. B. Vélez, and R. Rosenfeld. “Bringing Crime Trends Back into Criminology: A Critical
Assessment of the Literature and A Blueprint for Future Inquiry.” Annual Review of Criminology 1,
no. 1 (2018): 39–61. doi:10.1146/annurev-criminol-032317-092339.
Bergman, M. More Money, More Crime: Prosperity and Rising Crime in Latin America. New York: Oxford
University Press, 2018.
Blumstein, A., and J. Wallman. The Crime Drop in America. New York: Cambridge University Press,
2006.
Blumstein, A., and R. Rosenfeld. “Explaining Recent Trends in U.S. Homicide Rates.” The Journal of
Criminal Law & Criminology 88, no. 4 (1998): 1175–1216. doi:10.2307/1144254.
Brauer, J. R., and P. L. McCall. “Social Welfare Support and Homicide: Longitudinal Analyses of
European Countries from 1994 to 2010.” Social Science Research 48, no. 4 (2014): 90–107.
doi:10.1016/j.ssresearch.2014.05.009.
Buonanno, P., F. Drago, R. Galbiati, and G. Zanella. “Crime in Europe and the United States:
Dissecting the ‘Reversal of Misfortunes’.” Economic Policy 26, no. 67 (2011): 347–385.
doi:10.1111/j.1468-0327.2011.00267.x.
Chon, D. S. “Medical Resources and National Homicide Rates: A Cross-national Assessment.”
International Journal of Comparative and Applied Criminal Justice 34, no. 1 (2010): 97–118.
doi:10.1080/01924036.2010.9678819.
Chu, D. C., and R. F. Tusalem. “The Role of the State on Cross-national Homicide Rates.” International
Criminal Justice Review 23, no. 3 (2013): 252–279. doi:10.1177/1057567713500790.
Cook, P. J., and J. H. Laub. “The Unprecedented Epidemic in Youth Violence.” Crime and Justice 24
(1998): 27–61. doi:10.1086/449277.
Cullen, F. T. “Social Support as an Organising Concept for Criminology: Presidential Address to the
Academy of Criminal Justice Sciences.” Justice Quarterly 11, no. 4 (1994): 527–559. doi:10.1080/
07418829400092421.
Donohue, J. J., III, and S. D. Levitt. “The Impact of Legalised Abortion on Crime.” The Quarterly Journal
of Economics 116, no. 2 (2001): 379–420. doi:10.1162/00335530151144050.
Durkheim, E. The Division of Labour in Society. New York: The Free Press, 1984.
Eisner, M. “Modernity Strikes Back? A Historical Perspective on the Latest Increase in
Interpersonal Violence (1960-1990).” International Journal of Conflict and Violence 2, no. 2
(2008): 288–316.
Farrell, G., A. Tseloni, J. Mailley, and N. Tilley. “The Crime Drop and Security Hypothesis.„ Journal of
Research in Crime and Delinquency 48, no. 2 (2011): 147–175. doi: 10.1177/0022427810391539
Farrell, G. “Five Tests for a Theory of the Crime Drop.” Crime Science 2, no. 1 (2013): 1–8. doi:10.1186/
2193-7680-2-5.
Fox, J. A. “Demographics and U.S. Homicide.” In The Crime Drop in America, edited by A. B. Blumstein
and J. Wallman, 288–318. New York: Cambridge University Press, 2006.
Fox, J. A., and A. R. Piquero. “Deadly Demographics: Population Characteristics and Forecasting
Homicide Trends.” Crime and Delinquency 49, no. 3 (2003): 339–359. doi:10.1177/
0011128703049003001.
Gordon, R. A. “Issues in Multiple Regression.” American Journal of Sociology 73, no. 5 (1968): 592–616.
doi:10.1086/224533.
Hirschi, T., and M. R. Gottfredson. “Age and the Explanation of Crime.” The American Journal of
Sociology 89, no. 3 (1983): 552–584. doi:10.1086/227905.
Johnson, B. D., A. Golub, and E. Dunlap. “The Rise and Decline of Hard Drugs, Drug Markets, and
Violence in Inner-city New York.” In The Crime Drop in America, edited by A. B. Blumstein and J.
Wallman, 164–206. New York: Cambridge University Press, 2006.
22 J. TUTTLE ET AL.

Jones, B. L., and D. S. Nagin. “A Note on A Stata Plugin for Estimating Group-based Trajectory
Models.” Sociological Methods & Research 42, no. 4 (2013): 608–613. doi:10.1177/
0049124113503141.
Kanis, S., S. F. Messner, M. P. Eisner, and W. Heitmeyer. “A Cautionary Note about the Use of
Estimated Homicide Data for Cross-national Research.” Homicide Studies 2, no. 4 (2017): 312–
324. doi:10.1177/1088767917715670.
Kelling, G. L., and W. J. Bratton. “Declining Crime Rates: Insiders’ Views of the New York City Story.”
Journal of Criminal Law and Criminology 88, no. 4 (1998): 1217–1232. doi:10.2307/1144255.
LaFree, G. “A Summary and Review of Cross-national Comparative Studies of Homicide.” In
Homicide: A Sourcebook for Social Research, edited by M. D. Smith and M. A. Zahn, 125–145.
Thousand Oaks, CA: Sage, 1999.
LaFree, G. “Evidence for Elite Convergence in Cross-national Homicide Victimisation Trends, 1956
to 2000.” The Sociological Quarterly 46, no. 1 (2005): 191–211. doi:10.1111/j.1533-
8525.2005.00009.x.
LaFree, G., and A. Tseloni. “Democracy and Crime: A Multilevel Analysis of Homicide Trends in Forty-
four Countries, 1950-2000.” The ANNALS of the American Academy of Political Social Science 605,
no. 1 (2006): 25–49. doi:10.1177/0002716206287169.
LaFree, G., K. Curtis, and D. McDowall. “How Effective are Our “Better Angels?” Evidence for a World-
wide Decline in Violent Crime since the 1990s.” European Journal of Criminology 12, no. 4 (2015):
482–504. doi:10.1177/1477370815584261.
Land, K. C., P. L. McCall, and L. E. Cohen. “Structural Covariates of Homicide Rates: Are There Any
Invariances across Time and Social Space?.” The American Journal of Sociology 95, no. 4 (1990):
922–963. doi:10.1086/229381.
Lappi-Seppälä, T., and M. Lehti. “Cross-comparative Perspectives on Global Homicide Trends.” Crime
and Justice 43, no. 1 (2014): 135–230. doi:10.1086/677979.
Levitt, S. D. “The Effect of Prison Population Size on Crime Rates: Evidence from Prison
Overcrowding Legislation.” The Quarterly Journal of Economics 111, no. 2 (1996): 319–351.
doi:10.2307/2946681.
Levitt, S. D. “Understanding Why Crime Fell in the 1990s: Four Factors that Explain the Decline and
Six that Do Not.” Journal of Economic Perspectives 18, no. 1 (2004): 163–190. doi:10.1257/
089533004773563485.
Lott, J. R., Jr. More Guns, Less Crime: Understanding Crime and Gun Control Laws. 3rd Ed. Chicago, IL:
University of Chicago Press, 2013.
Marx, K. Capital: Volume 1. New York: Vintage Books, 1977.
McCall, P. L., and J. R. Brauer. “Social Welfare Support and Homicide: Longitudinal Analyses of
European Countries from 1994 to 2010.” Social Science 48 (2014): 90–107.
McCall, P. L., K. C. Land, and K. F. Parker. “Heterogeneity in the Rise and Decline of City-level
Homicide Rates, 1976-2005: A Latent Trajectory Analysis.” Social Science Research 40, no. 1
(2011): 363–378. doi:10.1016/j.ssresearch.2010.09.007.
McCall, P. L., K. F. Parker, and J. M. McDonald. “The Dynamic Relationship between Homicide Rates
and Social, Economic, and Political Factors from 1970 to 2000.” Social Science Research 37, no. 3
(2008): 721–735. doi:10.1016/j.ssresearch.2007.09.007.
Messner, S. F., B. Pearson-Nelson, L. E. Raffalovich, and Z. Miner. “Cross-national Homicide Trends in
the Latter Decades of the Twentieth Century: Losses and Gains in Institutional Control.” In Control
of Violence: Historical and International Perspectives on Violence in Modern Societies, edited by W.
Heitmeyer, H. G. Haupt, A. Kirschner, and S. Malthauner, 65–90. New York: Springer, 2011.
Messner, S. F. “When West Meets East: Generalising Theory and Expanding the Conceptual Toolkit of
Criminology.” Asian Journal of Criminology 10, no. 2 (2015): 117–129. doi:10.1007/s11417-014-9197-3.
Messner, S. F., L. E. Raffalovich, and G. M. Sutton. “Poverty, Infant Mortality, and Homicide Rates in
Cross-national Perspective: Assessments of Criterion and Construct Validity.” Criminology 48, no. 2
(2010): 509–537. doi:10.1111/j.1745-9125.2010.00194.x.
Messner, S. F., and R. Rosenfeld. Crime and the American Dream. Belmont, CA: Wadsworth, 1994.
Messner, S. F., and R. Rosenfeld. “Political Restraint of the Market and Levels of Criminal Homicide: A Cross
National Application of Institutional-anomie Theory.” Social Forces 75, no. 4 (1997): 1393–1416.
 GLOBAL CRIME 23

Moffitt, T. E. “Adolescent-limited and Life-course-persistent Antisocial Behavior: A Developmental

Taxonomy.” Psychological Review 100, no. 4 (1993): 674–701. doi:10.1037/0033-295X.100.4.674.
Nagin, D. S., and K. C. Land. “Age, Criminal Careers, and Population Heterogeneity: Specification and
Estimation of a Nonparametric, Mixed Poisson Model.” Criminology 31, no. 3 (1993): 327–362.
doi:10.1111/j.1745-9125.1993.tb01133.x.
Neumayer, E. “Good Policy Can Lower Violent Crime: Evidence from a Cross-national Panel of
Homicide Rates, 1980-97.” Journal of Peace Research 40, no. 6 (2003): 619–640. doi:10.1177/
00223433030406001.
Nevin, R. “How Lead Exposure Relates to Temporal Changes in IQ, Violence Crime, and Unwed
Pregnancy.” Environmental Research 83, no. 1 (2000): 1–22. doi:10.1006/enrs.1999.4045.
Nivette, A. E. “Cross-national Predictors of Crime: A Meta-analysis.” Homicide Studies 15, no. 2 (2011):
103–131. doi:10.1177/1088767911406397.
Nivette, A. E., and M. Eisner. “Do Legitimate Polities Have Fewer Homicides? A Cross-national
Analysis.” Homicide Studies 17, no. 1 (2013): 3–26. doi:10.1177/1088767912452131.
Ouimet, M. “A World of Homicides: The Effect of Economic Development, Income Inequality, and
Excess Infant Mortality on the Homicide Rate for 165 Countries in 2010.” Homicide Studies 16, no. 3
(2012): 238–258. doi:10.1177/1088767912442500.
Pare, P. P. “Indicators of Police Performance and Their Relationships with Homicide Rates across 77
Nations.” International Criminal Justice Review 24, no. 3 (2014): 254–270. doi:10.1177/
1057567714548453.
Pare, P. P., and R. Felson. “Income Inequality, Poverty and Crime across Nations.” The British Journal
of Sociology 65, no. 3 (2014): 434–458. doi:10.1111/1468-4446.12083.
Paternoster, R., R. Brame, P. Mazerolle, and A. Piquero. “Using the Correct Statistical Test for the
Equality of Regression Coefficients.” Criminology 36, no. 4 (1998): 859–866. doi:10.1111/j.1745-
9125.1998.tb01268.x.
Peres, M. F. T., and A. Nivette. “Social Disorganisation and Homicide Mortality Rate Trajectories in
Brazil between 1991 and 2010.” Social Science & Medicine 190 (2017): 92–100. doi:10.1016/j.
socscimed.2017.08.013.
Phillips, J. A. “Explaining Discrepant Findings in Cross-sectional and Longitudinal Analyses: An
Application to U.S. Homicide Rates.” Social Science Research 35, no. 4 (2006): 948–974.
doi:10.1016/j.ssresearch.2005.07.002.
Pratt, T. C., and T. W. Godsey. “Social Support, Inequality, and Homicide: A Cross-national Test of an
Integrated Theoretical Model.” Criminology 41, no. 3 (2003): 611–643. doi:10.1111/j.1745-
9125.2003.tb00999.x.
Pridemore, W. A. “A Methodological Addition to the Cross-national Empirical Literature on Social
Structure and Homicide: A First Test of the Poverty-homicide Thesis.” Criminology 46, no. 1 (2008):
133–154. doi:10.1111/j.1745-9125.2008.00106.x.
Pridemore, W. A. “Poverty Matters: A Reassessment of the Inequality-homicide Relationship in Cross-
national Studies.” The British Journal of Criminology 51, no. 5 (2011): 739–772. doi:10.1093/bjc/azr019.
Raferty, A. E. “Bayesian Model Selection in Social Research.” Sociological Methodology 25 (1995): 111–
163. doi:10.2307/271063.
Rogers, M. L. “A Descriptive and Graphical Analysis of the (Lack Of) Association between Age and
Homicide Cross-nationally.” International Criminal Justice Review 24, no. 3 (2014): 235–253.
doi:10.1177/1057567714548192.
Rogers, M. L., and W. A. Pridemore. “Geographic and Temporal Variation in Cross-national Homicide
Victimisation Rates.” In The Handbook of Homicide, edited by F. Brookman, E. R. Maguire, and M.
Maguire, 20–43. Malden, MA: Wiley-Blackwell, 2017.
Rogers, M. L., and W. A. Pridemore. “The (Null) Effects of Percent Young on 15 to 24 Age-specific and
Male- and Female-specific Cross-national Homicide Rates.” Homicide Studies 20, no. 3 (2016): 257–
292. doi:10.1177/1088767915613105.
Rogers, M. L., and W. A. Pridemore. “Do National Homicide Rate Follow Supranational Trends?.”
Journal of Research in Crime and Delinquency 55, no. 6 (2018): 691–727. doi:10.1177/
0022427818785210.
24 J. TUTTLE ET AL.

Roldan, R. M., G. Fondevila, and E. García-Tejeda. “Female Homicide Victimisation in Mexico: A

Group-based Trajectory and Spatial Study.” Global Crime (2021). doi:10.1080/
17440572.2020.1869539.
Rosenfeld, R. “Crime and Inflation in Cross-national Perspective.” Crime and Justice 43, no. 1 (2014):
341–366. doi:10.1086/677665.
Rosenfeld, R., and S. F. Messner. “The Crime Drop in Comparative Perspective: The Impact of
Imprisonment on American and European Burglary Rates.” The British Journal of Sociology 60,
no. 3 (2009): 445–471. doi:10.1111/j.1468-4446.2009.01251.x.
Santos, M. R., A. Testa, L. C. Porter, and J. P. Lynch. “The Contribution of Age Structure to the
International Homicide Decline.” PLoS ONE 14, no. 10 (2019): 1–28.
Shelley, L. I. Crime and Modernisation: The Impacts of Industrialisation and Urbanisation on Crime.
Carbondale, IL: Southern Illinois University Press, 1981.
Skott, S. “Disaggregating Homicide: Changing Trends in Subtypes over Time.” Criminal Justice and
Behavior 46, no. 11 (2019): 1650–1668. doi:10.1177/0093854819858648.
Spelman, W. “The Limited Importance of Prison Expansion.” In The Crime Drop in America, edited by
A. B. Blumstein and J. Wallman, 97–129. New York: Cambridge University Press, 2006.
Stamatel, J. P. “Correlates of National-level Homicide Variation in Post-communist East-Central
Europe.” Social Forces 87, no. 3 (2009): 1423–1448. doi:10.1353/sof.0.0179.
Tuttle, J. “Increasing Incarceration Rates and the Homicide Decline among OECD Nations, 1993-
2005.” In Homicide: Risk Factors, Trends and Prevention Strategies, edited by D. Wattel, 37–64. New
York: Nova Science Publishers, 2020.
Tuttle, J., P. L. McCall, and K. C. Land. “Latent Trajectories of Cross-national Homicide Trends:
Structural Characteristics of Underlying Groups.” Homicide Studies 22, no. 4 (2018): 343–369.
doi:10.1177/1088767918774083.
Van Dijk, J., A. Tseloni, and G. Farrell, Eds. The International Crime Drop: New Directions in Research.
New York: Palgrave Macmillan, 2012.
Wallerstein, I. “The Rise and Future Demise of the World Capitalist System: Concepts for
Comparative Analysis.” Comparative Studies in Society and History 16, no. 4 (1974): 387–415.
doi:10.1017/S0010417500007520.
Weiss, D. B., M. R. Santos, A. Testa, and S. Kumar. “The 1990s Homicide Decline: A Western World or
International Phenomenon? A Research Note.” Homicide Studies 20, no. 4 (2016): 321–334.
doi:10.1177/1088767916634406.
Wirth, L. “Urbanism as a Way of Life.” American Journal of Sociology 44, no. 1 (1938): 1–24.
doi:10.1086/217913.
Zimring, F. E. The Great American Crime Decline. New York: Oxford University Press, 2006.
 GLOBAL CRIME 25

APPENDIX A. Data Sources and Measures

Sources
Links for the data sources follow: World Health Organisation (WHO) https://www.who.int/
healthinfo/statistics/mortality_rawdata/en/; United Nations Office on Drugs and Crime
(UNODC) http://www.unodc.org/unodc/en/data-and-analysis/United-Nations-Surveys-on-Crime-
Trends-and-the-Operations-of-Criminal-Justice-Systems.html; National Research Institute of
Legal Policy’s Comparative Homicide Time Series (NRILP-CHTS); World Bank http://databank.
worldbank.org/data/home.aspx; World Development Indicators (WDI) http://databank.world­
bank.org/data/reports.aspx?source = world-development-indicators; Health, Nutrition and
Population Statistics (HNPS) https://data.worldbank.org/data-catalog/health-nutrition-and-popu­
lation-statistics; and United Nations Statistics Division (UNSD) https://unstats.un.org/unsd/data­
bases.htm;.
Measures
Homicide rates as the total number of deaths by assault divided by the total population)*100,000;
(WHO, UNODC, & NRILP-CHTS); Population Total (WDI); GDP per capita in constant US dollars (WDI);
Note: Purchasing Power Parity (PPP) not used because its series did not begin until 1990; Infant
Mortality Rate per 1,000 live births (HNPS); Percent Males ages 15 to 29 (HNPS); Sex Ratio as ratio of
total male population to female population*100 (HNPS); and Percent Urban Population as percen­
tage of total population (WDI).

APPENDIX B. Change in Variables from 1989 to 2010 across Low, Medium, and High
Rate Groups (Means and Standard Deviations).
Low Low Medium Medium High High
1989 2010 1989 2010 1989 2010
Homicide Rate 1.81 .94 6.08 4.34 12.31 19.84
(1.39) (.47) (7.62) (2.58) (13.27) (13.89)
Population 20,820.62 23,598.17 22,845.01 27,605.91 23,340.17 27,737.59
(in thousands) (27,616.51) (30,346.11) (55,068.16) (68,323.39) (43,652.23) (51,719.55)
Percent Urban 73.81 77.72 56.89 61.82 55.27 58.67
(14.09) (13.98) (20.08) (21.45) (18.58) (20.82)
Sex Ratio 96.92 97.75 98.03 97.64 95.52 94.51
(4.12) (6.15) (3.40) (3.40) (5.34) (5.93)
Percent Males Aged 24.50 20.85 26.71 25.19 27.53 26.25
15–29 (1.92) (2.90) (3.34) (3.94) (2.86) (2.32)
Infant Mortality 13.25 4.97 33.59 16.06 30.24 15.36
(13.29) (3.79) (21.40) (10.55) (14.06) (6.91)
GDP per Capita 29,370.89 38,266.88 7,471.75 10,598.35 6,450.50 8,551.96
(15,400.62) (22,158.75) (10,843.12) (13,090.37) (5,639.89) (5,284.83)