Social Science Research 42 (2013) 584–595

Contents lists available at SciVerse ScienceDirect

Social Science Research

journal homepage: www.elsevier.com/locate/ssresearch

The effect of poverty and social protection on national homicide

rates: Direct and moderating effects q
Meghan L. Rogers ⇑, William Alex Pridemore
Indiana University, Department of Criminal Justice, Sycamore Hall 302, Bloomington, IN 47405, United States

a r t i c l e i n f o a b s t r a c t

Article history: Social protection is the ability of a government to insulate its citizens from the problems
Received 22 July 2012 associated with poverty and market forces that negatively affect their quality of life. Prior
Revised 8 November 2012 research shows that government policies that provide social protection moderate the influ-
Accepted 6 December 2012
ence of inequality on national homicide rates. Recent research, however, reveals a strong
Available online 20 December 2012
association between poverty and national homicide rates. Further, theory and evidence
suggest that social protection policies are meant to aid in providing a subsistence level
Keywords:
of living, and thus to alleviate the vagaries of poverty not inequality. To this point, however,
Homicide
Poverty
no studies have examined the potentially moderating effect of social protection on the
Social protection strength of the association between poverty and homicide rates cross-nationally. We do
Cross-national analysis so in the present study. Employing data for the year 2004 from a sample of 30 nations,
we estimate a series of weighted least squares regression models to test three hypotheses:
the association between poverty and homicide will remain significant and positive when
controlling for social protection, social protection will have a significant negative direct
effect on national homicide rates, and social protection will diminish the strength of the
poverty–homicide association. The results provided evidence supporting all three hypoth-
eses. We situate our findings in the cross-national empirical literature on social structure
and homicide and discuss our results in the theoretical context of social protection.
Ó 2012 Elsevier Inc. All rights reserved.

1. Introduction

Poverty is one of the most consistent predictors of homicide rates in the empirical literature on the structural covariates
of violent crime in the United States (Messner and Rosenfeld, 1999; Pridemore, 2002; Sampson and Lauritsen, 1994). After a
long absence in the analogous cross-national literature, a small number of recent studies have shown a positive and signif-
icant association between poverty and national homicide rates (Paré, 2006; Pridemore, 2008, 2011). Prior to this recent re-
search on poverty, inequality had been the focus of many cross-national studies of homicide (for reviews see LaFree, 1999;
Messner and Rosenfeld, 2006). As such, several studies that examined the potential buffering effects of social protection and
related phenomena like decommodification (Esping-Andersen, 1990) concentrated on the ability of these constructs to mod-
erate the effects of inequality on national homicide rates (Messner and Rosenfeld, 2006; Pratt and Godsey, 2002; Savolainen,
2000). While empirical results largely support this hypothesis, theoretical questions remain about the inequality-homicide
association and exactly what social protection policies are meant to alleviate.

q
We thank Mitch Chamlin and Rick Rosenfeld for their helpful critiques of an earlier draft of this manuscript. An earlier version of this paper was
presented at the 2011 annual meetings of the American Society of Criminology.
⇑ Corresponding author.
E-mail addresses: meglroge@indiana.edu (M.L. Rogers), wpridemo@indiana.edu (W.A. Pridemore).

0049-089X/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ssresearch.2012.12.005
 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595 585

Following the lead of earlier researchers (Conway and Norton, 2002; Paré, 2006; Pridemore, 2008, 2011), we argue that
the economic components of policies aimed at social protection, social and economic welfare, and decommodification are
directed at and meant to support those living in or near poverty in an attempt to provide at least a subsistence minimum
level of living. An indirect consequence of such support may be to reduce inequality, but its direct aim is to decrease the
social and economic harms caused by poverty. Therefore, we should not only expect a negative direct effect of social support
on homicide rates, but we would expect that the strength of the positive association between poverty and homicide rates
will be weaker in nations that offer greater social protection to its most vulnerable citizens. The poverty–homicide associ-
ation has been tested in only a handful of recent cross-national studies, and the moderating effects of social protection on the
association between poverty and homicide have yet to be assessed. Thus, our study (1) adds to the small but growing empir-
ical literature gauging if poverty matters for the variation in cross-national homicide rates and (2) tests for the first time if
any poverty–homicide association is moderated by social protection.

2. Literature review

2.1. Poverty and cross-national homicide rates

Theoretical explanations for an association between poverty and homicide at the cross-national level may be drawn from
multiple social structure theories, especially those situated within the anomie tradition. For example, Institutional Anomie
Theory (IAT) makes the claim that, taken on its own, poverty cannot independently explain differences in crime rates across
nations (Messner and Rosenfeld, 2006). Instead, it could be theorized that high poverty rates are symptoms of institutional
imbalances within nations. If high decommodification scores are indicative of a nation that maintains a balanced institu-
tional structure (Messner and Rosenfeld, 2006; Savolainen, 2000), and weak decommodification scores are symptomatic
of nations that have an unbalanced institutional structure where the economic institution overpowers other institutions,
then it seems logical high poverty rates would also be a symptom of an overemphasized economic institution with other
social institutions weakened by this overemphasis on economic goals.
In addition, Paré (2006) proposed that one of the possible explanations for the relationship between poverty and crime
could be understood through social disorganization theory. A socially disorganized community is a community that is
impoverished and has high levels of ethnic heterogeneity, residential instability, and family disruption (Bursik, 1988; Samp-
son and Groves, 1989), resulting in the inability of the community to exert informal social control over the behavior of its
members. At the cross-national level, it is suggested that there is an aggregating effect, meaning that on average those na-
tions with a greater number or proportion of disorganized neighborhoods and/or neighborhoods with deeper levels of dis-
organization than other nations will exhibit higher crime rates.
Despite these and other theoretical explanations that lead us to expect an association between poverty and homicide
rates, until recently a formal test of the poverty–homicide thesis was missing from the cross-national homicide literature
(Paré, 2006; Pridemore, 2008, 2011). The lack of research regarding poverty and homicide rates at the cross-national level
was intriguing because of the consistent relationship between poverty and homicide within the US literature (Pridemore,
2002; Sampson and Lauritsen, 1994). Utilizing various indicators of poverty (e.g., infant mortality rates and the UN’s Human
Poverty Index), the poverty–homicide thesis has recently been tested multiple times at the cross-national level, with results
providing consistent evidence of an association across different samples, different types of crime, different data sources, over
time, and with the inclusion of different control variables (Paré, 2006; Pridemore, 2008, 2011; see also Messner et al., 2010).
In one test, Pridemore (2008) found a poverty–homicide association when the usual structural covariates (including inequal-
ity) were included in the model, but found no association between inequality and national homicide rates. In another test,
the relationship between poverty and homicide remained significant even when measures of economic wellbeing (e.g.,
unemployment) and income inequality were included (Paré, 2006), though again inequality was not significant. In a
time-series test of the poverty–homicide thesis undertaken on a sample restricted to 16 developed nations, Messner et al.
(2010) found that when absolute and relative deprivation were included in the same model, both were associated with
homicide rates. These findings were with a relatively homogeneous sample of nations that also have low poverty rates.
Therefore, it appears that even within nations with low poverty, poverty still has an association with homicide rates. Finally,
Pridemore (2011) replicated two previously published studies (Fajnzylber et al., 2002; Savolainen, 2000) that had shown an
inequality-homicide association at the cross-national level. In two of three models (Savolainen had disaggregated homicide
victimization rates by sex), the inequality-homicide association disappeared when a proxy for poverty was added to the ori-
ginal models, though the poverty–homicide association was significant in all models.

2.2. Social protection

Other theories postulate that cultural values create variation in the feelings of obligation to the community, and this in
turn influences crime at the macro-level (Currie, 1997). Currie (1997) argues that market societies contribute to high crime
rates across multiple nations because they offer little social protection from changes in the economy (Currie, 1997, p.152).
According to Currie (1997), seven criminogenic mechanisms can be observed with the rise of a market society, including the
destruction of livelihood, economic inequality and material deprivation, a reduction and in some cases removal of public ser-
586 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595

vices, and the birth of a ‘‘hard culture’’ that focuses on materialistic gain. He argues that market societies devalue labor and
attempt to decrease or eradicate labor costs from production of goods and services. The devaluing of labor in conjunction
with the decrease in social protection from the government creates conditions that could result in higher levels of poverty
and economic inequality. These higher levels of poverty should exist within a society that also is focused solely on materi-
alistic gain or private gain. These societies leave individuals with little recourse to deal with market changes that affect them,
their family, and their community. The weakening of social controls, both formal and informal, further compounds these
conditions.
Currie (1997) also proposed a connection between the lack of social services and supports, which is one symptom of a
market society, and increased crime rates in market societies. By providing social protection, nations may (1) help prevent
the values and ethics of a market society from overrunning their social and political alternatives to the market structure and
(2) temper the development of a materialistic ‘‘hard culture.’’ By providing protection against the devaluation of labor and
the effect it has on the labor market, social protection could also reduce the economic inequality and material deprivation
that exist in a market society (Currie, 1997). All of these possibilities would decrease the influence of the mechanism that
Currie (1997) theorizes explains differences in crime rates across nations.
Within the criminological literature, there has been considerable research devoted to understanding the relationship be-
tween homicide and social protection, both within the United States and utilizing cross-national samples. Despite the mea-
sure of social welfare or social protection used, the results have shown a consistent significant negative relationship between
social welfare and homicide. Within the United States, for example, the significant negative relationship between social pro-
tection and homicide has been found across standard metropolitan statistical areas (DeFronzo, 1983; Messner, 1986), cities
(Sampson, 1987), and even a specific city over time (Chamlin et al., 2002). Cross-nationally, the research has found signifi-
cant negative relationships between social protection and homicide utilizing different measures of welfare spending, includ-
ing a decommodification index (Messner and Rosenfeld, 2006; Savolainen, 2000), percent GDP spent on health care and
education (Pratt and Godsey, 2002), welfare spending (Gartner et al., 1990; Pratt and Godsey, 2002), and the amount of social
welfare spent in US dollars for each nation (Savage et al., 2008).
Two studies that specifically examined the association between social protection and homicide included a measure of
poverty (Paré, 2006; Pridemore, 2011). Both studies found a significant direct negative effect of social protection on homi-
cide. Pridemore (2011, p. 764) stated ‘‘nations that score highly on decommodification are better at providing support for
citizens in the lowest economic strata, thereby reducing poverty.’’ In addition to the findings of the decommodification effect
on homicide, Paré (2006) found that the negative association between homicide and social welfare was likely the result of
the intended effect of social protection, reducing poverty by providing additional assistance to citizens that live in poverty.

2.3. The moderating effects of social protection

Outside of the criminological research, social protection is conceptualized as a way to reduce poverty or provide those in
poverty with a safety net (or the newer conceptualization, a trampoline). The safety net is meant to catch those who fall be-
low the standard of living and provide them with a buffer from changes in the economy and adverse events (Conway and
Norton, 2002). In recent years, researchers and policy makers have attempted to rebrand social protection into a focus on
the risk and vulnerability that individuals face when they enter or continue to experience chronic poverty (Barrientos
et al., 2005). Social protection not only provides a higher standard of living for those in or near poverty, but it also provides
them with a buffer against changes in the economy and other adverse circumstances that threaten their standard of living,
even for those who are chronically living in poverty (Barrientos et al., 2005). Therefore, the effects of poverty are less caustic
when social protection is provided. Social protection has been found to aid in the reduction of poverty at the national level.
According to Nolan et al. (2002), poverty would have been much higher within Denmark, France, Germany, Ireland, Sweden
and the United Kingdom without some form of social protection. Thus, social protections may moderate the effects of pov-
erty on homicide at the national or individual levels (Barrientos et al., 2005; Devereux, 2001; Nolan et al., 2002).
As discussed previously, theory argues that social disorganization results chiefly from family disruption, residential
mobility, ethnic heterogeneity, and poverty. Social protection – which can include protections such as old age, survivors,
incapacity, active labor market, family, housing, and unemployment benefits, both in cash and in kind (Adema and Ladaique,
2009) – may hedge against these elements of social disorganization. For example, it may be that social protection in the form
of family assistance, housing assistance, unemployment and active labor market protections helps reduce family disruption,
residential mobility, and poverty. Family disruption often results in the household entering poverty (Sampson and Groves,
1989), but assistance could either prevent family disruption or prevent the entrance into poverty by the household because
of the loss of income. Housing assistance may moderate the effect of residential mobility, as this assistance allows families to
remain in their homes or stay in a residential location longer then they may have been able to on their own income. Finally,
unemployment benefits and active labor market assistance may provide a living wage, so that even if an individual or a
neighborhood has a high poverty level, the assistance may aid in raising the standard of living or staving off the effects
on crime of concentrated poverty.
If the theoretical argument in which poverty is associated with homicide is based within the arguments made by IAT, then
it is also logical that social protection would moderate the relationship of poverty and homicide. As discussed earlier, IAT
argues that poverty, as an independent predictor, is unable to account for differences in homicide rates across nations (Mess-
ner and Rosenfeld, 2006). However, poverty may be a symptom of institutional imbalance. If so, social protection may help to
 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595 587

restore institutional balance. According to Messner and Rosenfeld (2006), the level of a nation’s decommodification index,
which includes social protection measures that are purely based on governmental or public expenditures, is indicative of
a balance of institutions within nations. Therefore, social protection (a more inclusive measure of assistance for those
who live in poverty) may also be indicative of a balanced system of institutions within nations with high social protection
expenditures.
Finally, if the mechanism through which poverty is related to differences in homicide rates across nations is through the
market forces discussed by Currie (1997), then we would also expect social protection to condition the association between
poverty and homicide. Specifically, Currie (1997) argues that nations that provide more social protection to their citizens are
more likely to have an altruistic nature and less likely to be focused solely on materialistic or private gain. Therefore, changes
in the market will have less of an effect on families and communities because a social safety net is provided through social
protection. This social safety net, or social protection, should then be associated with lower homicide rates.
If poverty and homicide are significantly associated, it is logical that the policy utilized to reduce poverty could also have
a significant relationship with homicide. If social protection is found to reduce poverty, and poverty is significantly related to
homicide, then social protection should also reduce homicide whether through direct effects or moderating effects. In spite
of this, an empirical exploration of the moderating effects of social protection on the association between poverty and homi-
cide is missing from the cross-national homicide literature. Cross-national research exploring the moderating effect of social
protection has focused exclusively on the inequality-homicide association, and the results have been mixed (Bjerregaard and
Cochran, 2008; Messner and Rosenfeld, 1999; Pratt and Godsey, 2003; Savolainen, 2000). Further, Batton and Jensen (2002)
did not find that decommodification moderated inequality’s association with homicide in the United States.
In the face of these inconsistent results, and given the recent literature that has shown poverty and homicide to be related
at the cross-national level (Paré, 2006; Pridemore, 2008, 2011), it makes sense to test the unexamined hypothesis that social
protection moderates the association between poverty and national homicide rates, especially given the theoretical mech-
anism through which social protection is expected to influence violence rates. Just as importantly, the moderating effect of
social protection on the association between poverty and homicide may be understood as a function of what social protec-
tion is meant and found to accomplish, poverty reduction (Conway and Norton, 2002). Thus, it could be theorized that social
protection’s association with homicide is most likely and naturally a moderating effect, as most forms of social protection act
as a ‘‘buffer’’ for citizens that live in poverty. Social protection aids in providing the basic necessities required for day to day
living (Conway and Norton, 2002). As a buffer, it would likely indicate a moderating effect.
Social protection can also have a direct effect on national homicide rates. According to Cullen et al. (1999, p. 190), social
protection ‘‘is the process of transmitting various forms of capital—human, cultural, social and material.’’ Social protection’s
effect on crime has been perceived as stemming variously from increasing the ability of the family to regulate behaviors,
aiding in prosocial reactions to negative stimuli, aiding in early desistance from crime across the life course, and acting as
a social control (Pratt and Godsey, 2002).

2.4. Hypotheses

The literature review logically provides three hypotheses. First, we hypothesize that poverty will be positively associated
with national homicide rates, net of the effects of the other common structural covariates in the model. Second, social pro-
tection will be negatively associated with national homicide rates net of controls. Third, we hypothesize that social protec-
tion will moderate the relationship between poverty and homicide rates. Specifically, we expect the strength of the positive
poverty–homicide association to be weaker in nations with greater levels of social protection.

3. Data and method

3.1. Data

To test the hypotheses, we gathered data for 30 nations. As discussed below, the sample size was limited to those nations
for which the key measure of social support was available. Unless otherwise stated below, all data were for the year 2004.
Appendix A contains a list of nations employed in the analysis.

3.1.1. Dependent variable

The dependent variable was the number of homicide victimizations per 100,000 residents for each of the 30 nations. The
homicide data were obtained from the World Health Organization’s (WHO) WHO mortality and burden of disease estimates for
WHO member states in 2004 (World Health Organization, 2009).
The WHO homicide data were utilized instead of alternative sources because WHO enforces a uniform definition of homi-
cide. Alternative sources allow for differences in the case definition of homicide, resulting in estimates that may not be com-
parable across nations due to which violent events are or are not defined as homicide in a given nation (Smit et al., 2012).
WHO’s uniform definition of homicide is ‘‘Homicides and injury purposely inflicted by another person,’’ which includes the
International Classification of Diseases 10th Revision categories X85-Y09 (World Health Organization, 2009).
588 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595

We chose 2004 because of data availability. Although an analysis that includes more years may be preferred, it is not pos-
sible to conduct a cross-sectional time series analysis due to the lack of data both in the dependent and predictor variables
across time and nations. Further, while our sample is small, it is not uncommon for such small sample sizes to be found in the
cross-national empirical literature on homicide (e.g., Messner et al., 2010, used a sample of 16; Rosenfeld and Messner
(2009) used a sample of 10; Savolainen (2000) used a sample of 32).

3.1.2. Predictor variables

Two main predictor variables were used to test the hypotheses. First, infant mortality was used as a proxy for poverty.
Unfortunately, at the cross-national level there is no single agreed upon measure of poverty (Paré, 2006; Pridemore,
2008). Infant mortality has been found to be a strong correlate of poverty, and is used in other fields of research (Antonovsky
and Bernstein, 1977; Firebaugh and Beck, 1994; Ross, 2006). In addition, infant mortality has been used as an indicator of
poverty both in cross-national research (Frey and Field, 2000; Messner, 1989; Messner et al., 2010; Paré, 2006; Pridemore,
2008 and Pridemore, 2011) and by criminologists in research focusing on the United States (Loftin and Parker, 1985; McDo-
wall, 1986), and Messner et al. (2010) used alternative measures of absolute and relative poverty to examine the criterion
validity of infant mortality as a proxy. Infant mortality is defined as the number of infant deaths (under the age of 1) per
1000 live births. The infant mortality scores were obtained for the year 2004 from the Organization of Economic Cooperation
and Development (OECD, 2011). South Korea had missing data for infant mortality in the year 2004. Since the 2003 infant
mortality rate was also missing, we substituted the infant mortality rate from South Korea for 2005.
The second predictor variable was social protection. In this study, social protection was defined as the proportion of the
overall GDP for each nation spent on benefits associated with old age, survivors, incapacitation, health, family, active labor
market, unemployment, housing assistance, and other social policies (e.g., food subsidies) (Adema and Ladaique, 2009, p. 10).
This is a much more comprehensive measure of social support than used in prior studies of cross-national homicide rates.
The social protection provided by the Organization of Economic Cooperation and Development (OECD) is an index of the dif-
ferent forms of social protection that are available both through public and private resources within nations (Adema and
Ladaique, 2009). Nations provide information on public and private expenditures on the different forms of social protection.
The OECD gathers all of the social protection spending data and provides a unified measure of social protection across na-
tions. For the purpose of this paper, we chose to utilize the amount spent on these social protections as a proportion of the
overall GDP (OECD, 2011). Percent GDP was chosen as opposed to other measures because it is a commonly understood mea-
sure and is available for most nations in our sample for the year 2004. The value on this measure from Turkey was missing
for the year 2004. Because the 2003 data for Turkey on this measure were also missing, we substituted data from 2005.

3.1.3. Control variables

Five other variables were included in the models as controls. The first was income inequality, measured by the Gini coef-
ficient and obtained from the OECD (OECD, 2011). The education index component of the Human Development Index also
was included as a control variable. This measure was obtained from the Human Development Report published by the United
Nations (United Nations, 2011). The education index is a combination of the mean years of school for adults age 25 (obtained
through estimates from national censuses from UNESCO), and the expected years schooling for children within the nation
(which is capped at 18 years). The percent of the population that lives in an urban area, the sex ratio (i.e., the number of
males per females), and the percent of the entire population that is male and aged 15–24 were also included. These variables
were obtained from the World Bank (World Bank, 2011). These control variables were included to make the model similar to
previous research exploring the relationship between poverty and homicide (Pridemore, 2008, 2011), and these five vari-
ables are also commonly included in other cross-national studies of social structure and homicide.

3.2. Method

Several of our measures – homicide, poverty, inequality, education, and young males – were highly skewed, with skew
statistics greater than twice their standard errors. To address this, we began by taking the natural logarithm of each. Nor-
mality was estimated utilizing the Shapiro–Wilk test for normality. The logarithmic transformations resulted in normal dis-
tributions for homicide, income inequality, and percent young males, but not for poverty and education. Poverty did not
approximate a normal distribution due to possible outliers. Therefore, we conducted a sensitivity test on the model, exclud-
ing the nations with the outliers (Mexico and Turkey). The inferences drawn were the same as those reported in Table 2.
The populations of the nations included in our sample varied widely, ranging from less than 300,000 in Iceland to nearly
300,000,000 in the United States. This raises the likelihood that the variance of the prediction errors would not be constant,
violating an important regression assumption. Therefore, to test our hypotheses we used weighted least squares (WLS)
regression, weighting each case by the square root of the population size, which reduces heteroskedasticity in the error var-
iance (Hanushek and Jackson, 1977, pp. 143, 152). Further, we tested the assumption of constant variance with the Breusch-
Pagan and Cook-Weisberg tests for heteroskedasticity. When the original values for homicide were used as the dependent
variable the null hypothesis was rejected, meaning the errors were heteroskedastic. However, transforming the rates via the
natural log, as mentioned above, generated a model in which the prediction errors were homoscedastic. When we used the
logarithmically transformed rates and the OLS estimator, the conclusions drawn were the same as those using the WLS
estimator.
 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595 589

Table 1
Descriptive statistics and correlation matrix (n = 30).

1 2 3 4 5 6 7 8
1. LN homicide
2. LN poverty .68
3. Social protection .54 .61
4. LN inequality .54 .68 .64
5. LN education .45 .79 .48 .54
6. % Urban .01 .15 .20 .27 .27
7. M–F sex ratio .00 .16 .37 .07 .09 .35
8. LN males 15–24 .59 .76 .76 .59 .59 .27 .28
Mean 1.78 5.59 20.41 0.31 0.96 74.06 0.97 0.07
SD 1.74 4.46 5.97 0.05 0.04 11.10 0.02 0.01

Note: Means and standard deviations are for original values before logarithmic transformations.

Table 2
Results of weighted least squares model estimation (n = 30).

Model 1: Direct effects Model 2: Reduced direct effects Model 3: Interaction effect
b (b) SE p-value b (b) SE p-value b (b) SE p-value
Intercept 9.214 4.767 0.034 8.169 4.553 0.043 11.992 4.378 0.006
LN poverty 1.307 (0.91) 0.503 0.009 1.294 ( 0.90) 0.325 <0.001 1.023 ( 0.71) 0.335 0.003
Social protection 0.042 ( 0.33) 0.023 0.042 0.047 ( 0.37) 0.02 0.014 0.053 ( 0.42) 0.019 0.056
LN inequality 0.624 (0.13) 0.851 0.236 0.79 ( 0.17) 0.788 0.163 0.646 (0.14) 0.744 0.198
LN education 7.820 (0.49) 3.113 0.010 8.183 ( 0.52) 2.996 0.006 13.34 (0.84) 3.817 0.001
% Urban 0.017 (0.20) 0.011 0.077 0.019 ( 0.22) 0.01 0.036 0.013 (0.15) 0.01 0.098
M–F sex ratio 11.703 ( 0.30) 5.469 0.022 9.353 ( 0.24) 5.006 0.037 11.839 ( 0.30) 4.868 0.012
% Young males 0.113 (0.17) 0.196 0.286
GDP per capita 0.000 (0.17) 0.000 0.133
Poverty  Social protection 0.007 ( 0.54) 0.038 0.029
Adj. R2 0.733 0.741 0.771

Note: Given theoretical expectations, p-values are for one-tailed tests.

Descriptive statistics are provided in Table 1. The means and standard deviations shown are for the original values of
these measures before some were logarithmically transformed as discussed above. The mean homicide victimization rate
among the sample was 1.78 per 100,000 residents. Mexico (9.28) had the highest homicide rate, followed by the United
States (5.94), and Turkey (2.91). Nations with the lowest homicide rates included Japan (0.54), Germany (0.69), Ireland
(0.73), and Austria (0.77). The highest values on the infant mortality rate (i.e., the proxy for poverty) were in Turkey (24.6
infant deaths per 1000 live births), Mexico (17.6), and Poland, the Slovak Republic, and the United States (all with 6.8). Japan
(2.8), Iceland (2.8), and Sweden (3.1) had the lowest values on this measure. Nations with the highest proportion of their GDP
spent on social protection were Sweden (29.5), France (29.0), Austria (27.7), and Denmark (27.7). The nations scoring the
lowest on the social protection measure were South Korea (6.0), Mexico (9.6), Turkey (9.9), the United States (15.9), and Ire-
land (16.0).
The correlation matrix is also shown in Table 1. Homicide rates were correlated with poverty at 0.68 and with social pro-
tection at 0.54. While inequality was significantly correlated with the homicide rate at 0.54, this was dependent upon three
influential cases: Mexico, Turkey, and the United States. When these three cases were dropped, the correlation dropped sub-
stantially and was non-significant. Similarly, while the education variable was significantly correlated with homicide at
0.45, the same three nations were also influential and the correlation became non-significant when they were dropped.
The results of model estimation are shown in Table 2. Model 1 shows the results for the direct effects model. It can be
observed that GDP per capita has little effect within the regression model (b = 0, p = 0.133). Therefore, the model does not
benefit from including GDP per capita in constant US dollars. In addition, we chose not to include GDP per capita because
(1) its inclusion seems redundant given that we include measures of both poverty and inequality and (2) the theoretical
mechanism associating the two is unclear (Pridemore, 2008). The percent of young males is also dropped from the reduced
model (Model 2, Table 2) due to high levels of multicollinearity (this is discussed further in Section 3.3 below).
Model 2 in Table 2 shows the results for the reduced model. As expected, poverty was positively and significantly asso-
ciated with homicide rates in this sample, and social protection was negatively and significantly associated with homicide
rates.
Model 3 in Table 2 shows the results for the interaction effect. The log of poverty and social protection were mean cen-
tered before creating the interaction term to avoid multicollinearity when estimating the model. First, we see that poverty
and social protection retain their significant direct effects on homicide rates in this model. Second, the interaction term con-
590 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595

Table 3
Full model with principal components (n = 30).

Model 4: Full direct model PCA Model 5: Full indirect model PCA Model 6: Full indirect model PCA
excluding % young
b (b) SE p-value b (b) SE p-value b (b) SE p-value
Intercept 0.188 1.536 0.452 0.171 1.568 0.457 0.498 1.170 0.337
LN poverty 1.452 (1.01) 0.506 0.005 1.416 (0.98) 0.530 0.007 1.289 (0.89) 0.353 0.001
Social protection 0.061 ( 0.49) 0.023 0.007 0.064 ( 0.50) 0.025 0.009 0.057 ( 0.45) 0.018 0.003
LN inequality 0.461 (0.10) 0.910 0.309 0.406 (0.09) 0.947 0.336 0.160 (0.03) 0.729 0.414
Component 1 0.312 (0.61) 0.111 0.005 0.326 (0.64) 0.124 0.008 0.264 (0.40) 0.069 0.001
Component 2 0.227 ( 0.34) 0.138 0.057 0.248 ( 0.37) 0.158 0.066 0.643 ( 0.99) 0.151 0.000
Component 3 0.211 (0.17) 0.146 0.081 0.208 (0.17) 0.149 0.088 0.039 ( 0.03) 0.129 0.382
Poverty  Social Protection 0.011 ( 0.08) 0.037 0.387 0.080 ( 0.56) 0.036 0.018
Adj. R2 0.686 0.673 0.790

Note: Given theoretical expectations, p-values are for one-tailed tests.

taining these two measures was negatively and significantly associated with homicide rates, providing support for our third
hypothesis. In other words, the strength of the association between poverty and homicide rates is weaker in nations that
provide greater social protection to their citizens.

3.3. Sensitivity analysis

We took several steps to ensure the stability of our results. First, as noted above, the variable measuring the proportion of
the entire population that was male and aged 15–24 was dropped from the analysis due to high levels of multicollinearity.
However, when it was included in both the direct effects and interaction models the inferences in both models for poverty,
social protection, and the interaction term remained the same as reported in Table 2. Similarly, the education measure was
also highly collinear with poverty, though not to the extent of the young males measure.1 Again, when it was excluded from
the two models the inferences for the key variables remained the same.2
Second, we examined leverage, residual, and dfbeta values for outlying and influential cases. Analyses such as ours that
are carried out on small samples can be especially sensitive to even an influential case or two. Leverage values suggested a
closer look at Japan, South Korea, Mexico, and especially Turkey. The inferences described above remained unchanged when
these individual cases were dropped. The studentized deleted residuals suggested a closer look at Finland, Sweden, and the
United States. Again, the inferences for the key variables described above remained the same when these cases were ex-
cluded from the analysis. Finally, standardized dfbetas suggested that Japan and the United States might be influential cases.
As just described, however, the exclusion of these nations did not result in changes to the inferences drawn about poverty,
social protection, and the interaction term.
Third, to reduce the regressor space and allow for the control variables to be included by reducing the correlation between
most of the control variables and the predictor variables, a principal component analysis was conducted utilizing the control
variables GDP, education, percent urban, percent of young males, and sex ratio. The eigenvalues, proportional and cumula-
tive proportion of the variance accounted for, and a screeplot all supported generating three principal components that ac-
counted for 87% of the variance in the control variables. Table B1 in Appendix B provides a table showing that the three
components all had eigenvalues greater than 1, except for the third that existed solely to account for percent urban. Table 3
shows the results of model estimation using this approach. With regards to the variables of interest in the direct effects mod-
el (poverty, social protection, and inequality), Table 3 shows that the same conclusions were drawn as shown in full direct
effects model in Table 2. The interaction effect model varied slightly, as can be observed in Model 5. The effect of the inter-
action became non-significant due to the correlation between the interaction term and percent young (r = 0.64), which re-
sults in correlations between the first two principal components and the interaction term. When percent young was
removed from the principal component analysis, and another principal component analysis was conducted (resulting again

1
This collinearity likely plays a role in the unexpected positive and significant association between education and homicide rates shown in both models. In
fact, when poverty was removed from Model 1, the variance inflation factor for the education variable was reduced substantially and the association between
education and homicide rate was non-significant.
2
The variance inflation factors (VIFs) for the variables in the direct effects model, including the percent of the population that is male and between the ages
15–24, are poverty = 12.40, percent young males = 6.77, education = 4.44, social protection = 3.56, inequality = 3.38, sex ratio = 1.80, and percent of the
population that lives in urban areas = 1.74. The VIFs for the model that excludes the percent of the population that is male and between 15 and 24 are
poverty = 5.69, education = 4.02, inequality = 2.83, sex ratio = 1.79, and percent urban = 1.50. The VIFs for the variables in the interaction effect model including
percent of the population that is male and between ages 15 and 24 are poverty = 13.52, interaction = 9.12, education = 7.79, young males = 6.77, social
protection = 3.63, inequality = 3.41, sex ratio = 1.92, and percent urban = 1.88. The VIFs for the variables in the interaction effect model excluding percent of the
population that is male and between ages 15 to 24 are interaction = 9.12, education = 7.37, poverty = 6.80, inequality = 3.14, social protection = 2.91, sex
ratio = 1.92, and percent urban = 1.64.
 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595 591

Table 4
Reduced model with principal components (n = 30).

Model 7: Reduced direct model PCA Model 8: Reduced indirect model PCA
b (b) SE p-value b (b) SE p-value
Intercept 0.042 1.139 0.486 0.465 1.081 0.336
LN poverty 1.371 (0.95) 0.294 <0.001 1.011 (0.70) 0.322 0.003
Social protection 0.043 ( 0.34) 0.019 0.015 0.054 ( 0.43) 0.018 0.004
LN inequality 0.842 (0.18) 0.772 0.143 0.637 (0.13) 0.727 0.195
Component 1 0.172 (0.21) 0.079 0.020 0.172 (0.21) 0.074 0.015
Component 2 0.434 (0.68) 0.111 0.001 0.617 (0.96) 0.134 0.000
Poverty  Social protection 0.074 ( 0.52) 0.034 0.022
Adj. R2 0.748 0.780

Note Given theoretical expectations, p-values are for one-tailed tests.

in three components), the moderating relationship was again significant (see Model 6 in Table 3), and the correlation be-
tween the principal components and the interaction effect was reduced.
An additional principal component analysis was estimated in order to reduce the number of variables within the
model and increase the regressor space for the reduced model. This PCA analysis included the education index, percent
urban, and sex ratio. We extracted two principal components that accounted for approximately 83% of the common var-
iance among the variables. Table B2 in Appendix B provides the eigenvalues and the proportional and cumulative pro-
portion of variance accounted for in each component. Regression based factor scores were then computed for each
nation. Table 4 includes the results of both the direct and interaction model when the principle components are submit-
ted for the control variables sex ratio, percent urban, and the education index. When the two factor scores were included
in the WLS estimation the inferences drawn were the same for both the direct and moderating effects models of the key
variables.

4. Discussion

The focus of most cross-national research on the structural covariates of homicide has been on inequality. As a re-
sult, most studies examining the moderating effects of social protection, social welfare, decommodification, and related
phenomena have focused on their ability to condition the effects of inequality on national homicide rates. However, a
small but growing number of recent cross-national studies have found an association between poverty and homicide
net of other structural covariates, including inequality (Paré, 2006; Pridemore, 2008 and Pridemore, 2011; see also
Messner et al., 2010). Our study first adds to this recent literature by examining the direct effects of poverty and social
protection on national homicide rates. Our results are consistent with those from these recent studies, showing poverty
to be positively and social protection negatively associated with homicide rates. Our study then extends beyond prior
work by testing if social protection moderates the effect of poverty on national homicide rates. Previous literature has
found that social protection moderates the effect of income inequality on national homicide rates. This is not simply
because income inequality and poverty are different theoretical concepts that may have different associations with
homicide, but more importantly because the goal of social protection is to alleviate poverty not inequality. Given the
latter, prior studies testing the moderating effects of social protection on the strength of the inequality-homicide
association may very well have been testing the wrong relationship. Our results revealed that the strength of the
positive association between poverty and homicide is weaker in nations that provide greater social protection to their
citizens.

4.1. Poverty and cross-national homicide rates

Our first hypothesis was that poverty would be positively associated with national homicide rates. This hypothesis was
based on recent cross-national findings (Paré, 2006; Pridemore, 2008 and 2011; see also Messner et al., 2010). In these recent
studies, the association between poverty and homicide was tested across different samples, model estimation techniques,
and measures of the key independent and control variables. In all instances, the significant positive association between pov-
erty and homicide remained consistent. In our tests, this hypothesis was supported: the association between poverty and
homicide remained significant and positive across all models. As can be observed from Table 2, the association remained
when social protection, income inequality, and other common structural covariates were included in the model. The signif-
icant positive association remained even when alternative model specifications were utilized to explore the sensitivity of our
results to multicollinearity, extreme values, and other threats.
The poverty–homicide thesis at the cross-national level is often situated within the social disorganization theoretical
explanation (Messner et al., 2010; Paré, 2006), as poverty is an important component of the social disorganization model.
Within cross-national research it could be theorized that the positive association between poverty and homicide is the result
592 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595

of social disorganization aggregating to the national level. This essentially means that nations with higher levels of poverty
possess more neighborhoods or cities that are socially disorganized or that are more deeply disorganized. The effect (in our
case, the homicide rate) in disorganized neighborhoods would also aggregate to the national level, and that is why we see
poverty is positively associated with homicide. This hypothesis has yet to be tested explicitly, however, probably due to the
lack of available data across nations that would allow for the exploration of social disorganization at the neighborhood or
city level.
Even given the consistent empirical findings, however, what is most apparent here is that the theory in this area is still
relatively weak and the theoretical mechanisms relating structural poverty to national homicide rates remain unclear. While
it is true that several theoretical traditions ‘‘claim’’ poverty, we are still uncertain about how exactly it leads to higher na-
tional homicide rates.

4.2. Social protection and national homicide rates: direct and moderating effects

Our second hypothesis was that there would be a significant negative association between social protection and na-
tional homicide rates. Our results provide support for this hypothesis across all models and alternative specifications.
Several related theories of social structure and crime provide potential explanations for this association. It could be
argued that nations that possess higher levels of social protection provide their citizens with protective barriers against
social and market forces, or even simply provide a higher standard of living for individuals (Chamlin et al., 2002;
Currie, 1997; Messner and Rosenfeld, 2006). The presence of this relationship and the way it is theorized to operate
provides further support to the idea that poverty is more important than inequality or GDP (Paré, 2006; Pridemore,
2011). That is, social protection is meant to provide help to those in need of obtaining the necessary resources for
day to day living (Adema and Ladaique, 2009; Conway and Norton, 2002), whether social protection is meant to pro-
vide protection in times of a decline in economic conditions or to provide a general minimum standard of living. Of
course, the direct effects of social protection shown here and elsewhere may also operate by providing human, cultural,
social, and material capital (Cullen et al., 1999) that reduce crime by strengthening the ability of the family to regulate
members’ behaviors, acting as a social control, and aiding in prosocial reactions to negative stimuli (Pratt and Godsey,
2002).
What remains to be answered is whether social protection conditions the poverty–homicide association. Our third
hypothesis was that the strength of the association between poverty and homicide would be weaker in nations that provide
greater social protection to their citizens. Our results provide support for this hypothesis. The results in Table 2 show that the
interaction term containing social protection and poverty was significant and negative. Again, this finding supports the state-
ment that poverty matters. One central aim of social protection is to provide assistance to those living in poverty (Adema and
Ladaique, 2009; Conway and Norton, 2002; Devereux, 2002). If social protection is having the intended effect, including pov-
erty reduction, then it is expected that social protection would not only have a direct effect on national homicide rates, but
would moderate the effect of poverty on homicide rates. A moderating effect of social protection on the relationship between
poverty and homicide means that nations that provide their citizens with higher levels of social protection should reduce the
effect of poverty on homicide. Therefore, social protection acts as a buffer against poverty, thereby reducing the negative
outcomes associated with it, including higher homicide rates.
The buffer could be conceptualized as a characteristic of nations that do not have unbalanced institutions, as preventing a
market society from reaching its full effect, or as limiting the effect of the precursors to social disorganization. As discussed in
the literature review, IAT proposes that nations with high decommodification index scores are less likely to have an over-
bearing economic institution that promotes success by any means necessary. To this point in the cross-national homicide
literature, however, researchers argued that decommodification was central to alleviating income inequality and have tested
that relationship (Messner and Rosenfeld, 2006; Savolainen, 2000). It may be that higher poverty rates and less social sup-
port are symptoms of institutional imbalance, which results in an ‘‘any means necessary’’ standard for economic success
(Merton, 1938) and thus high national crime rates. Thus, greater decommodification and social protection would serve to
moderate homicide rates via their impact on poverty reduction.
Currie (1997) provides a similar theoretical argument as IAT, though he is attempting to explain the variation in crime
across nations and not explain only the high violent crime rates in the United States compared to other nations. His idea
is that a market society creates conditions favorable to crime through the devaluation of labor, economic inequality and
material deprivation, withdrawal of public services, a ‘‘hard’’ culture, a decrease in informal and communal networks, and
weakening social and political alternatives to a market society. For each of these it could be argued that social protection
would help prevent or buffer the effect. By providing social protection a nation is able to buffer its citizens from the deval-
uation of labor. In addition, a nation is able to fight the consequences of the market by providing an alternative to a market
society via social protection, protecting against the withdrawal of public services.
For the social disorganization theoretical argument, it could be reasoned that the measure of social protection we
utilized could help moderate the effect of the precursors of social disorganization. Social protection provides protection
against most of the precursors to social disorganization. Housing assistance, for example, may reduce residential mobility
and thereby reduce social disorganization. Unemployment and active labor market assistance may help provide
living wages to those in poverty (Adema and Ladaique, 2009), and therefore potentially reduce poverty and possibly
family disruption, both of which are viewed as antecedents of social disorganization. These and other forms of social
 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595 593

protection were included in the measure of social protection utilized in our study, though this argument was not directly
tested.

4.3. Limitations

A few limitations must be considered when interpreting the results of our study. First, our sample contains nations
that vary widely in respect to population size. To reduce the potential threat this represents to a key regression assump-
tion, homoskedastic error variance, we utilized weighted least squares regression, weighting each nation by the square
root of its population size (for more information see Hanushek and Jackson, 1977). Second, the sample was relatively
small (N = 30). Unfortunately, small sample sizes are common to the empirical literature on social structure and
cross-national homicide rates given data availability. In our case, our sample size was restricted because only a small
number of nations provide reliable data for one of our key measures (i.e., social protection). Nevertheless, important
findings from cross-national studies have appeared in leading journals (Messner et al., 2010, N = 16; Rosenfeld and Mess-
ner, 2009, N = 10; Savolainen, 2000, N = 32) in spite of these small sample sizes, and our sample size was large enough to
discern significant associations on our variables of interest. Further, multiple post-estimation tests were conducted to
assess the sensitivity of our findings. Our exploration of leverage values, residuals, and dfbeta’s for outlying and influ-
ential cases revealed that our findings held even given the small sample size.
A third limitation was the use of infant mortality as a proxy measure for poverty. At the cross-national level, there is
not a single direct measure of poverty (Pridemore, 2008, 2011), which is especially problematic to gauge via an economic
measure for nations at differing levels of development. However, there is support for a strong correlation between poverty
and infant mortality, and infant mortality has been used in prior criminological studies as both a proxy and an instrumen-
tal variable for poverty rates (Loftin and Parker, 1985; McDowall, 1986; Messner, 1989; Messner et al., 2010; Paré, 2006;
Pridemore, 2008, 2011). Further, when alternative measures for poverty have been employed in recent cross-national re-
search on poverty and crime, the association held (Paré, 2006). In addition, some have questioned the use of infant mor-
tality as a proxy measure of poverty, despite the correlation, because there could be a confounding variable that explains
the correlation between poverty and infant mortality that would weaken the validity of infant mortality as a proxy for
poverty. Nevertheless, it will be important to future research to discover and use more direct measures of poverty
cross-nationally when examining crime and other negative outcomes.
The choice of the variables for this study was done with the goal of making our analyses similar to the previous empirical
literature exploring the association between poverty and homicide (see Pridemore, 2008, 2011; Pare, 2006). Future research
should increase the sample size, thereby allowing for more controls. Some of the controls that may be of interest in the fu-
ture are an index of modernity and a measure of ethnic heterogeneity. By including both a measure of modernity and ethnic
heterogeneity it is possible to test some of the theoretical speculations made in our discussion, specifically social disorgani-
zation theory. However, it should be noted that despite the lack of these controls the model still performs well, including a
high adjusted R2, meaning that a vast majority of the variance across national homicide rates is accounted for within the
reduced model presented within this paper.

5. Conclusion

This paper had three objectives. The first was to provide another test of the poverty–homicide thesis since this remains
relatively new to the cross-national literature. The second objective was to test if social protection has a significant negative
association with national homicide rates even when poverty is included in the model. Both of these variables had been
shown to be associated with national homicide rates in prior studies. Our goal was not simply provide another test, but
to set up our third objective. That final objective, and our key innovation and main contribution to the literature, was to
determine if the strength of the positive association between poverty and national homicide rates is moderated by social
protection. Employing data from 2004 on our sample of 30 nations, the results of weighted least squares model estimation
provided support for each of these hypotheses.
One key aspect of social protection is that it is meant to act as a safety net for citizens in times of economic downturn or
provide a better living standard for citizens who live below or near the poverty line. Ultimately, the aim of social protection is
not to reduce inequality but to address absolute deprivation by raising the standard of living of those who live in poverty to a
level that provides the minimum necessities to survive day to day. The moderating effect of social protection found within
this paper supports this view. We found that nations with higher levels of social protection not only have lower rates of
homicide, but also that the strength of the association between poverty and homicide in these nations is weaker. Therefore,
while there are many more direct goals of social protection, one important indirect effect of providing greater social protec-
tion is a reduction in violent crime.
If a key mechanism through which social protection reduces violent crime is poverty reduction, this provides further indi-
rect support that poverty matters. Nevertheless, what remains is the need for clearer theory about why poverty and national
homicide rates are associated. This will require more and careful theorizing together with continued tests of these theories
utilizing increasingly precise measures and sophisticated methods.
594 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595

Appendix A

List of nations used in analysis, and their scores on homicide rates, infant mortality rates, and social protection as percent
GDP.

Nation WHO homicide rates Poverty (infant mortality) Social protection (% GDP)
Australia 1.3 4.7 17
Austria 0.8 4.5 28
Belgium 1.6 3.8 26
Canada 1.4 5.3 17
Czech Rep 1.3 3.7 20
Denmark 1.0 4.4 28
Finland 2.6 3.3 26
France 0.8 4.0 29
Germany 0.7 4.1 27
Greece 0.9 4.1 20
Hungary 2.2 6.6 22
Iceland 1.0 2.8 17
Ireland 0.7 4.8 16
Italy 1.0 3.9 25
Japan 0.5 2.8 18
Korea 2.2 4.7 6
Luxembourg 1.2 3.9 24
Mexico 9.3 18.0 7
Netherlands 1.2 4.4 21
New Zealand 1.4 5.9 18
Norway 0.8 3.2 23
Poland 1.6 6.8 22
Portugal 1.7 3.8 22
Slovak Republic 2.0 6.8 16
Spain 1.4 4.0 21
Sweden 1.2 3.1 30
Switzerland 0.9 4.2 20
Turkey 2.9 25.0 10
United Kingdom 2.0 5.1 20
United States 5.9 6.8 16

Appendix B. Principal components

See Tables B1 and B2.

Table B1
Principal component analysis for control variables in full model.

Component Eigenvalues % of Variance Cumulative (%)

1 2.356 0.471 0.471
2 1.440 0.288 0.759
3 0.567 0.114 0.873
Principal component matrix
1 2 3
LN Education 0.501 0.268 0.072
% Urban 0.407 0.401 0.814
M–F ratio 0.083 0.765 0.347
GDP per capita 0.541 0.201 0.459
% Young 0.534 0.376 0.033
 M.L. Rogers, W.A. Pridemore / Social Science Research 42 (2013) 584–595 595

Table B2
Principal component analysis for control variables in reduced model.

Component Eigenvalues % of Variance Cumulative (%)

1 1.396 0.465 0.465
2 1.091 0.364 0.8288
Principal component matrix
1 2
LN education 0.372 0.794
% Urban 0.741 0.057
M–F ratio 0.559 0.605

References

Adema, W., Ladaique, M. 2009. How expensive is the welfare state? Gross and net indicators in the OECD Social Expenditure Database (SOCX). OECD Social,
Employment and Migration Working Papers. OECD Publishing. <http://ideas.repec.org/p/oec/elsaab/92-en.html> (Retrieved).
Antonovsky, A., Bernstein, J., 1977. Social class and infant mortality. Social Science and Medicine 11, 453–470.
Barrientos, A., Hulme, D., Shepherd, A., 2005. Can social protection tackle chronic poverty? The European Journal of Development Research 17, 8–23.
Batton, C., Jensen, G., 2002. Decommodification and homicide rates in the 20th-century United States. Homicide Studies 6, 6–38.
Bjerregaard, B., Cochran, J.K., 2008. A cross-national test of Institutional Anomie Theory: do the strength of other social institutions mediate or moderate the
effects of the economy on the rate of crime? Western Criminology Review 9, 31–48.
Bursik, R.J., 1988. Social disorganization and theories of crime and delinquency: problems and prospects. Criminology 26, 519–551.
Chamlin, M.B., Cochran, J.K., Lowenkamp, C.T., 2002. A longitudinal analysis of the welfare-homicide relationship: testing two (nonreductionist) macro-level
theories. Homicide Studies 6, 39–60.
Conway, T.I., Norton, A., 2002. Nets, ropes, ladders and trampolines: the place of social protection within current debates on poverty reduction.
Development and Policy Review 20, 533–540.
Cullen, F.T., Wright, J.P., Chamlin, M.B., 1999. Social support and social reform: a progressive crime control agenda. Crime and Delinquency 45, 188–207.
Currie, E., 1997. Market, crime and community: toward a mid-range theory of post industrial violence. Theoretical Criminology 1, 147–172.
DeFronzo, J., 1983. Economic assistance to impoverished Americans. Criminology 21, 119–136.
Devereux, S., 2002. Can social safety nets reduce chronic poverty? Development and Policy Review 20, 657–675.
Devereux, S., 2001. Social Protection for the Poor. IDS Working Paper 142. Brighton, Sussex.
Esping-Andersen, G., 1990. The Three Worlds of Welfare Capitalism. Polity Press, Cambridge.
Fajnzylber, P., Lederman, D., Loayza, N., 2002. Inequality and violent crime. Journal of Law and Economics 45, 1–39.
Firebaugh, G., Beck, F., 1994. Does economic growth benefit the masses: Growth, dependence, and welfare in the third world. American Sociological Review
59, 631–654.
Frey, R.S., Field, C., 2000. The determinants of infant mortality in less developed world: a cross-sectional test of five theories. Social Indicators Research 52,
215–235.
Gartner, R., Baker, K., Pampel, F.C., 1990. Gender stratification and the gender gap in homicide victimization. Social Problems 37, 593–612.
Hanushek, E.A., Jackson, J.E., 1977. Statistical Methods for Social Scientist. Academic Press, New York.
LaFree, G., 1999. A summary and review of cross-national comparative studies of homicide. In: Smith, M.D., Zahn, M.A. (Eds.), Homicide: A Sourcebook of
Social Research. Sage, Thousand Oaks, CA.
Loftin, C., Parker, R.N., 1985. An errors-in-variable model of the effect of poverty on urban homicide rates. Criminology 23, 269–287.
McDowall, D., 1986. Poverty and homicide in Detroit, 1926–1978. Violence and Victims 1, 23–34.
Merton, R.K., 1938. Social structure and anomie. American Sociological Review 3, 672.
Messner, S.F., 1986. Graphic mobility, governmental assistance to the poor, and rates of urban crime. Journal of Crime and Justice 9, 1–18.
Messner, S.F., 1989. Economic discrimination and societal homicide rates: further evidence on the cost of inequality. American Sociological Review 54, 597–611.
Messner, S.F., Raffalovich, L.E., Sutton, G.M., 2010. Poverty, infant mortality, and homicide rates in cross-national perspective: assessment of criterion and
construct validity. Criminology 48, 509–537.
Messner, S.F., Rosenfeld, R., 1999. Social structure and homicide: theory and research. In: Smith, M.D., Zahn, M.A. (Eds.), Homicide: A Sourcebook of Social
Research. Sage Publications, Thousand Oaks, CA.
Messner, S.F., Rosenfeld, R., 2006. Crime and the American Dream. Wadsworth.
Nolan, B., Hauser, R., Zoyem, J.-P., 2002. The changing effects of social protection on poverty. In: Gallie, D., Paugam, S. (Eds.), Welfare Regimes and the
Experience of Unemployment in Europe. Oxford University Press, New York, NY.
Organization of Economic Cooperation and Development, 2011. Statistical Portal. <http://www.oecd.org/statistics/> (Retrieved on 05.05.11).
Paré, P.-P. 2006. Income Inequality and Crime Across Nations Reexamined. (P.S.U. published doctoral dissertation). University Park, PA.
Pratt, T.C., Godsey, T.W., 2002. Social support and homicide: a cross-national test of an emerging criminological theory. Journal of Criminal Justice 30, 589–601.
Pratt, T.C., Godsey, T.W., 2003. Social support, inequality, and homicide: a cross-national test of an integrated theoretical model. Criminology 41, 611–644.
Pridemore, W.A., 2002. What we know about social structure and homicide: a review of the theoretical and empirical literature. Violence and Victims 17,
127–156.
Pridemore, W.A., 2008. A first test of the poverty–homicide hypothesis at the cross-national level. Criminology 46, 133–154.
Pridemore, W.A., 2011. Poverty matters: a reassessment of the inequality-homicide relationship in cross-national studies. British Journal of Criminology 51, 739–772.
Rosenfeld, R., Messner, S.F., 2009. The crime drop in comparative perspective: the impact of the economy and imprisonment on American and European
burglary rates. The British Journal of Sociology 60, 445–471.
Ross, M., 2006. Is democracy good for the poor? American Journal of Political Science 50, 860–874.
Sampson, R.J., 1987. Urban violence: the effect of male joblessness and family disruption. American Journal of Sociology 93, 348–382.
Sampson, R.J., Groves, W.B., 1989. Community structure and crime: testing social disorganization theory. American Journal of Sociology 94, 774–802.
Sampson, R.J., Lauritsen, J.L., 1994. Violent victimization and offending: individual-, situational-, and community-level risk factors. In: Reiss, A.J., Roth, J.A.,
(Eds.), Understanding and Preventing Violence. vol. 3: Social Influences. National Academy Press, Washington, DC.
Savage, J., Bennett, R.R., Danner, M., 2008. Economic assistance and crime: a cross-national investigation. European Journal of Criminology 5, 217–238.
Savolainen, J., 2000. Inequality, welfare state, and homicide: further support for the institutional anomie theory. Criminology 38, 1021–1042.
Smit, P.R., de Jong, R., Bijleveld, C., 2012. Homicide data in Europe: definitions, sources and statistics. In: Liem, M., Pridemore, W.A. Handbook of European
Homicide Research: Patterns, Explanations, and Country Studies. Springer, New York, NY.
United Nations. 2011. Human Development Report. <http://hdr.undp.org/en/statistics/data/> (Received on 05.05.11).
World Bank. (2011). World Bank Development Indicators. Retrieved on May 5, 2011 from http://data.worldbank.org/data-catalog.
World Health Organization. 2009. Mortality and Burden of Disease Estimates for WHO member States in 2004. <http://www.who.int/healthinfo/
global_burden_disease/en/> (Retrieved).