Economic correlates of violent death rates in forty countries, 1962–2008: A cross-typological analysis

Abstract

Objectives

Our goal was to identify if there might be advantages to combining two major public health concerns, i.e., homicides and suicides, in an analysis with well-established macro-level economic determinants, i.e., unemployment and inequality.

Methods

Mortality data, unemployment statistics, and inequality measures were obtained for 40 countries for the years 1962–2008. Rates of combined homicide and suicide, ratio of suicide to combined violent death, and ratio between homicide and suicide were graphed and analyzed. A fixed effects regression model was then performed for unemployment rates and Gini coefficients on homicide, suicide, and combined death rates.

Results

For a majority of nation states, suicide comprised a substantial proportion (mean 75.51%; range 0–99%) of the combined rate of homicide and suicide. When combined, a small but significant relationship emerged between logged Gini coefficient and combined death rates (0.0066, p < 0.05), suggesting that the combined rate improves the ability to detect a significant relationship when compared to either rate measurement alone. Results were duplicated by age group, whereby combining death rates into a single measure improved statistical power, provided that the association was strong.

Conclusions

Violent deaths, when combined, were associated with an increase in unemployment and an increase in Gini coefficient, creating a more robust variable. As the effects of macro-level factors (e.g., social and economic policies) on violent death rates in a population are shown to be more significant than those of micro-level influences (e.g., individual characteristics), these associations may be useful to discover. An expansion of socioeconomic variables and the inclusion of other forms of violence in future research could help elucidate long-term trends.

1. Introduction

Since the mid-twentieth century, as incidences and mortality of certain infectious diseases and non-communicable diseases were dramatically reduced through public health measures, homicide and suicide rose and remained among the top leading causes of death worldwide (Dahlberg & Mercy, 2009; Krug, Dahlberg, Mercy, Zwi, & Lozano, 2002). In the 1990s, leading global organizations were instrumental in establishing the importance of understanding violence within a public health framework. Due to a dramatic worldwide increase in violence, the 49th World Health Assembly (WHA, 1996) officially declared violence as a “leading worldwide public health problem,” setting the precedence to identify and understand the social, economic, health, and other environmental factors that could prevent violence in society.

The WHO emphasizes that an ecological point of view of different types of violence is an important step of the public health approach to preventing violence (Krug et al., 2002). Notably, homicide and suicide account for 31.3% and 49.1%, respectively, of the total number of violence-related deaths worldwide. A number of studies have established important environmental causes and risk factors of these two types of violence, especially with regard to measures of economic inequality. In a 1996 Lancet article, Martikainen and Valkonen (1996) showed, after controlling for age, education, occupational class, and marital status, that individuals who experienced unemployment had greater mortality than those in employment. An influential empirical analysis by Stuckler, Basu, Suhrcke, Coutts, and McKee (2009) found that, among 26 European nations from 1970 to 2007, every 1% increase in unemployment was associated with a 0.79% rise in suicides at ages younger than 65 years and with a 0.79% rise in homicides. Some focused on suicide, such as a panel data analysis of 15 European countries between 1970 and 1998 that found the impact of income inequality and unemployment to be unequal across age groups for suicide rates (Andres, 2005). Others focused on homicide, such as a cross-sectional study of 165 countries that showed economic development, inequality, and poverty to be significant predictors of homicide, although many variations in homicide for developing countries remain inadequately explained (Ouimet, 2012). The existing literature makes it clear that the relationship between economic indicators and neither suicide nor homicide is straightforward and that a considerable amount of fine-tuning still needs to be done, especially for low-income countries but even for the European context.

For years, researchers have been cautious about performing a comprehensive global overview of homicide and suicide rates, usually due to regional and cross-national variations (Knox, Conwell, & Caine, 2004; Ouimet, 2012; Phillips & Cheng, 2012; Shah, 2007). On the other hand, some studies have proposed global trends of violence based on violent crime rates alone (Eisner, 2003; Pinker, 2011), leaving out the consideration of other forms of violence. Therefore, this article proposes a longitudinal study of homicide and suicide with the inequality measures of unemployment rates and Gini coefficients in as many countries and reaching as far back as the data will allow, which begins only in 1962 but combines homicide and suicide as one variable for the first time for worldwide data. The object is to determine whether observing the two together elucidates trends that would not be detectable through either alone, and to consider the ensuing implications. Longitudinal studies of homicide and suicide with inequality measures have been conducted extensively in the past, especially with regard to the association between unemployment and suicide rates (Inoue et al., 2007; Moser, Fox, & Jones, 1984; Moser, Goldblatt, Fox, & Jones, 1987; Moser, Jones, Fox, & Goldblatt, 1986). Previous studies have also explored the correlation between homicide and suicide (Bills & Li, 2005; Kennedy, Iveson, & Hill, 1999; Lester, 1988). However, homicide and suicide have never been combined as one variable, to the extent of our knowledge, under the assumption that the two types of violence indicate different, largely unrelated forms of violence.

The hypothesis is to test whether combining homicide and suicide into one variable will increase power, both statistically and theoretically, than when analyzing the two separately. The results can provide a more nuanced approach to deciphering the etiology of violence, in terms of where causes are shared between different types of violence and where the types of violence differ, as well as when a general tendency for violence might manifest as either homicide or suicide, depending on gender, ethnicity, personality, or immediate circumstances. Our hypothesis takes from the theory that homicide and suicide share similar causes and risk factors, but have different manifestations according to the social–cultural experience (Unnithan, Corzine, Huff-Corzine, & Whitt, 1994). Instead of treating homicide and suicide as two independent phenomena, suicide can be viewed as targeting violent behavior toward oneself, and homicide targeting violent behavior externally (Rezaeian, 2011), which has implications in a multicultural but rapidly Westernizing and polarizing world. Others have noted merit in examining homicide and suicide together within the same theoretical and empirical model (Batton, 2004; Wu, 2003). Our goal in integrating the study of homicide and suicide in relation to inequality and unemployment measures is to begin to address the general problem of studying different types of violence in silos, and of approaching violent behavior as a crime rather than a health issue that is of the whole population’s concern.

2. Methods

2.1. Sample and data

The sample consists of countries selected from 75 participating nations in the WHO. Countries that formed through the dissolution of larger nation states, and nations providing fewer than 20 years of data were excluded from analysis. After excluding ineligible countries, our dataset included: Argentina, Australia, Austria, Belgium, Brazil, Bulgaria, Canada, Chile, Colombia, Costa Rica, Cuba, Denmark, Finland, France, Germany, Greece, Guatemala, Hungary, Ireland, Israel, Italy, Japan, Kuwait, Mauritius, Mexico, the Netherlands, New Zealand, Norway, Poland, Portugal, Korea, Singapore, Spain, Sri Lanka, Sweden, Trinidad and Tobago, The United Kingdom, The United States of America, Uruguay, and Venezuela. The final sample size consisted of 40 nations and 732 observations.

The data were compiled through a number of public repositories of global information. Historical data on trends from officially reported homicide and suicide death rates were retrieved from the WHO Health Statistics and Information System’s Mortality Database. When investigating factors that influence worldwide homicide rates, extant literature relies predominantly upon cause of death data provided by the WHO (LaFree & Tseloni, 2006; Pampel & Williamson, 2001). We consider data provided by the WHO as the most reliable aggregation of global data available on homicide and suicide rates; the WHO functions as a repository of longitudinal data related to violent death rates of participating countries in the most uniform manner available to date. The collection of data occurs through civil registries and death certificates and aggregated by year, sex, age, and cause of death. The World Bank provided data on the unemployment rate (World Bank website: http://data.worldbank.org/indicator/SL.UEM.TOTL.ZS). The choice of the World Bank as a source was because of its standards for assessing data quality through internationally accepted guidelines, including those of the United Nations (UN) Fundamental Principles of Official Statistics. The United Nations University (UNU) provided data on Gini coefficients (UNU website: http://www.wider.unu.edu/research/Database/en_GB/database/). Similarly to the WHO and the World Bank data, we considered that UN constituent agency data archives would be most reliable for the most consistent inequality measures, despite the inherent difficulty of obtaining income distribution data. The particular database that we used was UNU-WIDER World Income Inequality Database, Version 2.0c, May 2008.

2.2. Measures

We created a variable measuring a combined violent death rate by summing the intentional homicide and self-inflicted death rates provided by the WHO (HR + SR = CDR). A second variable divided the suicide rate by the combined death rates to produce a measurement of the percentage of the combined death rates accounted for by suicide $\left(\frac{\text{SR}}{\text{CDR}}\text{Rate}\text{Promotion}(\text{RP})\right)$. A third variable calculated the rate ratio between suicide and homicide by dividing the suicide rate by the homicide rate $\left(\frac{\text{SR}}{\text{HR}}={\text{RR}}_{\text{SR}/\text{HR}}\right)$. The rate ratio measures the magnitude of difference between suicide and homicide rates for each nation. A value greater than one indicates a year in which the rate of suicide exceeded the rate of homicide.

2.2.1. The statistical model

In an effort to examine the relationship between homicide and suicide, we performed descriptive analyses on cross-national trends of rate ratios, rate proportions, as well as multivariate longitudinal analyses on homicide, suicide, and combined death rates in a sample of 40 nations from 1962–2008. Panel data is a form of multi-level data in which time points are observations nested within higher-order groups. In the case of our data, each time point is nested within a country. As a result, each of the year observations is not independent from its membership in a nation. It is important to select a statistical model that handles the hierarchical structure of the data appropriately. Since we are not interested in testing the association of any time invariant characteristics, the fixed effects model is the most appropriate analytical technique to approach our research objectives. The fixed effects model extinguishes the potential for variation in the higher-level units thus eliminating the potential for bias and confounding introduced by unmeasured time-stable nation-level variables (Lindstrom & Bates, 1990; Rabe-Hasketh & Skrondal, 2008).

The central concern of our study involves: (a) investigating the cross-national association between changes in the mean Gini coefficient, the mean unemployment rate, and the mean deaths of multiple outcome variables; and (b) elucidating the impact of combining homicide and suicide on understanding predictors of rates of violent death. The following equation was fit using a multilevel fixed effects model in Stata Version 13: In(DeathRate)_{it(1962 – 2008)} = β₁Gini_{it(1962 – 2008)} + β₂(Unemployment)_{it(1962 – 2008)} + ∝ + u_it. Our dependent variables are the log odds of homicide rates (HR), suicide rates (SR) and combined death rates (CDR), where i equals the specific nation and t represents the point in time of the observation. The intercept, ∝_i(i = 1 − n) represents the fixed effect for each nation state. β_{1 – 2} are measures of effect for the Gini coefficient and unemployment rate for each nation (i) at an observed point in time (t). Finally, u_it symbolizes the error term for each nation at time (t). Through a series of natural log transformations, death rates and unemployment rates were converted into a relatively normal distribution and entered into fixed effects regression models examining the association between unemployment rates and Gini coefficients on: (a) homicide rates (HR), (b) suicide rates (SR), and (c) combined death rates (CDR).

3. Results

3.1. Descriptive statistics

Table 1 presents homicide, suicide, and combined violent death rates for the 40 nations included in the study from 1962–2008. The basic structure of the table was adapted from an empirical article by LaFree and Tseloni (2006) that examined the association between democratic governments and homicide rates using this dataset. Minimum and maximum values embody the cross-nation range for all time-point observations for all nations included in the study. All rates are population standardized per 100,000 residents. The mean homicide death rate was 4.91 with a cross-national range of 0.74 (Ireland) to 50.63 (Colombia). Behind Colombia, mean rates of homicide were greatest in Guatemala (29.05), Mexico (20.54), Brazil (21.04), and Venezuela (14.69). Examining the within-national spread of mean homicide rates identified Guatemala (130.98) and Colombia (90.93) as having extreme outliers for homicide death rates. A similarly wide range in mean suicide rates was observed with a low of 0.97 (Kuwait) and a high of 30.22 (Hungary) with a mean suicide death rate of 6.53. Following Hungary, nations with the highest suicide death rates consisted of Finland (22.09), Austria (19.05), Denmark (18.84), Switzerland (17.78) and Japan (16.67). For combined death rates, the mean for the whole sample was 15.97 with a range of 0.66 to 131.70 (Guatemala). The lowest combined death rate in the sample was 2.39 (Kuwait) and the highest was 55.32 (Columbia). Fig. 1 provides a visual representation of the relationship between temporal changes in the homicide, suicide, and combined death rates.

Table 1.

Homicide, suicide, and combined death rates and descriptive statistics (n = 46).

Country	Points	Years	Homicide rate (HR)			Suicide rate (SR)			Combined death rates (CDR)			SR/CDR			RR (SR/HR)
			Mean	StdDv.	Min, max	Mean	StdDv	Min, max	Mean	StdDv	Min, max	Ratio	StdDv	Min, max	Ratio	StdDv	Min, max
Argentina	36	1966–2007	5.54	1.28	3.68, 9.42	7.65	1.17	6.15, 10.74	13.19	2.21	10.56, 17.98	58.00%	4.09%	.45, .67	1.41	0.24	.83, 2.03
Australia	44	1962–2006	1.70	0.31	.66, 2.14	12.22	1.66	7.43, 16.35	13.92	1.70	8.09, 17.74	87.73%	2.30%	.84, .93	7.5	1.85	5.23,12.38
Austria	47	1962–2008	1.16	0.31	.50, 1.73	19.05	3.48	11.15, 24.31	20.21	3.73	11.66, 25.66	94.30%	0.87%	.92, .96	16.96	2.72	12.01, 23.20
Belgium	39	1962–2004	1.31	0.46	.54, 2.37	15.69	2.67	11.21, 20.16	17.00	3.04	11.75, 21.83	92.46%	1.83%	.87, .95	13.02	3.39	6.46, 20.73
Brazil	24	1982–2005	21.04	4.08	13.84, 26.85	4.22	0.35	3.63, 4.78	25.26	4.34	17.66, 31.39	17.03%	2.33%	.14, .23	0.21	0.03	.17, .29
Bulgaria	45	1964–2008	2.78	0.85	1.46, 4.81	11.78	1.84	8.21, 14.47	14.55	2.52	9.80, 19.20	81.17%	3.08%	.74, .86	4.45	0.86	2.87, 6.19
Canada	43	1962–2004	1.94	0.41	1.33, 2.71	11.94	1.57	8.00, 14.53	13.88	1.92	9.49, 17.02	86.15%	1.58%	.82 .89	6.31	0.84	4.61, 8.18
Chile	44	1962–2005	3.71	1.51	1.93, 8.11	6.91	1.94	2.47, 10.68	10.62	3.17	4.80, 18.48	65.40%	6.63%	.50, .79	2.00	0.60	1.00, 3.82
Colombia	35	1962–2006	50.63	21.28	19.59, 90.93	4.70	1.42	2.75, 8.42	55.32	20.79	23.54, 94.46	10.17%	5.46%	.03, .22	0.10	0.05	.034, .22
Costa Rica	45	1962–2006	5.43	0.96	3.28, 7.44	5.60	1.25	3.19, 8.12	11.03	2.01	7.26, 15.56	50.56%	4.66%	.37, .60	1.04	0.19	.60, 1.52
Cuba	27	1964–2007	5.27	1.07	3.81, 7.49	15.99	3.93	9.43, 21.30	21.27	4.12	14.20, 27.91	74.50%	5.89%	.65, .84	3.15	1.05	1.83, 5.17
Denmark	45	1962–2006	0.94	0.29	.43, 1.46	18.84	5.27	8.97, 28.32	19.33	5.33	9.98, 29.67	94.72%	2.27%	.89, 98	21.61	9.37	8.44, 45.13
Finland	47	1962–2008	2.63	0.42	1.78, 3.54	22.09	2.71	16.23, 27.48	24.72	3.03	18.15, 30.49	89.38%	1.03%	.87, .91	8.50	0.92	6.63, 10.53
France	46	1962–2007	0.93	0.20	.57, 1.62	15.48	1.88	13.03, 19.47	16.41	2.01	13.60, 20.63	94.32%	0.97%	.89, .96	17.01	2.39	8.22, 22.77
Germany	45	1962–2006	1.09	0.23	.55, 1.45	15.23	3.51	8.64, 20.00	16.31	3.71	9.20, 21.23	93.33%	0.79%	.91, .94	14.19	1.67	10.62, 16.79
Greece	47	1962–2008	0.90	0.23	.50, 1.41	2.97	0.35	2.29, 3.72	3.88	0.45	2.92, 4.82	76.81%	4.61%	.66, .85	3.48	0.90	1.95, 5.55
Guatemala	36	1963–2006	29.05	21.26	1.56, 130.98	2.59	1.13	0, 5.02	31.64	21.14	1.60, 131. 70	9.47%	4.97%	0, .20	0.11	0.0623	0, .24
Hungary	47	1962–2008	2.34	0.54	1.53, 3.72	30.22	6.3	18.45, 39.80	32.56	6.49	19.98, 42.19	92.67%	1.69%	.89, .95	13.31	3.05	8.01, 18.79
Ireland	47	1962–2008	0.74	0.3	.21, 1.94	7.32	3.53	1.79, 13.17	8.07	3.71	2.03, 14.22	89.51%	4.26%	.78, .95	9.97	3.86	3.55, 19.08
Israel	32	1975–2006	2.31	1.48	.50, 7.76	6.65	0.89	5.27, 8.39	8.95	1.56	5.88, 13.66	75.55%	10.85%	.43, .92	3.92	2.39	.76, 11.93
Italy	44	1962–2007	1.35	0.48	.79, 2.73	5.60	0.61	4.57, 6.72	6.95	1.00	5.42, 8.78	81.08%	4.28%	.69, .86	4.52	1.09	2.21, 6.29
Japan	47	1962–2008	0.92	0.38	.36, 1.57	16.67	1.91	13.14, 19.65	17.59	1.93	13.70, 20.48	94.76%	2.01%	.91, .98	21.87	10.62	10.48, 51.97
Kuwait	25	1972–2008	1.42	0.79	0.41, 3.82	0.97	0.48	.12, 1.60	2.39	0.92	.66, 4.15	41.58%	17.45%	.08, .71	0.88	0.64	.09, 2.41
Mauritius	47	1962–2008	2.50	1.32	.65, 8.37	8.18	4.47	1.77, 17.00	10.72	4.67	2.84, 19.57	73.02%	14.95%	.26, .95	3.94	2.98	.35, 17.30
Mexico	46	1962–2007	20.54	6.26	8.17, 34.14	2.74	0.95	.91, 4.29	23.28	5.76	12.34, 36.53	13.28%	8.21%	.04, .34	0.16	0.12	.05, .51
Netherlands	47	1962–2008	0.86	0.29	.24, 1.33	8.36	1.16	6.09, 11.09	9.22	1.34	6.46, 12.03	90.10%	2.69%	.86, .96	10.95	4.07	6.33, 26.70
New Zealand	45	1962–2006	1.47	0.43	.61, 2.31	11.57	1.76	8.60, 15.02	13.04	2.06	9.82, 16.67	88.83%	2.41%	.84, .94	8.42	2.33	5.10, 16.64
Norway	46	1962–2007	0.89	0.29	.39, 1.55	10.73	2.44	6.34, 15.64	11.62	2.69	6.80, 16.76	92.42%	1.19%	.89, .95	12.51	2.11	8.41, 18.34
Poland	42	1962–2008	1.53	0.63	.82, 2.84	12.26	1.36	9.06, 14.31	13.80	1.83	9.97, 17.15	89.17%	3.19%	.82, .93	8.91	2.42	4.47, 13.12
Portugal	42	1962–2003	1.32	0.29	.77, 1.95	7.72	1.69	3.70, 10.14	9.03	1.74	4.60, 11.56	84.93%	3.99%	.77, .91	6.17	2.14	3.31, 10.51
Korea	22	1985–2006	1.51	0.29	1.00, 2.04	12.82	5.08	7.24, 22.40	14.33	5.24	8.37, 24.06	88.58%	3.13%	.83, .93	8.46	2.79	4.98, 13.49
Singapore	43	1963–2005	1.59	0.66	.54, 2.80	12.69	2.96	8.06, 19.42	14.27	3.42	8.61, 21.02	89.14%	3.43%	.77, .95	9.16	3.38	3.39, 18.08
Spain	44	1962–2005	0.70	0.35	.063, 1.27	5.24	1.10	3.67, 6.78	5.94	1.31	4.09, 7.59	88.58%	5.18%	.78, .99	12.92	14.50	3.60, 75.99
Sweden	46	1962–2007	1.10	0.22	.52, 1.49	15.16	3	10.34, 19.89	16.25	2.99	11.27, 20.75	93.05%	1.82%	.90, .97	14.67	5.21	9.01, 31.15
Switzerland	46	1962–2007	0.99	0.28	.53. 1.63	17.78	2.78	13.09, 22.79	18.77	2.90	13.71, 23.78	94.71%	1.33%	.91, .97	19.11	5.01	10.43, 30.74
Trinidad and Tobago	37	1962–2002	7.85	2.87	2.46, 15.12	9.98	3.91	2.57, 16.17	17.84	5.92	6.77, 28.86	55.23%	10.95%	.21, .70	1.35	0.49	.27, 2.35
United Kingdom	46	1962–2007	0.87	0.31	.40, 1.60	7.30	0.98	5.73, 10.09	8.17	1.02	6.13, 10.67	89.37%	3.51%	.81, .95	9.45	3.45	4.33, 17.37
United States of America	44	1962–2005	8.31	1.56	5.34, 10.56	11.21	0.75	9.53, 12.67	19.52	2.11	15.73, 22.69	57.78%	4.02%	.52, .68	1.39	0.25	1.08, 2.11
Uruguay	37	1963–2004	3.98	1.01	2.03, 5.85	11.00	1.83	7.96, 15.65	14.98	2.59	11.05, 21.40	73.60%	4.15%	.65, .84	2.89	0.70	1.82, 5.04
Venezuela	42	1962–2007	14.69	6.95	8.50, 33.79	6.51	1.47	3.49, 9.79	21.20	6.25	13.40, 38.75	33.16%	10.48%	.09, .49	0.53	0.23	.10, .97
n = 40			4.91	10.07	0.06, 130.98	11.06	6.53	0, 39.80	15.97	10.22	.66, 131.70	75.51%	24.98%	0, .99	8.13	7.46	0, 75.99

Fig. 1.

Homicide, suicide, and combined death rates, per 100,000 population per year, worldwide, 1962–2008.

3.2. Ratio of suicide to combined death

A quartile analysis revealed that for most of the nations in this study, suicide comprised a majority of the combined death rates. On average, the suicide rate accounted for 75.51% of the overall violent death rate with a range of 0–99% in the 40 nations from 1962–2008. In the fourth quartile, suicide accounted for 75% or more of the combined death rates for over two thirds or 26 countries in the sample. Suicide accounted for between 25.00% and 74.99% of the combined death rates (2nd and 3rd quartiles) for 20% of the sample (10 nations). At its midpoint, suicide comprised more than half of the combined death rates for 85% of the sample (34 nations). For 10% (4 nations) of the overall sample, suicide accounted for 25% or less of the combined mean death rate. Quartile analysis of the ratio between suicide to combined death rates illustrates heavy clustering of nations in the upper fourth quartile when compared to the distribution of nations at lower quartiles

3.3. Ratio of suicide to homicide

The ratio of suicide to homicide rates provided an indication of the overall magnitude of difference between the two rates. Overall, the mean suicide rate was 8.54 times the homicide rate with a range of 0–75.99. The suicide rates of Austria (16.96), Denmark (21.61), France, (17.01) Japan (21.87), and Switzerland (19.11) were at least 14 times the homicide rates. Conversely, for Mexico (0.16), Guatemala (0.11), Colombia (0.12), and Brazil (0.21), the suicide rate was at most 0.21 times the homicide rate. Analysis of the rate ratio also illustrates that nations with the greatest suicide rates are characterized with very low homicide rates. This pattern is also true for nations with the greatest homicide rates, in that they are also characterized with notably low suicide rates. Fig. 2 plots changes in time in the ratio of suicide to homicide from 1962–2008. Table 2 gives descriptive statistics, while Figs. 3 and 4 plot longitudinal changes, in unemployment rates and Gini coefficients from 1962–2008.

Fig. 2.

Suicide/homicide ratio per year, worldwide, 1962–2008.

Table 2.

Unemployment rates and Gini coefficient descriptive statistics.

	Mean	Std. deviation	Min, max
Gini	36.7	10.24	17.8, 64.0
UR	7.04	4.4	0, 28.1

Fig. 3.

Unemployment rate, per 100,000 population per year, worldwide, 1962–2008.

Fig. 4.

Gini coefficient, per 100,000 population per year, worldwide, 1962–2008.

3.3.1. Multivariate analysis

Preliminary tests identified the presence of serial correlation and heteroskedasticity in all of the fixed effects regression models using the Wooldridge Test for Serial Correlation (Wooldridge, 2010) and the Modified Wald Test for Group-wise Heteroskedasticiy (Drukker, 2003) (results available upon request). In order to correct for the presence of bias, Driscoll and Kraay (1998) standard errors were applied to all of the regression models (Hoechle, 2007). We also have reason to suspect, given the spatial clustering of our data, that the presence of cross-sectional, spatial correlation affects our standard error estimates. The Driscoll and Kraay (1998) method is robust to heteroskedasticity, and temporal serial dependence and spatial correlation and has found successful application in prior empirical inquiry into cross-national suicide rates (see Matsubayashi & Ueda, 2011). Table 3 reports results of the fixed effects estimation for all ages and age-stratified estimates. For all of the tables, three columns present estimates for the dependent variables. Unemployment and Gini coefficients were regressed on death rates in separate models for logged homicide, suicide, and combined death rates. Each increase in one unit of the unemployment rate was associated with an increase of 0.09 in homicide deaths (p < 0.001) and an increase of 0.039 (p < 0.05) in suicide rates. When combined, a small albeit significant relationship was observed between logged Gini coefficient and the combined death rates (0.0066, p < 0.05). This suggests that the combined homicide and suicide rate in a single measure improves the ability to detect a significant relationship when compared to either rate measurement alone. A significant association between logged unemployment rate persisted when death rates were combined (0.061, p < 0.01).

Table 3.

Association between Gini coefficient and log unemployment rate on logged suicide, homicide and combined death rates.

	All ages
	LogHomicide	LogSuicide	LogCombined death rates
Gini	.002 (.004)	.005 (.003)	0.0066* (.0029)
Unemployment rate (Log)	.09*** (.018)	.039* (.044)	0.061** (.017)
Model F-test	12.98***	3.27*	6.46*
R-Sq	0.0265	0.0219	0.0391

Note: Parameter estimates are from fixed effect regression models, the Driscoll–Kray standard errors are in parentheses. Significance for parameter estimates are as follows:

^*p < .05,

^**p < .01, and

^***p < .001.

Table 4 presents fixed effects models for the association between Gini coefficients and unemployment rate on suicide, homicide and combined death rates for 5 age bands. Logged unemployment rate was associated with an increase in the suicide rate in the 0–14 and 15–29 age bands. Similarly, a significant association was found between the logged unemployment rate and an increase in homicide death rates for all of the age bands with the exception of ages 0–14. In older national age bands, a weaker association was found between the unemployment rate and homicide rates. Where the association was strong in statistical significance and magnitude, combining death rates into a single measure improved statistical power. For instance, combining homicide (0.152, p < 0.001) with suicide (0.137, p < 0.001) resulted in a strong aggregate measure (0.158, p < 0.001). Combining rates containing parameter estimates with divergent significance resulted in a weaker aggregate overall death rate. Combining without first looking at the independent variables dampens the true power of the significant variable while simultaneously misrepresenting the variable without statistical significance.

Table 4.

Age-banded association between Gini coefficient and unemployment rate on homicide, suicide, and combined death rates.

	Ages 0–14			Ages 15–29			Ages 30–44			Ages 45–59			Age 60+
	Homicide	Suicide	Overall	Homicide	Suicide	Overall	Homicide	Suicide	Overall	Homicide	Suicide	Overall	Homicide	Suicide	Overall
Gini	.008** (.004)	.008 (.005)	.008* (.004)	.003 (.003)	.005 (.003)	.007* (.003)	−.0009 (.005)	0.0027 (.0047)	.005 (.003)	−.005 (.003)	.006 (.003)	.007 (.003)	.004 (.003)	.005 (.003)	.005 (.003)
UR (Log)	.009 (.033)	.17*** (.027)	.065** (.023)	.152*** (.030)	.137*** (.031)	.158*** (.029)	.09*** (.007)	0.054 (.020)	.07*** (.007)	.074* (.031)	−.024 (.006)	−.010 (.024)	.074* (.028)	−.04 (.027)	−.027 (.028)
R2	0.0055	0.0319	0.012	0.0403	0.0878	0.1	0.02	0.024	0.0306	0.019	0.023	0.022	0.011	0.026	0.022
Model F-test	4.05**	5.59**	5.68**	14.29***	11.78***	16.53***	8.26**	5.28*	9.83***	4.63*	2.92	2.81	3.54*	4.13*	3.68*

Note: Parameter estimates are from fixed effect regression models, the Driscoll–Kray standard errors are in parentheses. Significance for parameter estimates are as follows:

^*p < .05,

^**p < .01, and

^***p < .001.

4. Discussion

Applying longitudinal epidemiological research methodology, the present study sought to identify associations between a major public health problem, i.e. violent deaths resulting from homicide and suicide, and socioeconomic factors, i.e. unemployment rates and income inequality. By joining homicides and suicides as a combined measure of violence in different countries, the study aimed to find stronger associations to environmental determinants than when analyzing homicide and suicide in isolation, as previous studies have done. Such associations can provide the foundation for future research in better determining the etiology of these major causes of death and possible prevention methods when compared to variables previously observed in qualitative connections. The effects of macro-level factors (e.g., social and economic policies) on violent death rates in a population have been shown to be more significant than those of micro-level influences (e.g., individual characteristics), making the combination of related violent deaths a worthwhile undertaking so as to capture more subtle fluctuations or “transfers” from one modality of death to another over time. Hence, the rationale for the associations between macro-level determinants and violent death rates could be strengthened by combining homicide and suicide rates to give larger, more comprehensive insight into the etiology of violence, and to identify where similarities and differences occur in the determinants of homicide and suicide.

This analysis was inspired by a preliminary project whereby combined violent death (homicide and suicide) rates were found to be associated with political party (Democrat vs. Republican) in the United States over a time span of 111 years (Lee, Wexler, Gilligan, & Stolar, 2014-in this issue). Results from this study indicated that violent deaths were associated with an increase under Republican presidents and a decrease under Democratic presidents, and increasing unemployment and falling growth rate of per capita gross domestic product (GDP) are among the factors that correlate with violent death. Whether political party was a cause, a consequence, or a confounding factor co-occurring with “the socioeconomic health of the nation” was impossible to decipher, but it gave rise to the possibility that democratic policies or climates that served to reduce unemployment or inequality also served to decrease violent death rates, while conservative policies or climates that served capital (with the side effect of increasing unemployment and inequality) also increased violent deaths. A multinational longitudinal study, for which the present paper is merely a speculative beginning, should either refute or add strength to this comparative analysis, as well as better delineate the underlying situation of which the violent deaths serve as barometers.

This study compared the effects of unemployment rates and Gini coefficients in 40 countries over a time span of 46 years (1962–2008), in terms of changes in homicide and suicide as a combined variable. No previous study had examined the integration of homicide and suicide into one variable with the intent of exploring their similarities and differences in etiology, in association with socioeconomic factors in an overarching, multinational empirical study over time. Integrating the two related violent death rates into one variable would increase statistical and theoretical power of the associations. This method would capture more general violent tendencies that may be differently manifested across countries, depending on specific environments, but which have shared causes and risk factors. The results confirmed this hypothesis.

To increase confidence in the associated finding, the longitudinal design focused on examining the largest number of years possible, given the data, which ensured that the identified associations would have to persist over the socioeconomic variables (unemployment rate and Gini coefficient). As a result, one-time events occurring throughout the course of a country’s history that result in inordinately high levels of homicide or suicide—i.e., civil wars, economic depressions, natural disasters, etc.—were likely to have a lessened effect on the associations. Secondly, a sample size of 40 countries from all the major regions of the world lessened the likelihood that country-specific socio-cultural factors would dominate the results. Significant associations were present in the combined homicide–suicide variable despite vast social and cultural differences between sample countries, both high- and middle- to low-income, confirms the advantage of using an adequate sample size of countries in unveiling patterns of violence in relation to socioeconomic factors.

The positive association between unemployment rates and death rates persisted across different age groups, strengthening the findings. Specifically, similarities in economic factors driving temporal changes in homicide and suicide rates were found to be a function of age. For these data, homicide and suicide rates shared similarities in Gini coefficients and unemployment rates for the age band of 15–29. However, these relationships shifted during the analysis of different national age bands. It is critical to note, therefore, that the benefits of combining different types of violence do not negate the complexity of human violence and the need to have nuanced deliberations over it. We therefore suggest that combined rates always be considered a heterogeneity that will need intricate consideration of context, age group, and other influences. In certain contexts, combining death rates may mask underlying relationships between independent predictors while in others may improve statistical power to detect effects that might not otherwise be discoverable. Furthermore, the distinctions may not end at the level of homicide and suicide: some contexts may call for usefulness in distinguishing different subtypes, while others for further combining with other types. The point is that human violence, like all human behavior, should not be reduced to a uniform manifestation with a singular cause or singular trend, even when analyzing it at macro level. In other words, the results of this study should not be taken to mean that differences between types of violence can be ignored. Rather, subtle consideration of human-level dynamics should accompany large-scale interpretations, wherever possible at all times, given the complex nature of human behavior.

The nature of this study as a preliminary analysis for elucidating stronger associations between violent death and socioeconomic factors resulted in limitations, and calls for a more complete study of violent deaths. More dependent variables could be included to give a comprehensive assessment of how a country’s socioeconomic landscape may prevent or promote violence; since it was our goal to include as many years, and as great a variety of countries, as possible. In the literature, long-term trends of violence have been drawn without proper data (e.g., Pinker, 2011), generalizing violent crime to overall violence and early European trends to worldwide ones. While the effort is certainly worthwhile, premature conclusions without an adequate understanding of the variables can lead to misleading perceptions. For example, if one believed that violence rates are decreasing based on interpersonal violence alone, without regard to the very concerning rise of worldwide rates of suicide, which are currently greater than all other forms of behavioral violence combined (Krug et al., 2002; World Health Organization Violence Prevention Alliance, 2012), not to mention structural violence, which kills more people than all behavioral violence combined (Galtung, 1969; Farmer, 2003), one would be prone to misleading deductions. Additionally, if one generalized trends from the global North without regard to concerns that deaths, injuries, and disabilities due to violence are expected to increase in low- and middle-income countries (United Nations Office on Drugs & Crime, 2014), and that the bulk of violent deaths occur in societies where data accuracy is still wanting but experience the greatest effects and the largest population growth, one would reach premature predictions. These prevailing concerns led this study to err on the side of a longer time interval and greater inclusion of countries of varying economic levels, over comprehensiveness of socioeconomic variables. A future study, in process at the time of this article, will include more economic variables and political indicators within a smaller time range to add depth to the present findings.

For the time being, we decided to narrow our socioeconomic variable focus on unemployment rates and the Gini coefficient in order to assess the suitability of the analysis. The Gini coefficient is an effective means of measuring inequality, but there are some limitations: for example, there are different income distributions with the same Gini coefficient (De Maio, 2007); extreme wealth inequality can have low-income Gini coefficients (Domeij & Flodén, 2010); there is a small sample bias, as sparsely populated regions are more likely to have low values (Deltas, 2003); and the Gini coefficient is unable to discern the effects of structural changes in populations (Kwok, 2010). Therefore, unemployment rates have been added as an indirect measure of inequality, since they are often its underlying cause. Inclusion of more socioeconomic indicators would hence be valuable for adding important dimensions to this analysis.

Another limitation of the current study is that it focused on violent death rates without accounting for the numerous incidents of nonlethal violence that occur for each loss of life. Some important forms of violence, such as neglect or psychological violence, can potentially result in more severe effects but were not included in this analysis due to inconsistency of data across countries. Medical care in the global North has developed to such a degree as to skew the information on violence, for homicidal or suicidal intent and action may have existed in many cases but not resulted in lethality, such that death rates reflect in large part the distribution of medical care (Monkkonen, 2001). Inclusion of these numbers would bring the study into a broader context, and may even lead to more significant associations in future study. Overall, this study recognizes that violence is a conintual, intricate process of which homicide and suicide are the extreme endpoints, sometimes incidental, of its manifestations.

The stronger statistical and theoretical power of integrating homicide and suicide as one variable, as opposed to treating them separately, may allow for a more sensitive treatment of this process. The idea that different manifestations of violence can be grouped together results from their deeply interconnected psychological processes (Gilligan, 1996); others have also described them as illness processes akin to an infectious disease (Slutkin, 2012). An example of this dynamic, in the form of a modified “germ theory,” is shown in Fig. 5. The integration approach may also account for possible “transfers” from one mode of expression to another through time. For example, a spike in homicides and other minor conflicts has been noted throughout the twentieth century, especially following democratization (LaFree & Tseloni, 2006). However, we find that the spread of democracy generally has a stabilizing effect on violence, in terms of reducing interstate wars and eventually smaller-scale violence and homicides as well. Was the spike in the latter half of the twentieth century a fluke, then, to be dismissed amid a general trend of declining violence, as some authors (Pinker, 2011) have noted? More likely, since one form of violence can beget other forms, the great rise in interpersonal violence following the spread of democracy throughout the world is an after-effect of the crumbling autocracies, of which Russia and South Korea are examples. And according to the respective cultures, it manifested largely in the form of escalating homicide rates in Russia and rising suicide rates in South Korea. Were the repressive autocracies, then, less violent? Few would argue that they were, which points to the sudden “spike” being merely a transfer from one type of violence to another.

Fig. 5.

A modified “germ theory” of violence.

In light of this paper’s consideration of combining typologies of violence, one might consider a model such as in Fig. 6 to explain the prevailing types of violence through recent history. In spite of the transfers, the democratization process has the potential to accelerate progress in violence reduction and that humankind has enjoyed a modest reprieve from its tendency to harm itself. However, the rise of global capitalism as it has been practiced over the last 25 years, with increasing concentrations of power, rising inequalities, and decreasing legitimacy of governance, has the potential of undoing many of these gains (Brada & Bah, 2014). In this case, one might expect to see not a continuation of this decline but a reversal in violence trends in the future. In just what form this violence will take seems impossible to predict, given the variable course of the past, with new forms of violence emerging with changing circumstances. While the taste for individual gore may have decreased (Foucault, 1975), indifference toward other forms of violence has remained, if not increased: e.g., large-scale civilian deaths through distance and drone fighting, negligence about wide-reaching policies that threaten the lives of millions, more insidious torture such as mass imprisonment and solitary confinement, genocide, and sweeping catastrophes such as global warming or nuclear expansion. Do these attitudes represent less violence? By numbers of deaths, the sheer extent and level of suffering, and a potential for self-directed harm at species-scale, all indicate a very disturbing trend toward violence that has not yet been resolved.

Fig. 6.

Examples of typological progression of violence through time.

Combining different types and expressions of violence, rather than considering them in isolation, can be a first step to wrapping our heads around some of these problems. The question then becomes not whether violent crime is rising or falling, whether suicide is increasing or decreasing, or whether there are more or fewer wars, but do our collective actions contribute to human violence? A deeper understanding of the nature of violence makes it easier to detect, to avoid, and finally to remedy the tendency when it exists, and to do more meaningful work than to predict forms based only on a partial picture. Approaching violence from a health perspective, rather than simply one of security and criminal justice, has allowed for a more intricate examination of root causes, of how psychological and socio-cultural dynamics play out, and of what effective interventions might consist. More research that focuses on these underlying processes of violence, rather than external manifestations, therefore, would not only allow for a more sophisticated appreciation of trends but also a means of prognosticating and preventing where it is most productive. While these speculations go far beyond the current analysis, they drive a series of analyses, of which this combined study of homicide and suicide is only the first, minuscule step.

5. Conclusion

Combining homicide and suicide created a more robust single variable that had stronger associations with macro-level socioeconomic determinants. As factors rooted in institutional structure are shown to have an increased effect compared to immediate micro-level factors in the causation of individual violence, these associations may be useful to discover. An expansion of socioeconomic variables, especially in conjunction with political factors, and the inclusion of other forms of violence in future research could help elucidate long-term trends.