Sources and Methodology

We have taken as a proxy of interpersonal violence the rate of intentional homicides per 100,000 inhabitants, a standard procedure in the literature. Intentional homicide is usually defined as an unlawful death deliberately inflicted on one person by another person, excluding casualties from interstate wars’ and civil wars (Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014; Cantor and Cohen Reference Cantor and Cohen1980; Eisner Reference Eisner2008; OECD 2011).Footnote ¹⁰ Why the reliance on homicide rates? When assessing personal security in a country, the rate of homicides is usually preferred to alternative crimes such as robberies because data on robberies is less reliable than data on homicides: a large portion of robberies are neither reported nor recorded (Fajnzylber et al. Reference Fajnzylber, Lederman and Loayza2002a; Gurr Reference Gurr1981; Mares Reference Mares2009; OECD 2011). Other serious crimes such as sexual violence or violence against children are also underreported.

In contrast, because it is (arguably) the most serious of all crimes, intentional homicide is the crime that is most effectively recorded in most countries (Eisner Reference Eisner, Heitmeyer and Hagan2003b; Gurr Reference Gurr1981; Monkkonen Reference Monkkonen2005; United Nations 2010). Homicide statistics suffer less from underreporting than robberies because corpses are more difficult to ignore than loss of property; homicides are less likely to go undetected (Fajnzylber et al. Reference Fajnzylber, Lederman and Loayza2002b; Mares Reference Mares2009); and less prone to statistical manipulation given the grievous nature of the crime (Cantor and Cohen Reference Cantor and Cohen1980).Footnote ¹¹ Homicides are also better recorded than other alternative measures of interpersonal violence such as violence against children or intimate partner violence. There is general agreement that homicide records are perhaps the only crime figure that provide a consistently accurate measure of serious crime levels (Gurr Reference Gurr1981; OECD 2011), and that allow us to conduct comparative analysis across countries and over long periods (Eisner Reference Eisner, Heitmeyer and Hagan2003b, Reference Eisner2008). Homicide, as a crime, has always attracted attention from all institutions of power in all societies, while its significance (for authorities) has remained stable over the course of the last centuries (Eisner Reference Eisner, Heitmeyer and Hagan2003b).

Homicide is also preferred to other variables of interpersonal violence because it has more severe and long-lasting negative effects on well-being (OECD 2011). Furthermore, it is well defined, enjoying a lower degree of measurement error, and it is highly correlated with other violent crimes, especially as far as long-term trends are concerned (Eisner Reference Eisner2008). For example, there is a strong correlation between the rate of homicides and the rate of people who have reported being assaulted (Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014; OECD 2011). For Eisner (Reference Eisner, Heitmeyer and Hagan2003b), homicides are the only form of interpersonal violence for which meaningful judgments on long-term developments can be made. Perhaps the only drawback to using homicide statistics is that there are often low rates of homicide in society, but that is why it is expressed as a rate per 100,000 inhabitants (OECD 2011).

To produce the series of homicides in Chile we have used two major sources. First, from 1911 onward, the number of homicides in Chile have been available from the Institute of National Statistics (a printed source), which provides annual counts of homicides, based on causes of death statistics, better known as mortality statistics.Footnote ¹² They are based on information provided by the National Register Office, which in turn is based on death certificates, usually provided by medical examiners. Statistics on causes of death are the best way of making comparisons across countries because they are based on standardized classifications of medically confirmed causes of death (Eisner Reference Eisner, Heitmeyer and Hagan2003b).

Second, for the period before 1911, we collected the same information, but directly from the death record books at the National Register Office of Chile, which are available for the whole country from 1887 only,Footnote ¹³ also based on death certificates or causes of death statistics provided by civil servants, and verified by doctors or witnesses. This information is available in manuscripts, rather than printed texts. The main difficulty of working with this type of source relates to the nature of the certificates, which were mostly provided by witnesses, resulting in a wide variety of causes of death. However, because we wanted to identify intentional homicides, the description provided by witnesses did not pose a problem.Footnote ¹⁴ In addition to the deaths explicitly recorded as “homicide” and “murder,” we also included those clearly identifiable cases of unlawful death deliberately inflicted on one person by another between 1887 and 1908. For example, we included cases in which a victim was “stabbed to death.” To the best of our knowledge, we are the first research team to have been given access to the death records of Chile from 1887 to 1910, after signing a confidentiality agreement (not to disclose information on any particular case). Approximately 1.3 million files containing all deaths in Chile from 1887 to 1910 were consulted, to identify all homicides (around 5,700 for the years we entered the data).

To complete our homicide rates per 100,000 inhabitants, population data was obtained from Díaz et al. Reference Díaz, Lüders and Wagner2016, the most trustworthy source for this variable on an annual basis because it relies on extrapolated official census data. Because accurate population data is needed for reliable estimates of homicide rates (Gurr Reference Gurr1981), it is worth stressing that Chile was a pioneer Latin American country in the compilation of national statistics, from the early nineteenth century onward (Estefane Reference Estefane2004, Reference Estefane2019). The first full national census was taken in 1835, and from then on censuses were taken regularly almost every 10 years. Despite the usual problems and drawbacks of any census, Chilean demographic data is regarded as high quality. From the second half of the century onward, there was centralized data collection for all censuses, following the creation of the National Statistics Centre in 1843, which improved the quality and coverage of the data. From 1907 in particular, further improvements were introduced (Loveman Reference Loveman2014; Pérez Reference Pérez2010; Ventresca Reference Ventresca1995).

Other type of sources used to collect data on homicides are police statistics, conviction statistics, and judicial statistics, although the most reliable are the ones used here: mortality statistics (Eisner Reference Eisner2008; Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014). There was no available data from conviction records in Chile, except for the period 1880–86 (Oficina Central de Estadísticas, 1880–88), which we used to complete our series on mortality statistics from the beginning of that decade, and which were close to the data based on mortality rates for the rest of the 1880s.Footnote ¹⁵ We accepted them as good quality; excluding them would not have changed our conclusions. Police statistics were also elusive, while judicial records are incomplete for the period before 1880 (these are available only for certain locations), so we preferred not to use them (on the difficulties of working with criminal proceedings files, see Eisner Reference Eisner, Heitmeyer and Hagan2003b).Footnote ¹⁶ Additionally, searches within manuscript judicial records are extremely time consuming.

Data on elite homicides, deaths by agents of the state during periods of social unrest or workers’ strikes, and deaths from military conflicts were obtained from a careful review of relevant secondary published works. Data on robberies (only used to test its correlation with homicides) was gathered from the UN World Crime Surveys from 1970 and for earlier periods from Chile’s Attorney General.Footnote ¹⁷ From Díaz et al. (Reference Díaz, Lüders and Wagner2016) came the series on average numbers of years of schooling, GDP growth rates and urbanization rates, while income inequality was taken from Rodríguez Reference Rodríguez2017. The poverty rate was taken from Prados de la Escosura (Reference Prados de la Escosura, Hatton, O’Rourke and Taylor2007) and the World Bank (for the most recent years), while police numbers came from the same UN World Crime Surveys and other national official sources, such as the Statistical Yearbook of the Office of National Statistics. The age structure of the population was obtained from the censuses of the republic and the demographic yearbooks.

When assessing the evolution of homicide rates across long periods, and across regions, it is important to be aware of potentially distorting factors. First, consider improvements in medical knowledge, which Eisner has defined as technologies of healing. Why is this important? Many of those who died from a wound before the twentieth century could have been saved with modern medical technology, including the increased use of antibiotics from the 1940s onward (Eisner Reference Eisner2003a, Reference Eisner2008; OECD 2011). Improvements in transport and communications, which reduced the time until an emergency team attended, also reduced the lethality of violence during the twentieth century (Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014; Eisner Reference Eisner2008). These changes affected urban and rural areas in different ways. However, modern weapons are increasingly dangerous (OECD 2011).Footnote ¹⁸ This factor has led to what Eisner (Reference Eisner2008) has defined as improvements in technologies of killing. Technologies of healing and technologies of killing do affect homicide trends in the long run, although it is difficult to estimate the precise net impact.

This article is the first academic work to analyse the evolution of intentional homicides in Chile. Previous important works have dealt with social and political violence in Chile, in particular for the twentieth century.Footnote ¹⁹ There are also some relevant general works on crime and criminology in Chile.Footnote ²⁰ Yet, despite the importance of these works, none of them has provided a comprehensive approach to the evolution of interpersonal violence for a sustained period.

General Trends in Homicide Rates in Chile, 1880s–2010s

We have produced a 140-year series on homicide for Chile, adding around 75 years to the previously available series from the Clio Infra Project,Footnote ²¹ thus making an important contribution to the study of trends in interpersonal violence in Chile. What did we find? In decadal annual averages, the number of homicides per 100,000 inhabitants in Chile was high during the late nineteenth century and the early decades of the twentieth century (between 16 and 19 between the 1880s and the 1930s), with an increasing trend between the 1880s and the 1920s, as seen in figure 2. These high rates were slightly below the homicide rates in the notably violent environment of medieval England and at a similar level to those of the Netherlands in the sixteenth century (Eisner Reference Eisner, Heitmeyer and Hagan2003b).

Figure 2. Average homicide rates in Chile and selected European countries, 1880s–2000s. Source: Authors for Chile data; Eisner (Reference Eisner2008) for all others.

Yet, if these figures are put into international perspective, the equivalent ratio in Western Europe (by now a society characterized by low levels of interpersonal violence) was 2.5 and 2.4 for the last two decades of the nineteenth century (Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014). By the early 1880s, most countries in Western Europe already had low homicide levels (Eisner Reference Eisner2008): the transition to a less violent society had been made.Footnote ²² Chilean society, however, remained subject to widespread violence.

It took some time for Chilean society to change: the “civilizing process” referred to by Elias, Spierenburg, Thome, Eisner, Mares, and other authors trying to explain the downward trend in homicide rates in Western Europe, belatedly arrived in Latin America.Footnote ²³ Chilean homicide rates remained high during the 1880s–1920s, declining moderately during the 1930s. However, in the 1940s they declined to 13.3, and the 1950s witnessed the most dramatic decline in homicide rates in Chile; the rate of the previous decade was more than halved; there were 4.9 homicides per 100,000 inhabitants, a rate comparable to that in the United States at the time (Mares Reference Mares2009). During the 1960s this ratio nearly halved again, reaching 2.7, and remained at a similar level until the 1990s.Footnote ²⁴

By the 1990s Chile had reached low levels of homicide by international standards, even if compared to developed nations. Between the 1930s and the 1960s Chilean homicide rates converged with the average levels in Europe, which had the lowest global homicide levels. In the 1920s homicide rates in Chile were 12 times higher than in Europe; during the 1980s–1990s they were only slightly more than twice as high. Homicide rates have increased during the last two decades (slightly above 4 per 100,000 inhabitants during the 2000s–2010s).

This increase may be due to the rise of organized crime and to the increase in drug trafficking in Chile. The average ratio of drug offenses (per 100,000 people) increased from 12 and 16 in the 1980s and 1990s, respectively, to 60 and nearly 120 in the 2000s and 2010s, respectively. There is also a perception of growing disparity between the rich and the poor, as the October 2019 riots have evidenced, leading to a decline in the legitimacy of the state, which often triggers violence (Mares Reference Mares2009). This was a period when the macro-economy did well, but the lower strata of society felt more frustrated. In such a situation, people are more likely to commit violent acts because the poorest become increasingly aware of their low status and their lack of social mobility (Gurr Reference Gurr1981).

The annual series in five-year moving averagesFootnote ²⁵ shows that the lowest level of homicide rates in Chile was achieved during the mid-1960s, and for a few years the rate fell below 2.0 (figure 3).Footnote ²⁶ As a reference point, that figure is roughly the current average of OECD countries (OECD 2011), while the lowest worldwide level of interpersonal lethal violence was attained during the 1950s by some Western European countries, at about 0.4–0.6 homicides per 100,000 inhabitants (Eisner Reference Eisner2003a).

Figure 3. Number of homicides per 100,000 inhabitants: Chile, 1880–2017 (5-year moving average). Source: Authors.

The transition toward low crime rates started in Chile comparatively late, in the early 1940s (figure 2). Developed countries such as the United Kingdom achieved similar rates in the second half of the eighteenth century, while Germany, the Netherlands, Belgium, and Switzerland followed suit during the earlier part of the nineteenth century. From the 1840s in particular, there was a sustained decline in criminal violence across Europe (Eisner Reference Eisner2003a, Reference Eisner2008), a century earlier than in Chile. In some southern European countries this process took longer, but, by the early 1940s, even countries such as Spain, which clearly lagged behind the European average, already had homicide rates below 2 (figure 2).

What needs explanation here is why homicide rates declined so significantly in Chile during the 1930s–1960s, and then again during the 1980s and 1990s. Chile was a special case; Eisner (Reference Eisner, Heitmeyer and Hagan2003b) found “an almost universal increase in homicide rates from the late 1950s to the early 1990s,” while for most of this period Chile’s homicide rates were declining or, at worst, remained stagnant. In the following section we document the results of regression conditioning on some important quantitative variables available to us. We regress homicide rates against some of these variables. There are other important variables that may be nonmetric, may not be statistic, and may not be linear, but that are equally important (Thome Reference Thome1995).

Following Mares (Reference Mares2009) and Thome (Reference Thome1995), it is important to look at the economic development of the country and at changes of large scale social structures during this period. The 1930s–1960s coincides with the period of inward looking development, or industrialization led by the state (Bértola and Ocampo Reference Bértola and Ocampo2012). There was a profound economic change in the country, characterized by greater intervention by the state in economic affairs, increasing investments in public works, and the promotion of national industrialization. State institutions grew in number, while many policies of social incorporation were promoted, leading to the emergence of a welfare state (Arellano Reference Arellano1985). The poverty rate of the country declined from 52 percent in the 1920s to 32 percent in the 1960s. The governmental social expenditure in per capita (real) terms increased nearly fivefold between the 1920s and the 1960s, while social mobility improved. Education and health care also improved; the average number of years children spent in school increased from 6.3 in the 1920s to more than 9 by the early 1970s. These changes probably triggered a greater degree of restraint in interpersonal behavior, lowering interpersonal violence and promoting internal pacification.

Another related development is the evolution of suffrage. Historians of crime should pay attention not only to the process of state modernization but also to the process of democratization of the country under analysis (Thome Reference Thome1995). In Chile between the 1810s and the 1870s, only a minority of the population was able to vote, while the government controlled most elections. In 1842, for example, only 2 percent of the population was able to vote (Urzúa Reference Urzúa1992; Valenzuela and Valenzuela Reference Valenzuela and Valenzuela1983).Footnote ²⁷ During the 1880s and 1890s some reforms were introduced (increasing the share of the population able to vote), but electoral fraud persisted and political power remained within the same hands. The 1920s and 1930s brought about a radical change. The franchise was extended; women were now able to vote (from 1934 in the council elections, and from 1949 in all elections), while new political parties emerged, including the socialist and communist parties. In 1932 only 7 percent of the population voted; by 1973 more than a third did so (Campos Reference Campos1984). So important was this transformation that according to Valenzuela and Valenzuela (Reference Valenzuela and Valenzuela1983), by the 1960s Chile was one of the most democratic countries on earth, despite being a developing nation. This is important because when the state transfers some power back to citizens, fostering political legitimacy, interpersonal violence decreases: the state monopoly of force gains further legitimacy (Mares Reference Mares2009). This process was taking place when homicide rates were declining most sharply in Chile.

The rise of bureaucratic state structures is also linked with a decline in violence because the state is regarded as a pacifying institution, which monopolizes violence (Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014; Eisner Reference Eisner2003a). Societies that experience a modernization process are less prone to high levels of criminality (Thome Reference Thome1995). In Chile from the late nineteenth century onward there emerged a strong bureaucratic state structure that encompassed the modernization of legal institutions, the reform of ministries, the creation of a modern prison system, the emergence of social security, the ordering of public finances, and general improvement in the regulation of social activities. These institutions were further improved during the following decades. Carabineros de Chile (the national police force) was created in 1927 and modernized thereafter (Barría Reference Barría2015; Cleaves Reference Cleaves1974; Faúndez Reference Faúndez2007; Henríquez Reference Henríquez2014; Islas Reference Islas2017; López Reference López2013, Reference López, Jaksic and Rengifo2017; Silva Reference Silva2008; Urzúa Reference Urzúa1970; Urzúa and García Reference Urzúa and García1971).

Another variable worth examining is the country’s level of alcoholism, and in particular the state’s attempts to decrease it. It has been widely established that alcohol consumption is one of the main determinants of crime. It is believed that both drugs and alcohol fuel violent crime and higher levels of interpersonal violence in general (Cea et al. Reference Cea, Ruiz and Matus2006; Ensor and Godfrey Reference Ensor and Godfrey1993; Gerson and Preston Reference Gerson and Preston1979; Lightowlers et al. Reference Lightowlers, Pina-Sánchez and McLaughlin2021; Murdoch and Ross Reference Murdoch and Ross1990). Alcohol consumption has been used as an important predictor of violent crime (Saridakis Reference Saridakis2004, Reference Saridakis2011). A study by Fundación Paz Ciudadana (1999) has shown that 59 percent of murderers in Chile consumed alcohol before committing homicide. Therefore, lower rates of alcohol consumption lead to a reduction in aggressive behavior; the adoption of humanistic values leads to less interpersonal violence.

For Chile in particular, during the late nineteenth century and early twentieth century many contemporaries linked the high consumption of alcohol to high levels of criminality in the country (in particular to homicides): most criminals were either alcoholics or were drunk when committing a serious crime. Alcoholism was perceived as a “social plague,” exacerbating violent behavior (Revista Médica de Chile Reference Médica de Chile1892; Zilleruelo Reference Zilleruelo1909). The year 1902 saw the passing of the first Alcohol Law to regulate taxation and consumption. Its aim was to increase alcohol prices to lower consumption, and at the same time to increase fiscal revenues, part of which went to finance campaigns against alcoholism (Fernández Reference Fernández2006). Subsequently, all presidents of the republic have announced other measures to combat alcoholism in Chile, including the construction of sports facilities, gyms, and cultural establishments, in particular in low-income neighborhoods (ibid.).

Salvador Allende, appointed minister of health in 1939, was an early and eloquent voice in the campaign against alcoholism. The government of the 1930s was one of the first to understand the importance of treating alcoholism as a serious social disease, that needed an integral and holistic treatment, rather than seeing it solely as a public order issue, as previous governments had done. The Ministry of Health implemented several policies to tackle alcoholism, aiming to improve the country’s human capital, social stability, and general well-being, and these helped to reduce the rate of homicide. However, this is a topic that calls for further research. There are many substantial gaps in the data on per capita alcohol consumption in Chile, for example.

We also wanted to provide a chronology of death penalty legislation in Chile, given the acknowledged deterrent effect of capital punishment on murder (Cantor and Cohen Reference Cantor and Cohen1980), and of the legislation on homicides. In Chile, the death penalty was established in 1874, when the first penal code of the country was promulgated; it applied to homicides, parricides, and infanticides.Footnote ²⁸ In the case of parricide, its application was automatic. In the case of other homicides, the minimum penalty was 10 years in prison, and the maximum the death penalty. Yet, it took a decade and a half before the death penalty was first applied in Chile, in 1890. Capital punishment has never been widely implemented in Chile; although it remained legal for more than a century, until 2001Footnote ²⁹ (the last execution was in 1985).

Some changes were made to the penalties for homicide between 1874 and 2001. While the death penalty had previously been automatically awarded for parricide, in 1970 the sentencing options were changed to between 15 and 20 years of jail as a minimum or the death penalty as a maximum; in the same year, the highest penalty for homicide was changed from execution to a life sentence, which after 20 years could be waived, while the minimum penalty was between 10 and 15 years of jail. Between 1890 and 1985 the death penalty was applied only 57 times (Baeza Reference Baeza2016),Footnote ³⁰ roughly once every two years. Although it was in operation for almost the whole of our period of study, for the next section we created a dummy variable for the period when the death penalty was applied to most homicides, 1890–1969 (excluding parricides); and another for 1890–1985 (including parricides), rather than for the entire period it was legal (1875–2001). Whether the abolition of the death penalty in 2001 contributed to the rising homicide rates of the last two decades remains uncertain.

What is puzzling, though, is that Chile’s convergence toward low homicide rates took place when its per capita GDP was diverging in comparison to the leading world economies. In figure 1 we provided evidence of the relative positions of Chile and the United States only, but these could have been paralleled with those of other leading economies such as the United Kingdom or Germany. However, despite this divergence, Chilean per capita GDP did increase during this period, while inequality decreased (Rodríguez Reference Rodríguez2017). In decadal average, real per capita GDP increased 25 percent, 17 percent, and 17 percent in the 1940s, 1950s, and 1960s, respectively, if compared to the previous decade.

Chile’s position during the last half century has improved dramatically, being now close to those countries with the lowest homicide rates worldwide, an exemplary case of convergence within developing countries. In Latin America, Chile, Argentina, and Uruguay are notable for their comparatively low homicide rates (figure 4). The average homicide rates for the region during the 1960s–1970s were around 12, increasing to 17 and 21 during the 1980s and 1990s–2000s, respectively (Baten et al. Reference Baten, Bierman, Foldvari, Van Zanden, Van Zanden, Jörg Baten, Rijpma, Smith and Timmer2014), between four and eight times higher than in Chile. It is yet to be seen if after the social unrest that started in October 2019 these rates will remain at that comparatively low level.

Figure 4. Homicide rates in some selected Latin American countries. Source: Authors for Chile data. For all others, Clio Infra Project (https://clio-infra.eu/).

When dissecting the national figures at a regional level, we managed to find data for the 1880s–1950s only. We have gathered information for the seven large regions of Chile,Footnote ³¹ shown in figure 5, excluding the Extreme South. There is a huge variation between regions. This is not unusual: it has always been acknowledged that homicide rates can vary importantly across provinces within a country (Gurr Reference Gurr1981). The Extreme South was not included in the figure for ease of visual exposition; homicide rates were too high in this region (reaching 68 and 51 during the 1890s and 1900s, respectively) compared to all others (the South was sparsely populated at that time). Other regions suffering from high rates of interpersonal violence were the Large North and the capital, Santiago. These were the two regions with the most concentrated urban populations in the country. At the opposite extreme, the Little North enjoyed the lowest rate of homicides. During the 1940s and 1950s all regions experienced a decline in homicide rates, as was to be expected, and this decline started a decade earlier in many regions (the Large North, Centre-North, Santiago, South, and Extreme South).

Figure 5. Homicide rates across Chilean regions, 1880s–1950s. Note: The 1880s contain data for 1887–89 only; the 1890s for 1890–94 only, the 1900s, for 1900, 1902, 1904, 1906, 1908; the 1910s, for 1913, 1913, 1915–1918; the 1920s for all years except for 1920; the 1930s and 1940s for all years; while the 1950s contain data only for 1950–51 and 1953.

Descriptive Regression Analysis of Lethal Violence in Chile

We consider how observed homicides described in this article move with other determinants mentioned previously. Given the longitudinal nature of these data, as well as the possibility for additional confounding factors or reverse causalities, we note that this analysis is correlational and exploratory, allowing us to observe how independent variation of a number of factors of interest predicts rates of homicide, as well as to examine whether leads of key socioeconomic factors predict future rates of homicides. To aim to minimize concerns related to reverse causality, we first consider contemporaneous movements between independent variables and homicides, before examining lags of these independent measures, which are entirely predetermined with respect to homicides. We provide descriptive plots of all variables, as well as how these track alongside homicide rates, in appendix 1. To undertake this analysis we estimate models of the following form:

(1) $$\matrix{ {log{{\left( {Homicides} \right)}_t}}{ = {\beta _0} + {\beta _1}Polic{e_t} + {\beta _2}Gin{i_t} + {\beta _3}Pop20 - {{34}_t}} \cr {}{ + {\beta _4}Schoolin{g_t} + {\beta _5}Povert{y_t} + {\beta _6}GDPgrowt{h_t}} \cr {}{ + {\beta _7}Urbanizatio{n_t} + {\beta _8}SocialSpendin{g_t}} \cr {}{ + {\beta _9}DeathPenalt{y_t} + {\varepsilon _t}} \cr } $$

where the independent variable of interest (the natural logarithm of the homicide rate per 100,000 inhabitants) at time t is allowed to depend on a number of covariates also measured at time t. The data used in this article are a time series of homicides, socioeconomic, and social outcomes with a single national-level average for more than 100 years. Given the time-series nature of this data, subnational fixed effects cannot be included in conditional regression models. The covariates included in equation 1 (listed in order) are the rate of police personnel per 100,000 inhabitants, the Gini coefficient based on personal income, the proportion of population aged between 20 and 34 years, the average years of schooling in the country, the poverty rate, the rate of real GDP growth, urbanization, social spending per capita in real terms, and an indicator for whether the death penalty was in place for all homicides. Here we estimate the linear coefficients β ₁ to β ₉ and the constant term β ₀ by Ordinary Least Squares (OLS). The ϵ _t term is an unobserved stochastic component, and standard errors for each estimate are robust to heteroscedasticity. In this specification, we use the natural logarithm of the homicide rate so that each coefficient can be interpreted as the approximate percentage change in rates of homicide with a one unit increase in the independent variable of interest.

To address concerns that independent variables may move in response to homicide rates rather than vice versa, we augment the prior specification to include lags of each variable. Namely, call Determinants _t the full set of righthand side variables in the preceding model (a vector of variables). Similarly, Determinants _(t-1) is the lagged realization of each measure, and Determinants _(t-k) is the k^th lag of each variable. We thus additionally estimate:

(2) $$\matrix{ {log{{\left( {Homicides} \right)}_t} = \;\alpha + {{\boldsymbol \beta} ^\prime }Determinant{s_t} + {{\boldsymbol \gamma} ^\prime }{\mkern 1mu} Determinant{s_{(t - 1)}}} \cr { + {{\boldsymbol \delta} ^\prime }{\mkern 1mu} Determinant{s_{(t - 2)}} + {\eta _t}} \cr } $$

where β, γ, and δ are parameter vectors corresponding to contemporaneous measures, the first lag of these measures, and the second lag of these measures, respectively. Once again, we will estimate this model by OLS, η _t is a stochastic error term, and standard errors are robust to arbitrary forms of heteroscedasticity. Here, we restrict lags to two years for each determinant considered. In practice, adding additional temporal lags has relatively little impact on higher order coefficients. Formal tests of prediction error based on the Akaike information criteria and Schwarz’s Bayesian information criteria suggest optimal lag lengths of between 1 and 3 lags when considering the natural logarithm of the rate of homicides.

Finally, we will address potential uncertainty in measurement by taking entirely predetermined five-year moving averages of each independent variable of interest. This consists of taking averages for each variable listed in the above models over the five years preceding each moment t and regressing the homicide rate in period t on these moving averages.

In table 1 we present correlational evidence from equations 1 and 2. In column 1, we first display equation 1, progressively including additional lags in columns 2 and 3. In column 1 we observe that police personnel per capita, urbanization, and real social spending per capita are all inversely related to the homicide rate. In the case of police personnel, for every 1 unit increase per 100,000 inhabitants, we estimate a reduction in 0.3 percent in the homicide rate. Surprisingly, in this model we note a positive relationship between schooling and rates of homicide, though return to this relationship below when we examine its stability over time.

Table 1. Correlational evidence on long trends in homicide rates

In columns 2 and 3 we add additional lags of each measure, where equation 2 is presented in column 3. In this specification where full lags are included for each relevant measure we now observe that both present and past rates of poverty are negatively related with homicides, and observe some evidence that growth in real GDP is also negatively correlated with homicides (increases in real GDP are conditionally correlated with falling rates of homicides).Footnote ³² Of interest, now police personnel is still negatively correlated with homicide rates, however this obtains principally in lagged models, suggesting some feedback process between additional police presence, and future rates of homicide. For example, a 1 unit increase per 100,000 inhabitants in period t-2 is estimated to be associated with a 0.7 percent decline in homicides in period t.

In table 2 we examine these relationships by period, both prior to, during, and posterior to the sharp reduction in homicides observed between the 1930s and the 1960s. Columns 1 and 2 present regression results for only the pre-1930 period both with and without lagged outcomes and suggest a negative correlation between homicides and each of GDP growth, the rate of poverty, and social spending. Interestingly, this relationship appears to lessen over time. Indeed, in the period of sharp reduction in homicides (1930s–1960s), the only robust correlation obtains when considering police personnel. In this case, we observe that for each additional member per 100,000 inhabitants, rates of homicide are approximately 1 percent lower. This relationship is similarly observed in the post-1960 period.

Table 2. Correlational findings by period

Finally, in table 3 we examine an alternative measure of the independent variables available, namely taking predetermined five-year moving averages of each measure. In general, these similarly point to the changing nature of determinants over time, as observed with regressions in table 2. For example, when taking longer period moving averages, we observe relatively consistent correlations between (higher) inequality as measured by the Gini coefficient and reductions in rates of homicide. We similarly observe relatively consistent correlations between higher social spending and reductions in homicide, particularly in the pre-1960 period, as well as higher GDP growth and reduced rates of homicide. However, correlations between years of schooling and rates of police personnel do not exhibit such stability, limiting our ability to comment on the sign of correlations at least when longer moving average estimates are employed.

Table 3. Homicide rates and predetermined 5-year moving averages of correlates