Abstract

Homicide is a leading cause of death for young people in the United States aged 15–34, but it has a disproportionate impact on one subset of the population: African American males. The national decline in homicide mortality that occurred from 1991 to 2014 thus provides an opportunity to generate evidence on a unique question—How do population health and health inequality change when the prevalence of one of the leading causes of death is cut in half? In this article, we estimate the impact of the decline in homicide mortality on life expectancy at birth as well as years of potential life lost for African American and white males and females, respectively. Estimates are generated using national mortality data by age, gender, race, and education level. Counterfactual estimates are constructed under the assumption of no change in mortality due to homicide from 1991 (the year when the national homicide rate reached its latest peak) to 2014 (the year when the homicide rate reached its trough). We estimate that the decline in homicides led to a 0.80-year increase in life expectancy at birth for African American males, and reduced years of potential life lost by 1,156 years for every 100,000 African American males. Results suggest that the drop in homicide represents a public health breakthrough for African American males, accounting for 17 % of the reduction in the life expectancy gap between white and African American males.

Introduction

From 1991 to 2014, the U.S. homicide rate fell by more than one-half. This trend, which was entirely unanticipated by most criminologists and came after decades of extreme violence across much of urban America, has generated an enormous research literature designed to shed light on why violence fell (e.g., Levitt 2004; Roeder et al. 2015; Zimring 2006). But a second question has received very little empirical attention—How has the “Great American Crime Decline” (Zimring 2006) changed inequality in the United States?

This question is particularly relevant when considering inequality in population health. Homicide is one of the leading causes of death among young people in the United States from early adolescence through early adulthood, but statistics on homicide for the population as a whole obscure the large disparities in death rates by race and gender (Centers for Disease Control and Prevention n.d.). Among 15- to 24-year-olds in 2014, the rate of homicide mortality was 2.4 per 100,000 non-Hispanic whites and 38.6 per 100,000 non-Hispanic African Americans. Within each of these groups, the death rate for males is several times greater than the death rate for females. The extreme racial and gender gaps in homicide mortality mean that the health impact of violence is felt disproportionately by African American males, who represented 44 % of all homicide victims in the United States during 2014 despite making up just 7 % of the population (Federal Bureau of Investigation (FBI) 2015; U.S. Census Bureau 2017).

The same disparities in mortality also mean that African American males benefit most from large-scale drops in the national homicide rate. In this sense, the decline in the U.S. homicide rate that has occurred since the early 1990s has important implications for racial and gender health inequality, and provides an opportunity to generate evidence on a unique question—How does inequality in population health change when the prevalence of a leading cause of death, unevenly distributed across the population, is cut in half? Research has assessed how homicide contributes to disparities in life expectancy, but no prior research has considered how the long-term decline in violence beginning in the 1990s has changed the degree of racial and gender health inequality.

Our goal in this study is to generate evidence on how the long-term decline in homicide mortality has affected trends in life expectancy at birth (LE) for African American and white males and females, respectively. We do so by estimating trends in LE for all four groups from 1991, when the U.S. homicide rate reached its latest peak, through 2014, when this rate had reached its lowest point in several decades (FBI 2017). We then simulate trends in life expectancy at birth for each group under the counterfactual assumption that rates of mortality, by age group, race, gender, and education level, remained at 1991 levels. The same exercises are conducted using an alternative measure that captures the loss of years of life due to premature mortality, years of potential life lost (YPLL) (Gardner and Sanborn 1990).

Results indicate that the decline in homicide mortality from 1991 to 2014 led to a 0.80-year increase in LE for African American males and a reduction in YPLL of 1,156 years for every 100,000 African American males. Because any analysis of changes in life expectancy relies on strong counterfactual assumptions, these results represent descriptive rather than causal evidence revealing how life expectancy changes for different segments of the population when homicide mortality declines. However, sensitivity analyses indicate that our estimates change very little even in the presence of extensive confounding; further, there is good reason to think that the main results are conservative because they do not consider all the potential indirect effects of violence on health.

Our analysis has two main sets of implications. First, it sheds light on a major social trend that has had pronounced consequences for population health and represents one of the most important advancements in the health of African American males in the past several decades. Whereas a tremendous amount of research has examined the causes of the crime drop, very little work has focused on how it has affected inequality in the United States. Second, the evidence brings attention to the potential role of violence prevention in reducing racial and gender health inequality. Violence has never received the same attention or resources as other major causes of mortality and morbidity, and yet confronting violence is one of the few clear pathways by which public policy can address persistent gaps in life expectancy among African American males, one of the most disadvantaged segments of the population (Fuchs 2016).

The Crime Decline and Life Expectancy

The decline in American violence from the early 1990s to 2014 is a trend often associated with policing, criminal justice, and urban policy, but it also has important implications for public health. Although the crime drop is rarely mentioned in discussions or analyses of major trends in population health, we argue that these distinct research literatures should be brought together (Pridemore 2003).

The Fall of Violence

We focus on the years spanning from 1991 to 2014, a period in which the national homicide rate and the rates of other types of violence were cut by at least 50 %. After fluctuating between 7.9 and 10.2 homicides per 100,000 from 1970 through the early 1990s, the U.S. rate of homicides dropped to 4.5 per 100,000 in 2014, the lowest rate of the past 50 years (FBI 2017).1 The fall of violence was experienced differently across U.S. cities, but the homicide rate dropped to some degree in just about every major city across the country. It was a national phenomenon that had its greatest impact on the most violent cities and the most violent neighborhoods across the country (Friedson and Sharkey 2015).

Although a large research literature has examined the causes of the crime drop, there is no single definitive answer about why crime began to fall, and few explanations have received unequivocal empirical support. Zimring (2006) focused on the drop in violence during the 1990s and argued that it can be explained by a combination of improving economic conditions, shifts in the age of the population, changes in policing tactics, and the expansion of the incarcerated population. Levitt (2004) argued that growth in law enforcement, increases in incarceration, the end of the crack cocaine epidemic, and the legalization of abortion account for the decline of violence. A recent report from the Brennan Center for Justice included a state-level analysis of 13 factors that have been proposed as explanations for the crime drop and argued that factors such as the growth of police departments and the decline of alcohol consumption likely played meaningful roles in explaining state-level variation in the scale of the crime drop (Roeder et al. 2015).

In addition to these broad overviews, a set of recent studies has provided convincing causal evidence on two particular factors. The first is the decline in lead exposure. Lead-based paint in residential settings has become much less common after regulatory changes limited its use beginning in the 1950s and then banned its use for residential purposes in 1978. Similarly, the Clean Air Act of 1970 led to an almost complete removal of lead from gasoline between 1975 and 1985. As a result of these regulations, the average level of lead in the blood of Americans declined rapidly, and uniformly across all demographic groups, from the mid-1970s to the early 1990s (Reyes 2007). Aizer and Currie (2017) provided strong evidence suggesting that drops in blood lead levels have a causal effect on behavioral problems and early delinquency, and Reyes (2007) argued that the long-term drop in lead exposure likely had a large impact on levels of violence.

Sharkey et al. (2017), on the other hand, brought attention to the proliferation of community nonprofit organizations in the 1990s and presented causal evidence that these organizations, which focused on controlling crime, providing social supports, and providing opportunities and resources for youth, had a substantial impact on violence. Rigorous evaluations of specific types of organizations provided additional guidance on programs that have been shown to be effective in reducing violence. The establishment of Business Improvement Districts, the development of organizations that offer cognitive behavioral therapy combined with after-school sports or summer jobs programming, and efforts to “green” abandoned lots throughout cities have all been shown to lead to substantial reductions in violent crime in experimental or rigorous quasi-experimental evaluations (Branas et al. 2011; Cook and MacDonald 2011; Cook et al. 2015; Heller 2014).

The extensive literature on the causes of the crime drop does not lead to a simple conclusion about the precise set of factors that explain this trend, but the literature does provide guidance for social policy designed to reduce violence further. Efforts to improve public health by reducing exposure to lead and excessive alcohol consumption are likely to have spillover effects on violence, and investments in an array of community-based programming also represent a promising approach to confronting violent crime. Our central argument in this article is that these approaches should also be thought of as parts of an overall strategy to reduce inequalities in population health.

Trends in Life Expectancy at Birth

The second relevant research literature has focused on trends in life expectancy in the United States, with a particular focus on variation by race and gender. Descriptive research on trends has documented steady growth in LE for all groups. Gaps between whites and African Americans remain large but have narrowed over time. From 1995 to 2014, LE for African Americans has risen more than twice as much as for whites, increasing by 6 years compared with 2.5 years for whites (Fuchs 2016). The greatest gains in LE have been experienced by African American men and by the most vulnerable segment of the African American population (that is, individuals in the bottom half of the survivor distribution).

A set of studies provides evidence on the specific causes of death that help explain trends in LE for different groups as well as gaps in LE across groups. Focusing on the period before homicide began to fall, Harper et al. (2007) found that increases in homicide deaths accounted for 26 % of the two-year increase in the life expectancy gap between black and white males from 1983 to 1993. Homicide was the second-leading contributor to the increase in this gap, exceeded only by HIV, which accounted for more than 50 % of the increase. In the period from 1993 to 2003, on the other hand, Harper et al. (2007) found that the gap between black and white males narrowed by 2.1 years. Declines in homicide were the leading cause of this narrowing, accounting for 0.6 years, while declines in HIV deaths were the second leading cause. In a similar vein, Geronimus et al. (2011) found that the excess death rate of working-age black males (in comparison with whites) declined by 14 % from 1980 to 2000 and that the homicide drop of the 1990s accounted for 62 % of this decline.

Firebaugh et al. (2014) further decomposed changes in the black–white life expectancy gap from 2000 to 2010 to assess the relative importance of changes in causes of death versus changes in age of death due to various causes. They showed that most of the decline in the racial gap was due to delays in the age of death from major causes such as heart disease, cancer, and cerebrovascular disease. The rest of the narrowing of the gap was due to shifts in the distribution of the causes of death. Homicide played little role in explaining the drop in racial gap in LE that Firebaugh et al. documented because the analysis focused only on 2000 to 2010, a period in which homicide rates fell less substantially than in the prior decade.

Focusing on current gaps in LE rather than trends, Kochanek et al. (2013) estimated that LE in 2010 was 3.8 years lower for African Americans than for whites, primarily because of greater rates of mortality from heart disease, cancer, homicide, diabetes, and perinatal conditions. Homicide accounted for 0.50 years of this gap and was its third-leading contributor, trailing only heart disease and cancer. African American males were estimated to live an average of 4.7 years less than white males, due to the same factors plus stroke. Homicide accounted for 0.87 years of this gap among males and was its second-leading contributor, trailing only heart disease. This finding is consistent with longer-term patterns: heart disease and homicide were likewise found to be the two leading contributors to the racial gap in life expectancy among males in 1980 (Keith and Smith 1988) and were two of the three leading contributors to this gap (along with cancer) in 1998 (CDC 2001).

Among the full list of significant causes of death, Fuchs (2016) showed that homicide has the second greatest racial disparity in death rates, trailing only HIV infection (see also Harper et al. 2012). In 2014, the age-adjusted death rate for homicide among African Americans was almost six times as high as for whites. According to a CDC analysis (2017), homicide was the leading cause of death for black males aged 15–34, accounting for almost one-half of deaths between ages 15 and 24 and almost one-third of deaths between ages 25 and 34. The next leading cause, unintentional injuries, accounted for less than one-quarter of deaths. By contrast, homicide was the third leading cause of death for white males aged 15–24, accounting for approximately 7 % of deaths, and trailing unintentional injury and suicide.

Our review of the literature leads to two main conclusions. First, trends in life expectancy show that racial and gender gaps have narrowed considerably over time, and declines in homicide mortality played an important role in contributing to the reduction in health inequality between black and white men. Whereas the rise of homicide mortality in the 1980s increased the racial gap in LE among males, the drop in homicide in the 1990s reduced this gap. In this article, we build on this literature, using refined methods and focusing our attention on the full period of 1991–2014, over which the national homicide rate fell dramatically. The trend of declining violence over these years provides a unique chance to assess how racial and gender health inequality changes when a leading cause of death that is unevenly distributed across the population is cut by more than half. We analyze how the drop in the national murder rate that occurred from the most recent peak of violence to the most recent trough affected LE among white and African American men and women as well as what role it played in contributing to the narrowing of racial and gender gaps in LE.

Despite the progress in reducing racial and gender gaps in LE, the second conclusion from our review of the literature is that substantial gaps persist, and black males continue to have by far the lowest life expectancy among these four groups. Fuchs (2016) argued that reducing homicide is one of the few clear pathways to further reducing the racial gap in life expectancy. Our conclusions from this analysis thus have implications beyond the goal of better understanding trends in population health, and they bear directly on the question of how to confront racial and gender gaps in health moving forward.

Methods

Analytic Strategy

For each calendar year in the analysis, we use a standard actuarial procedure to estimate LE at birth for African American males and females and white males and females (Phillips n.d.). This procedure consists of using an iterative process to construct a life table that predicts the lifespans of 100,000 hypothetical individuals, on the basis of age-, race-, and gender-specific mortality rates. Deaths during each year of age (x), from 0 to 99, are calculated as
dx=lx×qx,
1
where lx is the number of survivors to age x; qx is an individual’s risk of dying between reaching ages x and x + 1; and dx is the number of individuals dying between reaching ages x and x + 1.2
Assuming 100,000 live births (i.e., l0 = 100,000), Eq. (2) shows the calculation of the number of survivors to each year of age, from 1 to 100:
lx=lx1dx1.
2
For all deaths occurring beyond the first year of life, death is assumed to occur exactly halfway through the year of life in which the individuals die. Those dying in their first year are assigned the average lifespan of all individuals who die in their first year (calculated using CDC data). Survivors to age 100 are assigned the sex- and race-specific life expectancy at age 100 given in CDC data. Life expectancy at birth is calculated as
LE0=c=099dxx+cx+l100×100+LE100100,000,
3
where LE0 is life expectancy at birth; cx equals 0.5 for all values of x from 1 to 99; c0 equals the average lifespan of those dying in their first year of life; and LE100 is life expectancy at age 100.
In a second step, we recalculate life expectancies at birth under the counterfactual assumption that mortality rates due to homicide stayed fixed at 1991 levels (Carey 1989; Lai and Hardy 1999). Under this counterfactual scenario, the proportion of the population (at each age) that would not have died anyway from some cause besides homicide is subject to an increased risk of mortality from homicide, equal to the difference between the 1991 and actual homicide mortality rates. Accordingly, we calculate race- and gender-specific counterfactual mortality rates for each year of age (x), as in Eq. (4):
mx=mx+(MxMx×1mxMx),
4
where mx and mx, respectively, are the counterfactual and actual mortality rates; and Mx and Mx, respectively, are the counterfactual (i.e., 1991) and actual rates of mortality from homicide. We then convert these counterfactual mortality rates to counterfactual risks of dying (qx) and construct life tables based on these risks.
We estimate YPLL for actual homicide deaths and for those that would take place under the counterfactual. YPLL from homicide per 100,000 births is calculated for each race/gender group, standardized for the population’s age distribution to allow comparisons over time and between groups (Aragón et al. 2008). This calculation is shown in Eq. (5):
YPLL=x=084YPLLx×wx,
5
where YPLLx is years lost from homicide per 100,000 persons of age x, and wx is the proportion of U.S. residents under age 85 who are of age x.3 We calculate YPLLx as follows:
YPLLx=Dx×LExCxpx×100,000,
6
where Dx is the number of homicide deaths at age x;4LEx is life expectancy at age x;5Cx is the average time (in years) since the last birthday of a homicide victim of age x;6 and px is the population of age x. In this formulation, a homicide death is considered premature by LEx– Cx, the average life expectancy of a person at the age when the death took place (Aragón et al. 2008). Deaths at age 85 or older should not be considered premature (given current life expectancies at birth) and thus are not considered in determining YPLL from homicide.

We use Monte Carlo simulations to estimate the standard errors of LE and YPLL (Mooney 1997). In each iteration of our simulations, the input variables (i.e., mortality rates) are randomly assigned values based on the probability distributions implied by their standard errors. We calculate confidence intervals pertaining to the LE and YPLL estimates, and for differences between the two estimates (e.g., actual LE and counterfactual LE).

Counterfactual estimates of LE and YPLL are carried out under two alternative sets of assumptions. For the first set of models, we follow prior literature and assume that homicide victims would have the same LE as others of the same age, race, and gender (Harper et al. 2007; Levine et al. 2001; Redelings et al. 2010). Although this approach is standard in the literature, we argue that it is likely to generate results biased away from 0. Homicide victims are drawn disproportionately from segments of the population with lower-than-average educational attainment (Wong et al. 2002), who have lower LE than the rest of the population within race and gender groups (Meara et al. 2008; Olshansky et al. 2014; Wong et al. 2002). If educational attainment of homicide victims is not taken into account, the estimated effect of the decline in homicide will be overestimated.

In a second set of models, we assume that homicide victims had the same LE as others of the same race, gender, and level of educational attainment. We use the following educational attainment categories: did not finish high school, completed high school but did not attend college, and attended college. Applying the same methods as outlined earlier, we calculate the actual and counterfactual number of deaths for each year of age, from 20 to 74, and the number of individuals surviving in each year, using rates of mortality specific to each educational attainment category and the proportion of the population in each category.7

This approach is still subject to bias if homicide victims have lower expected LE within age/gender/education categories, as is likely. However, we argue for our approach on three separate grounds. First, main results are not sensitive to violations of our assumption. Even if homicide victims have potential LE that is substantially lower than individuals of the same race, gender, and education level, our point estimates change only slightly, and the main patterns of results do not change. Results from the sensitivity analysis are discussed later. Second, our approach goes beyond the standard practices in the literature, which assume that individuals who die from particular causes of death would have had the same LE as others of the same race and gender. Third, from a practical standpoint, it is not possible to gather more detailed data on mortality rates within education groups. We are unable, for instance, to obtain detailed, age-specific data on mortality due to homicide within income groups.

Although we have taken steps to go beyond the existing literature and minimize the degree of bias in our estimates, this is a descriptive exercise revealing how LE and YPLL would change, in a mechanical manner, if homicide mortality never fell from the levels of 1991. The estimates are vulnerable to potential bias away from 0 due to unobserved confounding. As an example, declines in lead exposure may affect homicide rates and may also affect life expectancy through other health-related pathways. Sensitivity analyses show that even extensive confounding of this type would not lead to meaningfully different results, but it is impossible to know the degree of confounding precisely.

The estimates from our analysis may also be biased toward 0 because we are considering only the direct effects of homicide mortality and thus are ignoring the indirect pathways by which violence affects stress, educational attainment, economic outcomes, and the health of victims and everyone connected to them. Considering the range of potential indirect pathways connecting the decline in violence to the long-term health of victims, family members, peers, and neighbors, it is reasonable to think that our estimates may be quite conservative.

The results should not be interpreted as causal estimates of the impact of violence on health but instead as providing descriptive evidence revealing how life expectancy changes for different segments of the population when homicide mortality declines. For simplicity and to avoid awkward language, we continue to use language that connotes causal relationships, but we emphasize that the results should be thought of as causal only if the set of assumptions behind our model are valid.

Data

Annual mortality rates—for each race/gender at each year of age from 0 to 74—were calculated by dividing number of deaths in that group by its population. Death counts were obtained from the Multiple Cause of Death mortality data published by the Centers for Disease Control and Prevention (CDC). Population counts consist of the bridged-race population estimates provided on the CDC’s website.8 Race- and gender-specific mortality rates for ages 75–99, as well as life expectancies at age 100, were derived from data in the CDC’s U.S. life tables publication series.9,10 Annual rates of mortality due to homicide, by race and gender, were calculated by dividing counts of homicide deaths (from the Multiple Cause of Death data) by population counts (from the bridged-race population estimates). Deaths were coded by race in accordance with the classification given in the CDC data source, in which all decedents are designated as being white, black, or members of other racial groups.11 A distinct homicide mortality rate was calculated for each year of age from 0 to 84, while a single rate was determined for ages 85 and over.12

The analysis using educational attainment requires counts by educational attainment of age-specific populations, total deaths, and homicide deaths in each race and gender group. Population counts by educational attainment were estimated by multiplying a race and gender group’s total population (from the bridged-race population estimates) by the proportion of that group in each educational attainment category (from Current Population Survey estimates of educational attainment in the United States by age, race, and gender),13 for each year of age. Counts of total deaths and homicide deaths by age, race, gender, and educational attainment were obtained from the aforementioned Multiple Cause of Death data. Deaths with missing educational attainment data were distributed into the educational attainment categories according to the proportions of decedents (by age, race, and gender) with known levels of educational attainment in these categories.14

Results

Life Expectancy at Birth

The top panel of Table 1 displays estimated LE for the four subgroups in 2014. In the left columns of the table are estimates of LE based on actual mortality rates in 2014, which show that white females had the highest LE, followed by African American females, white males, and African American males. The middle columns of the table show estimates of LE under the counterfactual condition in which mortality rates due to homicide remained at 1991 levels. Results presented in the top panel of Table 1 are based on the assumption that homicide victims would have the same LE as individuals of the same race and gender. The columns on the right show the estimated impact of the drop in mortality due to homicide on LE for each group.

We find that the decline of homicide-specific mortality led to increases in LE of 0.04 years (95 % confidence interval = 0.01 to 0.08) for white females, 0.14 years (95 % confidence interval = 0.10 to 0.17) for white males, 0.28 years (95 % confidence interval = 0.18 to 0.37) for African American females, and 1.00 years (95 % confidence interval = 0.89 to 1.10) for African American males.

The estimates in the top panel of Table 1 are based on the most complete data available and are estimated with the greatest precision possible. However, the assumption that homicide victims have the same LE as other members of the same race and gender likely leads to bias away from 0 in our counterfactual estimates.

Results in the bottom panel of Table 1 rely on the more conservative assumption that homicide victims would have the same LE as individuals of the same race, gender, and education group. Under this assumption, point estimates of gains in LE from the homicide drop are slightly smaller for all groups, and the variance of the estimates is larger, making our estimates less precise. The impact of the homicide decline for white males and females is small in magnitude, and the impact for African American females is slightly larger but is estimated imprecisely. We cannot reject the null hypothesis that the impact of the homicide decline on LE for African American females is 0. For African American males, we estimate that the effect of the decline in homicide mortality is to increase LE by 0.80 years (95 % confidence interval = 0.45 to 1.15).

Figure 1 extends these results by displaying actual trends in LE over the full period from 1991 to 2014 (solid lines), as well as trends in LE under the counterfactual condition in which mortality rates due to homicide remained at 1991 levels (dashed lines). The results presented in the figure display counterfactual life expectancy trends under the assumption that homicide victims would have the same LE as individuals of the same race, gender, and education group. For African American and white females, and for white males, there is minimal difference between trends in life expectancy with and without the drop in homicide mortality. Homicide is not a common cause of death for these groups, at any age. For African American males, on the other hand, the trend in life expectancy would look noticeably different if the national rate of homicide did not decline from 1991 to 2014.

Over this period, the gap in LE between white and African American males dropped from 8.21 years in 1991 to 4.17 years in 2014. We estimate that if mortality due to homicide had not fallen over this period, the gap between white and African American males would have been 4.86 years in 2014.

Years of Potential Life Lost

Table 2 shows actual and counterfactual YPLL for each of the four groups under the assumption that homicide victims have the same LE as members of the same race and gender (top panel) and the more conservative assumption that homicide victims have the same LE as those with the same race, gender, and education level (bottom panel). We focus our attention on the latter estimates.

We estimate that the decline of mortality due to homicide reduced YPLL by 46 years for every 100,000 white females (95 % confidence interval = –53 to –40), by 160 years for every 100,000 white males (95 % confidence interval = –170 to –151), by 321 years for every 100,000 African American females (95 % confidence interval = –353 to –289), and by 1,156 years for every 100,000 African American males (95 % confidence interval = –1,250 to –1,063).

For all four groups, we can reject the null hypothesis of no change in YPLL due to the reduction of homicide. However, by far the largest impact is found among African American males. For this group, YPLL would have been 93 % higher if mortality due to homicide had not declined from 1991 to 2014.

Discussion

Homicide is a unique public health challenge because its victims are typically young, mostly under the age of 34. For this reason, drops in mortality due to homicide have a substantial impact on LE and on YPLL (FBI 2015; Fox and Zawitz 2017). In the United States, the impact of homicide is felt disproportionately by one segment of the population: young African American males. These two features of homicide mean that any overall change in the prevalence of homicide mortality will have its greatest effect on a group that has had persistently low LE, and thus has important implications for racial and gender health inequality. Analyzing how the long-term drop in homicide mortality affects LE of different groups allows us to answer a question that is crucial to understanding disparities in population health—How does health inequality change when the prevalence of one of the leading causes of death for a specific group (in this case, homicide) is cut in half?

Our results suggest that the drop in homicide represents a public health breakthrough for African American males, adding 0.80 years to life expectancy at birth and accounting for 17 % of the reduction in the LE gap between white and African American males from 1991 to 2014.15 Our estimates indicate that if homicide had not declined, more than 1,100 years of potential life would have been lost for every 100,000 African American males. Although prior research has examined the role of homicide in contributing to gaps in LE, the analysis in this article is the first to track trends in LE over the full period in which violence fell across the country.

To provide a sense of the magnitude of the improvement in life expectancy of African American males generated by the decline of homicide, it is useful to compare the estimated impact from this study with the impact of other major changes in public health. Olshansky et al. (2005), for instance, conducted a set of simulations designed to estimate how LE for different groups would be expected to change if the problem of rising obesity in the United States ended and the level of obesity went to 0. According to their estimates, if obesity disappeared, LE of the population would rise by between one-third and three-quarters of a year, and LE for African American men would rise by between 0.30 and 1.08 years. From this perspective, the decline in the homicide rate is associated with an improvement in LE of African American men that is similar to or greater than the potential impact of eliminating obesity among this segment of the population.

Comparisons to other social trends reinforce the argument that the fall of violence is one of the most important developments in population health over the past several decades. For example, Stewart and Cutler (2014) estimated that population LE increased by 1.26 years and 0.43 years, respectively, due to reductions of about 50 % in the prevalence of cigarette smoking and in motor vehicle accident fatalities between the 1960s and 2010.16 Another recent analysis estimated that increases in drug poisoning deaths reduced population LE by 0.28 years between 2000 and 2015 and that at least 75 % of this increase was associated with opioids (Dowell et al. 2017). Our results thus suggest that the improvement in LE of African American males due to the drop in violence is slightly smaller than the improvement in population LE driven by the long-term drop in smoking, but it is substantially larger than the improvement driven by the drop in motor vehicle deaths and the reduction in LE driven by the much-publicized opioid epidemic.

Although the results from the analysis are striking in their implications, they should be interpreted with a clear sense of the limitations of the study—some of which are shared by all research estimating the impact on longevity of specific causes of death, and some of which are unique to our analysis.

First, our estimates rely on strong assumptions about the counterfactual world in which homicide mortality remained at 1991 levels, and there are good reasons to expect that our estimates are biased away from 0 because of unobserved confounding. As an example, declines in lead exposure may affect homicide rates and may also improve life expectancy through other health-related pathways. By contrast, the rise of incarceration is a trend that may have had a negative causal effect on levels of violence yet may also have had negative effects on health, leading to bias in the opposite direction as the reduction in lead exposure.17

This type of potential for bias due to confounding is present in any analysis of the impact of one specific cause of death on LE. We have gone further than prior research in our efforts to make more reasonable assumptions about life expectancy of individuals who are likely victims of homicide by adjusting for race, age, gender, and education level in generating counterfactual estimates of LE. However, our methods remain vulnerable to confounding. We assessed the sensitivity of our results to this type of confounding and found that even if it were extensive, it would not lead to meaningfully different results.18

Second, our estimates are likely to contain bias toward 0—that is, to be conservative—because they consider only the direct effects of homicide on longevity through the loss of life. Individual experience with all forms of interpersonal violence as well as indirect exposure to violence is linked with cognitive skill development, stress, and mental health (Coker et al. 2002; Fowler et al. 2009; McEwen 2000; Sharkey 2010). Through these and other pathways, the direct and indirect impact of the national decline of violence that began in the early 1990s is likely to be substantially larger than the impact due solely to declines in homicide mortality.

A third key limitation of this study is that we have presented a single national estimate of the impact of the decline in homicide on LE. However, the decline in violence was not spread evenly across the country. The impact on LE is undoubtedly much larger in cities that experienced the most substantial drops in homicide, such as Fort Worth, Los Angeles, New York City, and Washington, DC. Further, the concentration of violence in specific neighborhoods within cities means that our national estimates mask tremendous variation in the public health impact of the crime decline (Friedson and Sharkey 2015). More research is necessary to understand how levels and trends in homicide mortality affect LE in specific cities and neighborhoods where violence remains an urgent problem. Moving outward from the United States, this research should motivate similar analyses in other nations that have experienced declines in violence over a roughly similar timeframe, including England and Wales, Canada, and several other European nations (Farrell et al. 2014).

With these limitations in mind, we argue that the analysis provides new evidence to assess how the trend of declining violence has affected racial and gender inequality in one of the core indicators of population health. But beyond the contribution to scholarly understanding of trends in life expectancy, we argue that our results should bring added urgency to the effort to confront violence as a central public health challenge in the United States (Pridemore 2003). Data from 2014 indicate that homicide remains the leading cause of death for African American males aged 15–34 and is the fifth leading cause of death for all African American males (CDC 2017). And yet homicide does not garner the same attention or resources as other major public health challenges. In 2017, the U.S. National Institutes of Health designated about $65 million for research on homicide, youth violence, and violence prevention combined (National Institutes of Health 2017). To provide a point of comparison, more than $900 million was designated for obesity research.

Confronting violence requires research designed to refine scholarly understanding of why violence has fallen as well as research evaluating interventions to reduce violence. Although there is no single accepted explanation for why homicide has fallen since the 1990s, a range of factors have received at least some empirical support, including the following: changes in the size of police forces and in policing tactics, the decline of the crack cocaine epidemic, the rise of incarceration, the steep drop in children’s exposure to lead, changes in abortion law, shifts in the age structure of the population, improvements in the national economy and associated drops in joblessness, and declines in alcohol use (Aizer and Currie 2017; Chalfin and McCrary 2013; Levitt 2004; Roeder et al. 2015; Zimring 2006). This set of possible explanations provides little clarity on how to confront violence, but the approaches used to confront violence in the present are not constrained by the forces that have reduced violence in the past. Efforts to reduce alcohol consumption and lead exposure are likely to generate spillover effects on violence, and evidence from a wide range of interventions—including cognitive behavioral therapy for youth, enhanced summer jobs programs, and “greening” of abandoned city lots—provides novel models for how to confront violence moving forward (Branas et al. 2011; Cook et al. 2015; Heller 2014).

Recent trends in homicide reinforce the call for urgency. After more than two decades of falling violence, the U.S. national homicide rate rose sharply (by 10 %) in 2015 and rose again in 2016, and the increase was most pronounced for African American men (FBI 2016).19 Just as the decline in homicide mortality had its greatest impact on African American men, any increase in violence will have its greatest impact on the same group and will exacerbate existing disparities in health. In short, confronting violence is crucial to addressing persistent racial and gender inequality in health. The results from our analysis and the rise in violence since 2014 should bring renewed attention and resources to the effort to preserve, and extend, one of the most important public health improvements of the past several decades.

Acknowledgments

We thank Amar Hamoudi for taking the time to review and talk through our methods and code in detail. His feedback and guidance were enormously valuable. Thanks also to Glenn Firebaugh, Robert Sampson, and Larry Wu for helpful feedback on the project.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Notes

1

The national homicide rate rose to 4.9 murders per 100,000 residents in 2015 (FBI 2016).

2

Risk of death for each age (x) is calculated as

qx = 1 – exp(–mx),

where qx is risk of death, and mx is the mortality rate (Phillips n.d.).

3

These proportions were calculated using estimates of the 2005 U.S. population by single years of age, obtained from https://wonder.cdc.gov/bridged-race-population.html.

4

This value is obtained by multiplying the rate of mortality from homicide at age x by the population at age x.

5

Life expectancy at age x is calculated by dividing the total number of years that survivors to age x live (after reaching age x) by the number of such survivors from among the 100,000 live births assumed when constructing a life table.

6

This value is assumed to equal 0.50 years for victims aged 1–84 and 0.48 years (calculated based on CDC data on the timing of infant homicide deaths) for deaths before age 1.

7

We obtain the number of excess deaths (in the counterfactual vs. the actual scenario) before age 20 in each educational attainment group by multiplying the total number of such excessive deaths before age 20 by the proportion of homicide victims aged 20–74 in that group.

9

In its life tables, the CDC adjusts mortality statistics for the oldest ages using models that combine data from the CDC’s vital statistics, the U.S. Census, and Medicare. For further details, see Arias (2012).

10

For the years 2001–2013, we obtained mortality statistics regarding those aged 75 or older from the CDC’s annual life tables. For 1991–2000, we obtained these mortality statistics by applying linear interpolation to estimates from the U.S. Decennial life tables for 1989–1991 (treated as applying to 1990) and the U.S. Decennial life tables for 1999–2001 (treated as applying to 2000). Because annual life tables for 2014 were not available when we conducted our analysis, we applied 2013 mortality statistics regarding those aged 75 or older to this year.

11

The CDC obtains death records from states. Some states report a single race per subject, while others report one or more races. When multiple races are reported, the CDC uses a bridging procedure to determine a single racial classification. See CDC (2004) for a summary of this procedure. For consistency, we use population counts that were bridged by the CDC using the same procedure, when we calculate mortality rates.

12

Our data source for population counts (the CDC’s online bridged-race population estimates) makes available a single estimate of the population aged 85 and over (of each race/gender), and a separate estimate for each year of age from 0 to 84.

14

We tested our results’ sensitivity to this means of handling missing educational data by running our models with two alternative approaches to handling missing data: (1) excluding deaths with missing educational attainment data from the calculation of mortality rates and reducing population counts (used in calculating these rates) by the percentage of all deaths that were so excluded; and (2) assigning all deaths with missing educational attainment data to the “less than high school” educational attainment category. (The latter approach is conservative because it increases homicide deaths under the counterfactual scenario among the population’s least-educated segment, which has the shortest lifespan.) Using these alternative approaches, we obtained results similar to those in our model (see Table 1, panel B). For the first alternative approach, the difference between actual and counterfactual LE for black males was 0.715 years (95 % confidence interval = 0.360 to 1.070); for the second alternative approach, this difference was 0.705 years (95 % confidence interval = 0.353 to 1.058).

15

Firebaugh et al. (2014) found that declines in homicide from 2000 to 2010 had little impact on the black–white gap in life expectancy. Consistent with this finding, Fig. 1 shows that most of the effect of declining homicide rates on this gap occurred between 1993 and 2000.

16

These estimates pertain to LE at age 25.

17

We should be clear that both the effect of incarceration on violence and the overall effect of rising incarceration on population health are not entirely settled. The rise of incarceration through the 1980s likely had some negative impact on violent crime, although the magnitude is not clear. The continuing increase in incarceration in the 1990s is thought not to have had a large impact on crime (see National Research Council 2014). Incarceration has been found to have short-term positive effects on the health of inmates but may have long-term negative effects on individuals who have experienced incarceration as well as on their family members and friends, and particularly on their children (see Wildeman and Wang 2017).

18

We tested our results’ sensitivity to the assumption that LE of potential homicide victims is equivalent to that of others in the same race/gender/educational attainment groups in two ways. First, we assumed that unmeasured characteristics of homicide victims would lead to their having lifespans that were, on average, five years shorter than other individuals in the same age/race/gender/education groups. Second, we assumed the difference between LE of homicide victims and LE of others in the same age/race/gender/education groups would be equivalent to the difference in LE between the highest and lowest educational attainment categories for individuals of the same race/gender. Using these alternate assumptions, our results were similar to those in our original models (see Table 1, panel B). Under the first alternative assumption, the counterfactual LE of black males was 71.763 years (95 % confidence interval = 71.515 to 72.012), resulting in a gap between actual and counterfactual LE of 0.745 years (95 % confidence interval = 0.397 to 1.093). Under the second alternative assumption, African American males’ counterfactual LE was 71.806 years (95 % confidence interval = 71.555 to 72.056), resulting in a gap between actual and counterfactual LE of 0.702 years (95 % confidence interval = 0.352 to 1.052).

19

There were a total of 1,494 more murder victims, and 906 more African American male murder victims, nationally in 2015 than in 2014 (FBI 2016).

References

References
Aizer, A., & Currie, J. (
2017
).
Lead and juvenile delinquency: New evidence from linked birth, school and juvenile detention records
(NBER Working Paper No. 23392).
Cambridge, MA
:
National Bureau of Economic Research
. Retrieved from http://www.nber.org/papers/w23392
Aragón, T. J., Lichtensztajn, D. Y., Katcher, B. S., Reiter, R., & Katz, M. H. (
2008
).
Calculating expected years of life lost for assessing local ethnic disparities in causes of premature death
.
BMC Public Health
,
8
(
116
). https://doi.org/10.1186/1471-2458-8-116
Arias, E. (
2012
).
United States life tables, 2008
(National Vital Statistics Reports Vol.
61
, No.
3
).
Hyattsville, MD
:
National Center for Health Statistics
. Retrieved from https://www.cdc.gov/nchs/data/nvsr/nvsr61/nvsr61_03.pdf
Branas, C. C., Cheney, R. A., MacDonald, J. M., Tam, V. W., Jackson, T. D., & Ten Have, T. R. (
2011
).
A difference-in-differences analysis of health, safety, and greening vacant urban space
.
American Journal of Epidemiology
,
174
,
1296
1306
. 10.1093/aje/kwr273.
Carey, J. R. (
1989
).
The multiple decrement life table: A unifying framework for cause-of-death analysis in ecology
.
Oecologia
,
78
,
131
137
. 10.1007/BF00377208.
Centers for Disease Control and Prevention (CDC)
. (
2001
).
Influence of homicide on racial disparity in life expectancy—United States, 1998
.
Morbidity and Mortality Weekly Report
,
50
,
780
783
. Retrieved from https://www.cdc.gov/mmwr/preview/mmwrhtml/mm5036a2.htm
Centers for Disease Control and Prevention (CDC)
. (
2004
).
NCHS procedures for multiple-race and Hispanic origin data: Collection, coding, editing, and transmitting
.
Atlanta, GA
:
Division of Vital Statistics, National Center for Health Statistics, CDC
. Retrieved from https://www.cdc.gov/nchs/data/dvs/Multiple_race_documentation_5-10-04.pdf
Centers for Disease Control and Prevention (CDC)
. (
2017
).
Leading causes of death in males and females—United States, 2014
.
Atlanta, GA
:
CDC
. Retrieved from https://www.cdc.gov/healthequity/lcod/index.htm
Centers for Disease Control and Prevention (CDC), National Center for Injury Prevention and Control
. (n.d.).
10 leading causes of death by age group, United States—2013
.
Atlanta, GA
:
National Vital Statistics System, National Center for Health Statistics, CDC
. Retrieved from https://www.cdc.gov/injury/wisqars/pdf/leading_causes_of_death_by_age_group_2013-a.pdf
Chalfin, A., & McCrary, J. (
2013
).
The effect of police on crime: New evidence from U.S. cities, 1960–2010
(NBER Working Paper No. 18815).
Cambridge, MA
:
National Bureau of Economic Research
. Retrieved from http://www.nber.org/papers/w18815.pdf
Coker, A. L., Davis, K. E., Arias, I., Desai, S., Sanderson, M., Brandt, H. M., & Smith, P. H. (
2002
).
Physical and mental health effects of intimate partner violence for men and women
.
American Journal of Preventive Medicine
,
23
,
260
268
. 10.1016/S0749-3797(02)00514-7.
Cook, P. J., Dodge, K., Farkas, G., Fryer, R. G., Guryan, J., Ludwig, J., . . . Steinberg, L. (
2015
).
Not too late: Improving academic outcomes for disadvantaged youth
(Working Paper No. 15-01).
Evanston, IL
:
Northwestern University, Institute for Policy Research
. Retrieved from https://www.ipr.northwestern.edu/publications/docs/workingpapers/2015/IPR-WP-15-01.pdf
Cook, P. J., & MacDonald, J. (
2011
).
Public safety through private action: An economic assessment of BIDS
.
Economic Journal
,
121
,
445
462
. 10.1111/j.1468-0297.2011.02419.x.
Dowell, D., Arias, E., Kochanek, K., Anderson, R., Guy, G. P., Jr., Losby, J. L., & Baldwin, G. (
2017
).
Contribution of opioid-involved poisoning to the change in life expectancy in the United States, 2000–2015
.
JAMA
,
318
,
1065
1067
.
Farrell, G., Tilley, N., & Tseloni, A. (
2014
).
Why the crime drop?
Crime and Justice
,
43
,
421
490
. 10.1086/678081.
Federal Bureau of Investigation (FBI), Uniform Crime Reporting Program
. (
2015
).
Crime in the United States 2014
.
Washington, DC
:
FBI
. Retrieved from https://ucr.fbi.gov/crime-in-the-u.s/2014/crime-in-the-u.s.-2014
Federal Bureau of Investigation (FBI), Uniform Crime Reporting Program
. (
2016
).
Crime in the United States 2015
.
Washington, DC
:
FBI
. Retrieved from https://ucr.fbi.gov/crime-in-the-u.s/2015/crime-in-the-u.s.-2015
Federal Bureau of Investigation (FBI), Uniform Crime Reporting Program
. (
2017
).
UCR data tool
[Data set]. Retrieved from https://www.ucrdatatool.gov
Firebaugh, G., Acciai, F., Noah, A., Prather, C. J., & Nau, C. (
2014
).
Why the racial gap in life expectancy is declining in the United States
.
Demographic Research
,
31
,
975
1006
. 10.4054/DemRes.2014.31.32.
Fowler, P. J., Tompsett, C. J., Braciszewski, J. M., Jacques-Tiura, A. J., & Baltes, B. B. (
2009
).
Community violence: A meta-analysis on the effect of exposure and mental health outcomes of children and adolescents
.
Development and Psychopathology
,
21
,
227
259
. 10.1017/S0954579409000145.
Fox, J. A., & Zawitz, M. W. (
2017
).
Homicide trends in the United States
.
Washington, DC
:
U.S. Department of Justice, Office of Justice Programs, Bureau of Justice Statistics
. Retrieved from https://www.bjs.gov/content/pub/pdf/htius.pdf
Friedson, M., & Sharkey, P. (
2015
).
Violence and neighborhood disadvantage after the crime decline
.
ANNALS of the American Academy of Political and Social Science
,
660
,
341
358
. 10.1177/0002716215579825.
Fuchs, V. R. (
2016
).
Black gains in life expectancy
.
JAMA
,
316
,
1869
1870
. 10.1001/jama.2016.14398.
Gardner, J. W., & Sanborn, J. S. (
1990
).
Years of potential life lost (YPLL)—What does it measure?
Epidemiology
,
1
,
322
329
. 10.1097/00001648-199007000-00012.
Geronimus, A. T., Bound, J., & Colen, C. G. (
2011
).
Excess black mortality in the United States and in selected black and white high-poverty areas, 1980–2000
.
American Journal of Public Health
,
101
,
720
729
. 10.2105/AJPH.2010.195537.
Harper, S., Lynch, J., Burris, S., & Davey Smith, G. (
2007
).
Trends in the black-white life expectancy gap in the United States, 1983–2003
.
JAMA
,
297
,
1224
1232
. 10.1001/jama.297.11.1224.
Harper, S., Rushani, D., & Kaufman, J. S. (
2012
).
Trends in the black-white life expectancy gap, 2003–2008
.
JAMA
,
307
,
2257
2259
. 10.1001/jama.2012.5059.
Heller, S. B. (
2014
).
Summer jobs reduce violence among disadvantaged youth
.
Science
,
346
,
1219
1223
. 10.1126/science.1257809.
Keith, V. M., & Smith, D. P. (
1988
).
The current differential in black and white life expectancy
.
Demography
,
25
,
625
632
. 10.2307/2061326.
Kochanek, K. D., Arias, E., & Anderson, R. N. (
2013
).
How did cause of death contribute to racial differences in life expectancy in the United States in 2010?
(NCHS Data Brief No. 125).
Atlanta, GA
:
National Center for Health Statistics
. Retrieved from https://www.cdc.gov/nchs/data/databriefs/db125.pdf
Lai, D., & Hardy, R. J. (
1999
).
Potential gains in life expectancy or years of potential life lost: Impact of competing risks of death
.
International Journal of Epidemiology
,
28
,
894
898
. 10.1093/ije/28.5.894.
Levine, R. S., Foster, J. E., Fullilove, R. E., Fullilove, M. T., Briggs, N. C., Hull, P. C., . . . Hennekens, C. H. (
2001
).
Black-white inequalities in mortality and life expectancy, 1933–1999: Implications for healthy people 2010
.
Public Health Reports
,
116
,
474
483
.
Levitt, S. D. (
2004
).
Understanding why crime fell in the 1990s: Four factors that explain the decline and six that do not
.
Journal of Economic Perspectives
,
18
(
1
),
163
190
. 10.1257/089533004773563485.
McEwen, B. S. (
2000
).
The neurobiology of stress: From serendipity to clinical relevance
.
Brain Research
,
886
,
172
189
. 10.1016/S0006-8993(00)02950-4.
Meara, E. R., Richards, S., & Cutler, D. M. (
2008
).
The gap gets bigger: Changes in mortality and life expectancy, 1981–2000
.
Health Affairs
,
27
,
350
360
. 10.1377/hlthaff.27.2.350.
Mooney, C. Z. (
1997
).
Monte Carlo simulation
(Sage University Paper Series on Quantitative Applications in the Social Sciences, Series No. 07-116).
Thousand Oaks, CA
:
Sage
.
National Institutes of Health
. (
2017
).
Estimates of funding for various research, condition, and disease categories (RCDC)
.
Bethesda, MD
:
National Institutes of Health
. Retrieved from https://report.nih.gov/categorical_spending.aspx
National Research Council
. (
2014
).
The growth of incarceration in the United States: Exploring causes and consequences
.
Committee on Causes and Consequences of High Rates of Incarceration
, J. Travis, B. Western, & S. Redburn (Eds.). Committee on Law and Justice, Division of Behavioral and Social Sciences and Education.
Washington, DC
:
National Academies Press
.
Olshansky, S. J., Antonucci, T., Berkman, L., Binstock, R. H., Boersch-Supan, A., Cacioppo, J. T., . . . Rowe, J. (
2014
).
Differences in life expectancy due to race and educational differences are widening, and many may not catch up
.
Health Affairs
,
31
,
1803
1813
.
Olshansky, S. J., Passaro, D. J., Hershow, R. C., Layden, J., Carnes, B. A., Brody, J., . . . Ludwig, D. S. (
2005
).
A potential decline in life expectancy in the United States in the 21st century
.
New England Journal of Medicine
,
352
,
1138
1145
.
Phillips, D. P. (n.d.).
The life table
.
San Diego, CA
:
University of California, San Diego
. Retrieved from http://pages.ucsd.edu/~dphillips/lifetable_detailed_construction.txt
Pridemore, W. A. (
2003
).
Recognizing homicide as a public health threat: Toward an integration of sociological and public health perspectives in the study of violence
.
Homicide Studies
,
7
,
182
205
. 10.1177/1088767903251105.
Redelings, M., Lieb, L., & Sorvillo, F. (
2010
).
Years off your life? The effects of homicide on life expectancy by neighborhood and race/ethnicity in Los Angeles County
.
Journal of Urban Health
,
87
,
670
676
. 10.1007/s11524-010-9470-4.
Reyes, J. W. (
2007
).
Environmental policy as social policy? The impact of childhood lead exposure on crime
.
B. E. Journal of Economic Analysis & Policy
,
7
(
1
). https://doi.org/10.2202/1935-1682.1796
Roeder, O., Eisen, L. B., & Bowling, J. (
2015
).
What caused the crime decline?
New York, NY
:
Brennan Center for Justice, New York University School of Law
. Retrieved from https://www.brennancenter.org/publication/what-caused-crime-decline
Sharkey, P. (
2010
).
The acute effect of local homicides on children’s cognitive performance
.
Proceedings of the National Academy of Sciences
,
107
,
11733
11738
. 10.1073/pnas.1000690107.
Sharkey, P., Torrats-Espinosa, G., & Takyar, D. (
2017
).
Community and the crime decline: The causal effect of local nonprofits on violent crime
.
American Sociological Review
,
82
,
1214
1240
. 10.1177/0003122417736289.
Stewart, S. T., & Cutler, D. M. (
2014
).
The contribution of behavior change and public health to improved U.S. population health
(NBER Working Paper No. 20631).
Cambridge, MA
:
National Bureau of Economic Research
. Retrieved from http://www.nber.org/papers/w20631.pdf
U.S. Census Bureau
. (
2017
).
American fact finder
[Data set]. Retrieved from https://factfinder.census.gov/faces/nav/jsf/pages/index.xhtml
Wildeman, C., & Wang, E. A. (
2017
).
Mass incarceration, public health, and widening inequality in the USA
.
Lancet
,
389
,
1464
1474
. 10.1016/S0140-6736(17)30259-3.
Wong, M. D., Shapiro, M. F., Boscardin, W. J., & Ettner, S. L. (
2002
).
Contribution of major diseases to disparities in mortality
.
New England Journal of Medicine
,
347
,
585
592
. 10.1056/NEJMsa012979.
Zimring, F. E. (
2006
).
The great American crime decline
.
New York, NY
:
Oxford University Press
.