|Home | About | Journals | Submit | Contact Us | Français|
We assessed whether New York City’s gun-related homicide rates in the 1990s were associated with a range of social determinants of homicide rates.
We used cross-sectional time-series data for74 New York City police precincts from 1990 through 1999, and we estimated Bayesian hierarchical models with a spatial error term. Homicide rates were estimated separately for victims aged 15–24 years (youths), 25–34 years (young adults), and 35 years or older (adults).
Decreased cocaine consumption was associated with declining homicide rates in youths (posterior median [PM] = 0.25; 95% Bayesian confidence interval [BCI] = 0.07, 0.45) and adults (PM = 0.07; 95% BCI = 0.02, 0.12), and declining alcohol consumption was associated with fewer homicides in young adults (PM = 0.14; 95% BCI = 0.02, 0.25). Receipt of public assistance was associated with fewer homicides for young adults (PM=−104.20; 95% BCI=−182.0, −26.14) and adults (PM=−28.76; 95% BCI=−52.65, −5.01). Misdemeanor policing was associated with fewer homicides in adults (PM=−0.01; 95% BCI=−0.02, −0.001).
Substance use prevention policies and expansion of the social safety net may be able to cause major reductions in homicide among age groups that drive city homicide trends.
Most large US cities experienced a decline in homicide during the 1990s. However, nowhere was the decline more publicized than in New York City.1–4 The city’s homicide drop in the 1990s was the largest in its postwar history1: homicides declined from 2245 in 1990 to 633 in 1998—a drop of 72% over 8 years.2
Two leading theoretical perspectives have guided interpretations of the mechanisms behind New York City’s decline in homicide. One of the most prominent theoretical perspectives is the theory of “broken windows” policing,5 which proposes that failure to control minor offenses creates a sense of public disorder and encourages the proliferation of crime. This theory motivated investment in increased policing of misdemeanors during the 1990s. Research studying this approach’s impact on homicide have produced conflicting evidence; studies exist that report no evidence of reduced homicide associated with misdemeanor policing,6,7 modest reductions in homicide,3,4,8 and significant reductions in crime.9,10
The second theoretical framework, the “crack cocaine” thesis, attributed the increase in homicide in the 1980s to the appearance of crack cocaine in the drug markets of large cities.11–13 This phenomenon was accompanied by heavy recruitment of young male dealers, creating an increased need for use of guns. Blumstein proposed that the drop in homicide in the 1990s was caused by shifting drug markets, police response to gun carrying by young men, efforts to decrease general access to guns, an increase in the prison population, and economic expansion.11,12
Previous studies of the determinants of the New York City homicide decline have built on these theoretical perspectives and used pooled cross-sectional time-series designs, as well as natural experiments, to investigate the impact on overall homicide rates of the following phenomena: improvement in policing of low-level, quality-of-life offenses3,4,9,10,14,15; a decline in drug-market activity, with fewer turf wars between drug-dealing crews11,12,16–18; reduced firearm availability as a result of federal and citylevel efforts to limit access to and licensing of handguns and assault weapons16,19,20; the removal of dangerous persons from the streets via implementation of proincarceration policies15,16,21,22; and a decline in alcohol consumption after the instatement of a consumption tax on beer and hard liquor.2
Previous time-series analyses3,4,8 have examined the influence of these social determinants on overall homicide, and the current study expands upon that literature by separately examining how such factors are related to homicide victimization in specific age groups. The risk of homicide varies sharply by age,23–25 and homicide rates in different age groups may result from different sets of risk factors. Indeed, the differential response to social changes by age group may have contributed to the significant shift in the age structure of homicide in the United States in the 20th century.15,26,27 In New York City, the dramatic drop in homicides in the 1990s benefited some age groups more than others2,28: adolescents aged between 16 and 19 years and young adults aged 20 to 29 years experienced the sharpest decreases in homicide of all age groups.
Focusing on age-specific rates rather than on overall population rates may allow us to understand the drivers of the homicide rates for the age groups that are causing the overall rates to rise and fall. Therefore, in this analysis, we replicated the pooled, cross-sectional time-series design used in previous studies of the overall homicide rate to compare how social determinants of changes in total homicide levels in New York City (i.e., misdemeanor policing, cocaine consumption, firearm availability, incarceration rates, and alcohol consumption) influenced age-specific gun-related homicide rates in the 1990s.
We collected data from 5 sources: the Office of the Chief Medical Examiner (OCME) of New York City, the New York City Police Department (NYPD), the New York City Human Resources Administration, the New York City Mayor’s Management Office, and the US Census Bureau. The units of analysis used for this research were New York City police precincts. Precincts are the most appropriate unit of analysis to study the impact of broken-windows policing, because law enforcement is organized at the precinct level.19 Precincts 33 and 34 were treated as 1 precinct because they were a single precinct until they were split in 1994. The Central Park precinct (precinct 22) was excluded because no one resides in this precinct.
We measured gun-related homicide rates for the following age groups of victims: 15 through 24 years, 25 through 34 years, and 35 years and older. These groups were chosen to capture different developmental stages (youth, young adulthood, and adulthood, respectively) and groups at varying risks of homicide; the risk of homicide decreases from its peak at the age group 15–24 years. Models were estimated separately for gun-related homicides for each age group. We focused on gun-related homicides because (1) previous research has demonstrated distinct trends for gun versus nongun homicide in New York City, and (2) the overall trend for gun-related homicide is more compatible with theoretical arguments about the impact of changes in policing and cocaine markets.1
All cases of homicides in New York City from 1990 to 1999 were identified through standard manual review and abstraction of the medical files of the OCME. The OCME investigates all deaths of people believed to have died from unnatural causes. Thus, all homicide deaths in New York City are reviewed by the OCME and were included in the charts used for data extraction. The same person has held the office of New York City Chief Medical Examiner since 1990, so classification of cases, toxicology, policies, and other aspects of the OCME’s work have remained the same over the study period.
Data regarding cause of death, circumstances of death (including use of a gun), and toxicology were collected from the OCME files by trained abstractors using standard protocols and data-collection forms. The OCME investigators used the decedent’s medical history, the circumstances and environment of the death, autopsy findings, and laboratory data to attribute the cause of death for each case reviewed. We used ArcGIS version 9.0 (ESRI, Redlands, CA) to geocode all OCME cases from 1990 through 1999 to a precinct by address of injury. Only cases with a valid address of injury were included in the analysis. We used data from the 1990 US Census (summary tape files 1 and 3)29 and the 2000 US Census (summary files 1 and 3)30 to calculate homicide rates per 100 000 population. The total annual precinct population was estimated through linear interpolation for the years between census population estimates of 1990 and 2000.
Data were collected from the NYPD on all misdemeanor/ordinance arrests and citizen complaints by precinct from 1990 through 1999 to represent “broken windows”—oriented policing net of misdemeanor behavior. The misdemeanor–ordinance arrest rate per 10 000 population was used as the predictor of interest, and the rate of complaints per 10000 population was used as a covariate to adjust for a potential endogenous response to underlying levels of crime.
The proxy measure for cocaine use was the percentage of deaths in each precinct per year from 1990 through 1999 that were classified as accidental by the OCME and that had positive toxicology results for cocaine, recorded from OCME data. We used the annual percentage of gun suicide deaths from 1990 through 1999, recorded from OCME data, as a proxy for precinct firearm availability. This measure correlates highly with survey-based measures of firearm availability.31 Incarceration rate per 100000 population was operationalized as the number of prison admissions by precinct of arrest from 1990 through 1999. This measure was originally obtained from the New York State Division of Criminal Justice Services. Alcohol consumption per precinct was measured as the annual percentage of accidental deaths with toxicology results positive for alcohol from 1990 through 1999, recorded from OCME data.
The selection of control variables was informed by previous research by Messner et al.3 and others.32 Time-varying covariates included: the ratio of felony arrests to number of felony complaints from 1990 through 1999, to assess whether increased police activity at all levels rather than increased broken-windows policing specifically was responsible for the decline in homicides2; police manpower, measured as the number of police officers assigned to each police precinct from 1990 through 1999 by the NYPD3; and receipt of public assistance, obtained from the Human Resources Administration at the community–district level and disaggregated to the precinct, as a measure of time-varying neighborhood disadvantage.33–36
Covariates only available for decennial years and measured as fixed at 1990 were percentage male, percentage under age 35 years, percentage Black, percentage Hispanic, percentage foreign-born, percentage unemployed, and concentrated poverty. These data were obtained from Census Summary File 3.29 Infoshare Online (http://www.infoshare.org) provided census data at the tract level, which were aggregated to the precinct level. We conducted principal component analysis to construct the composite measure of baseline concentrated poverty by summing the percentage of persons living below 200% of the poverty line (to account for the high cost of living in New York City), the percentage with less than a high school education, and the percentage of female-headed households, each weighted by its factor loading. Higher scores indicated greater levels of concentrated poverty. All time-invariant control variables were standardized to a mean of zero and a standard deviation of 1 to improve convergence and to enhance comparability and ease of interpretation.
where ΔYit was the change in the age-specific homicide rate between times t and t+1 for precinct i for time period t, ΔXarrestsit was the change in the misdemeanor arrest rate between times t and t+1, ΔXcoke.it was the change in proportion of accident decedents positive for cocaine toxicology, ΔXfirearmsit was the change in the proportion of gun suicides, ΔXincarcerationit was the change in the incarceration rate, ΔXalcoholit was the change in the proportion of accident decedents positive for alcohol toxicology, ΔLit was a vector of time-varying covariates, Vi was a set of baseline covariates, and λi was the random spatially structured effect.39 We used spatial error models to account for the spatial dependence of risk for homicide in nearby areas. The spatial random effect was modeled with a prior that has a conditionally autoregressive (CAR) distribution, with weights for first-order adjacent neighbors set at 1 (“neighbors” defined as precincts sharing a border).40 All models were estimated with Winbugs version 1.2 (MRC Biostatics Unit, Cambridge, England), with 2 parallel Markov chain Monte Carlo chains. Details about these spatial models are given in Appendix 1 (available as a supplement to the online version of this article at http://www.ajph.org).
The predictors of neighborhood homicide rates were examined separately for each homicide victim age group (15–24 years, 25–34 years, 35 years or older). For each age group, a model including the exposures of theoretical interest was first estimated. Next, the time-varying control variables were introduced into the model. Finally, spatial lags for the distribution of significant age-specific predictors in adjacent neighborhoods were included in model 4.
Of 14186 homicides that occurred in New York City between 1990 and 1999, 2027 (14.3%) were missing precinct-of-injury information, leaving 12159 homicides classified by precinct of injury. Of these, 8820 (72.5%) were firearm-related and were used in the analysis. Compared to those included in the analysis, firearm homicide decedents who were excluded from the analysis because of missing precinct of injury were more likely to be female (8.10% of excluded vs 7.54% of included) and White (9.27% of excluded vs 7.66% of included) or other race (4.93% of excluded vs 3.66% of included), but were less likely to be Black (48.51% of excluded vs 52.14% of included). There were no differences in the ages of decedents included and excluded from the analysis. Homicide counts geocoded by precinct on the basis of data obtained from the OCME correlated from 0.85 to 0.95 (depending on the year; by Sperman p correlation) with the NYPD homicide counts.
Figure 1 in Appendix 2 (available as a supplement to the online version of this article at http://www.ajph.org) illustrates changes in age-specific homicide rates over the decade. Gun-related homicide rates among youths aged 15–24 years in 1999 77% from the rate in 1990, and young adults aged 25–34 years experienced a decline of 74% from 1990 to 1999. Homicide among adults aged 35 years or older dropped 76% over the same time period.
Table 1 presents the demographic characteristics for the total sample averaged across the decade, as well as the beginning and the end of the decade. Although baseline measures were standardized for model-estimation purposes, we report means and standard deviations of the original distributions for descriptive purposes. Concomitant with the homicide drop, a series of social characteristics shifted. For example, proxy indicators of cocaine use and alcohol use (the proportion of deaths in the precinct with a positive toxicology report for cocaine and alcohol use) decreased from 8.6% to 5.2% and from 19.5% to 16.2%, respectively. The proportion of residents receiving public assistance diminished over the decade, from 12.3% to 9.3%. The rate of misdemeanor–ordinance arrests, by contrast, increased from 286.23 to 389.15 per 10000 people.
Table 2, Table 3, and Table 4 present the models used to test the association between changing precinct-level characteristics and changing age-specific rates of homicide. Table 2 presents models for youths aged 15–24 years. Model 4 reveals significant positive associations for change in cocaine consumption and change in incarceration rates (Table 2). Expressed with reference to standard deviations to illustrate magnitude, these results indicate that a standard deviation (or 10.17%) increase in the proportion of a neighborhood’s accidental deaths that were positive for cocaine was associated with 2.6 more homicides per 100 000 people (posterior median [PM] = 0.26; 95% Bayesian confidence interval [BCI]=0.07, 0.46), and a standard deviation increase in cocaine consumption in surrounding neighborhoods was associated with 3.7 more homicides (PM=0.36; 95% BCI=0.01, 0.71). A standard deviation increase in the incarceration rate (413.73 incarcerations per 100 000) was associated with a homicide increase of 16.55 (PM=0.04; 95% BCI=0.01, 0.07).
Table 3 shows comparable models for young adult (ages 25–34 years) victims of homicide. Model 4 (Table 3) indicates that a standard deviation (or 13.13%) increase in accidental deaths positive for alcohol was associated with 1.8 more homicides per 100 000 population (PM=0.14; 95% BCI=0.02, 0.25). Change in the proportion of those receiving public assistance had a negative association with change in homicide (PM=−104.20; 95% BCI=−182, −26.14). A standard deviation (or 10.11%) increase in public assistance resulted in a homicide decrease of 10.53.
Table 4 shows models for homicide among adults aged 35 years or older. As with youths, change in cocaine consumption was positively associated with change in homicide. A standard deviation increase in cocaine consumption was related to a homicide increase of 0.71 (model 4; PM=0.07; 95% BCI=0.02, 0.12). Change in public assistance also was significantly associated with change in homicide, similar to the results for young adults. A standard deviation increase in the rate of public assistance was associated with 2.9 more homicides per 100000 (PM=−28.76; 95% BCI=−52.65, −5.01).
In contrast to the findings for the other age groups, change in misdemeanor and felony arrests was significantly associated with change in homicide for the adult age group. A standard deviation increase in misdemeanor arrests (737 arrests per 10 000) was associated with 7.37 fewer homicides per 100 000 people (PM=−0.01; 95% BCI=−0.02, −0.001). A standard deviation increase in felony arrests (0.22 arrests per complaints) was associated with 2.62 fewer homicides (PM=−11.92; 95% BCI=−21.54, −2.34).
This is one of the first studies to use the full range of theoretically relevant predictors to study age-specific gun-related homicide victimization in New York City in the 1990s. We found that during a time of dramatic citywide decrease in homicide, social changes had distinctive relationships with the decline of age-specific homicide rates. A decline in cocaine-market activity was comparably associated with a falling homicide rate for people aged 15–24 years and for those aged 35 years or older. Decreased levels of alcohol consumption were associated with a reduced rate of homicide for those aged 25–34 years. Among both young adults and adults, provision of public assistance reduced levels of homicide, although less so for adults than for young adults. Finally, increased misdemeanor policing played the strongest protective role among adults.
The impact on youth of declining levels of cocaine consumption in focal and surrounding neighborhoods is consistent with the crack cocaine thesis, which implies that youth are particularly susceptible to the influence of changes in the drug markets. The decline in the crack cocaine trade and the transformation of the trafficking market away from open-air, competitive, free-for-all arenas into a structured, underground economy removed the need for highly armed and potentially violent youth as sellers.2 The impact of cocaine consumption on adults, in contrast, may reflect a decline in the number of adults who developed crack or cocaine dependence and an associated decline in the risk of engaging in violent acts (and of becoming a victim of a violent act), as well as a decrease in the level of violence associated with the drug trade, which would likely be run by serious criminals who were also cocaine-dependent. Evaluations of the drug-homicide link have found significant associations in cities across the United States12,41–44 and in other countries, including Australia45 and Brazil.46 Ousey and Lee, for example, found in a study of 132 cities that part of the homicide decline from 1984 to 2000 was caused by a drop in drug-market activity and the aging of drug-market participants.17
We found a positive association between alcohol consumption and homicide rates for young adults aged 25–34 years. Individuals are more vulnerable to victimization when they are under the influence of alcohol intoxication. This may be particularly the case for persons who abuse alcohol and are alcohol dependent, problems that increase in young adulthood.47 Previous studies have found that alcohol, which has been attributed to increased impulsivity and cognitive impairment,48 was responsible for an important proportion of gun-related homicides in New York City.49–51 The establishment of a special tax on hard liquor and beer in the mid-1980s2 may have contributed to lower alcohol consumption that persisted in the 1990s, lowering the risk of victimization among chronic alcohol abusers.
Increases in the relative size of the population on welfare were associated with declines in homicide among young adults and adults. This unexpected finding may reflect the salutary effect of a welfare safety net, conditional upon baseline levels of disadvantage. Previous research has suggested that more generous and expansive social-welfare policies reduce stressors in the environment and strengthen institutional controls, thereby reducing levels of lethal violence.52–54 The distinctive association between an increase in welfare receipt among neighborhood residents and a decrease in homicide among older age groups is consistent with the notion of a process of “aging out of crime,”55,56 as people in their 20s are more likely to be on the road to desistance from crime and in the process of forming families, so that income supplements could help get them out of situations that make them more vulnerable to victimization.
Increased incarceration was associated with an increase in homicide among youth aged 15–24 years. This result contradicts previous work by McCall et al.15 and Levitt,21 who found a significant relationship between increases in the use of prisons and decreases in crime in cities across the United States. However, our finding is consistent with Karmen’s finding that after 1993, the fraction of felony arrestees who received harsh prison sentences declined, thus making incarceration an ineffective strategy against homicide.2 At the same time, police may target high-crime areas and imprison a large number of low-level criminals in such precincts, thus contributing to co-occurring—but not causally connected—high incarceration levels in high-homicide areas.
Misdemeanor policing was protective against homicide victimization for adults aged 35 years or older. Previous research3,4,17,57,58 has found an association between misdemeanor policing and overall homicide, but to our knowledge, this is the first study to examine the impact that misdemeanor policing has on age-specific rates of victimization. Perhaps misdemeanor policing has a particular impact on adults by keeping chronic drug abusers off the streets, making them less vulnerable to victimization. Further research needs to delve into potential explanations for this pattern.
The study was subject to several limitations. The use of a nonexperimental design inevitably raises concerns about endogeneity and unobserved confounding. We attempted to address these issues by including time-varying and baseline measures of economic conditions, sociodemographic characteristics, and policing practices, as well as spatial lags for the distribution of significant predictors in adjacent neighborhoods. However, we cannot conclusively differentiate the impact of the main predictors of interest from correlated phenomena that may have been occurring at the same time in New York City neighborhoods. Further, our measure of firearm availability does not necessarily capture weapon carrying, which may have been affected by increased aggressive policing. Future studies should examine the impact of policing on gun carrying and consequently on homicide.
This study clarifies the specific ways in which changes in social conditions affected homicide victims of different age groups in New York City. Although the decline in crack and cocaine markets played a role in thehomicidedeclineamongthegroupwith the highest rate of homicide (youths aged 15–24 years), alcohol and cocaine consumption, misdemeanor policing, and public assistance contributed to changes in homicide among the older age groups. The heterogeneity in homicide rates by age group and the secular shifts in the age structure of homicide underscore the need to examine age-specific patterns to understand more fully the processes that drive aggregate trends in homicide.
This study was funded by the National Institute on Drug Abuse (grants DA 06354 and DA 017642) and the Robert Wood Johnson Foundation.
The data on complaints for misdemeanor arrests, ordinance violations, and felony complaints were kindly provided to us by Richard Rosenfeld and Robert Fornango.
ContributorsM. Cerdá conceptualized the study and led the writing of the article. S. F. Messner and S. Galea assisted with framing the arguments in the article, writing the article, and editing the article. M. Tracy and E. Goldmann assisted with the analyses. D. Vlahov and K.J. Tardiff assisted with design of the study, management of the study, and initial framing of the article. All authors assisted with conceptualizing ideas, interpreting findings, and reviewing drafts of the article.
Human Participant Protection
The research protocol was reviewed and approved by the New York Academy of Medicine institutional review board and the Cornell University Weill School of Medicine institutional review board.