Are the geographic disparities in U.S. violent crime rising?

Inequality in economic and social outcomes across U.S. regions has grown in recent decades. The economic theory of crime predicts that this increased variability would raise geographic disparities in violent crime. Instead, I find that geographic disparities in homicide rates decreased. Moreover, these same decades saw decreases in the geographic disparities in policing, incarceration, and the share of the population that is African American. Thus, changes in policing, incarcerations, and racial composition could have led to a decrease in inequality in homicide rates. Moreover, the joint provision of law enforcement by local, state, and federal authorities may have reduced the impact of economic distress on violent crime.


Homicides and cleared homicides from law enforcement agencies
I examine homicides reported by all law enforcement agencies in the United States (e.g., city police departments, sheriffs' offices, and university police departments), which are referred to as "murder and non-negligent manslaughter," and correspond to homicides from NCHS (excluding deaths caused by the intervention of the police or executions [1]).Their  The analysis excludes agencies that cannot be assigned to a commuting zone.These agencies serve multiple counties (state police, Indian reservations police).However, only 0.41% of all murders are reported by agencies that cannot be assigned to a commuting zone.
The 1962 data for Utah, Vermont, Virginia, Washington, West Virginia, Wisconsin, Wyoming, Alaska, and Hawaii was inadvertently erased [2].For this reason, I exclude homicides by all law enforcement agencies for this year.Cleared homicides are only available for the years 1964-2020.

Police officers
The number of police employees is from Jacob Kaplan's Concatenated Files: Uniform Crime Reporting Program Data: Law Enforcement Officers Killed and Assaulted (LEOKA) 1960-2020, Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2021-09-22, https://doi.org/10.3886/E102180V11.I corrected a few of the 1960 and 1961 values using the printed versions of Crime In The United States for 1960 and 1961, which can be found at archive.org/details/sim_ crime-in-the-united-states_1960/mode/2up?view=theater and archive.org/details/sim_crime-in-the-united-states_1961/page/124/mode/2up?view=theater.For 2018-2020, an unusually high number of law enforcement agencies reported zero police officers.For this reason, I exclude these years from the analysis.I link the law enforcement agency codes to counties the same way as with crime statistics.Only 0.54% of all law enforcement officers are reported by agencies that cannot be assigned to a commuting zone.

Prison statistics
The number of state prisoners from a particular county from 1986 to 2016 is obtained from the Vera Institute, Incarceration Trends Dataset, https://github.com/vera-institute/incarceration-trends,2024.I only examine counties with data for at least 28 out of the 34 years.enforcement divided by the number in NCHS (since I exclude from the NCHS data on homicides committed by police and executions [1]).Inequality in reporting, however, decreased from the 1960s until the 1980s but has been back to higher levels since the 1990s (see panel (b) in Fig 5).

Commuting zone demographics
Thus, between the 1970s and the 2000s, homicide rates in some commuting zones would have artificially decreased because a smaller share of homicides were reported by law enforcement in those communing zones (compared to other commuting zones).Indeed, inequality in homicide rates computed with law enforcement statistics increased from the 1970s to the 1990s (see panel (c) in In conclusion, law enforcement statistics provide ambiguous findings about the evolution of inequality of homicides.Thus, examining other non-lethal violent crimes reported in the law enforcement statistics may not be helpful given that they are widely believed to be subject to greater reporting error [7,8,[10][11][12]

Appendix A3: Interpretation of changes in Gini coefficients
This section explains why a 0.05 change in the Gini coefficient for homicides is substantial.
Let f (h) be the density function for homicides, and let µ be its mean.Blackburn constructs a density function for homicides with more inequality, f ⋆ (h), by increasing the homicide rate in high-homicide-rate areas by k and decreasing it by k in low-homicide-rate areas.Here, "high-homicide rate areas" are commuting zones with homicide rates that are above the median, while "low-homicide rate areas" are commuting zones with homicide rates that are below the median.Then, Blackburn shows that where G ⋆ and G are the Gini coefficients corresponding to the densities f ⋆ and f .In the 2010s, the mean and median homicide rates were 4.38 and 3.60 per one hundred thousand, while the Gini coefficient was 0.39.I compute the more unequal density by increasing homicide rates by 0.477 in all commuting zones with homicide rates above the median and decreasing them by 0.477 in all commuting zones with below the median homicide rates.Then, equation (1) implies that the more unequal homicide density function has a Gini coefficient of In order to interpret the results in percentages, I note that the average 2010s homicide rates for high and low homicide-rate areas were 7.63 and 2.4.Thus, increasing the homicide rate by 0.477 in high-homicide-rate commuting zones corresponds to a 6.25% increase, while decreasing it by 0.477 in low-homicide-rate commuting zones corresponds to a 19.88% decrease.
Clearly, a 19.88% decrease in homicides is substantial, and thus the 0.05 = 0.477/(2 • 4.38) increase in the Gini coefficient is also significant.

Appendix A4: Lorenz curves
Let p ∈ [0, 1] be a population percentage, X i the value of variable 'X' for commuting zone i (for instance, the 1960 homicide rate for commuting zone i), Q X (p) the highest value of X for commuting zones in the pth percentile, and L X (p) = i : Xi≤Q X (p) X i / j X j the value of the Lorenz curve.Then, X is more unequal than Y is for all p, L X (p) ≤ L Y (p).
Fig A plots Lorenz curves for selected years and variables, which were computed using [13].The graphs show that over the period analyzed, inequality in homicide rates, policing, imprisonment rates, and share of the population that is African American decreased while inequality in homicide clearance rates increased.
Commuting zones are defined in Tolbert, Charles M., and Molly Sizer, 1996, "U.S. Commuting Zones and Labor Market Areas: A 1990 Update," Economic Research Service Staff Paper 9614.The Aleutian Islands were not split into East and West boroughs in the National Center for Health Statistics data until 1994.So, for earlier years, I combine these two areas and thus only consider 740 commuting zones.Also, for the 1960s, I combine Dillingham with Bristol Bay and thus have 739 commuting zones for those years.I compute the commuting zone population, African American population, population of ages 15 to 29, and income per capita by aggregating the counties corresponding to each commuting zone.County population by age and race for 1969-2020 are obtained from the Survey of Epidemiology and End Results, U.S. State and County Population Data.County population by age and race for 1960 are from the 1960 Decennial Census.The data from the Survey of Epidemiology and End Results and the Bureau of Economic Analysis combines some counties in Alaska, Colorado, Hawaii, and Virginia.I obtained demographic data for those counties from the 1970, 1980, 1990, 2000, 2010, and 2020 Decennial Census and the 2006-2010 and 2016-2020 American Community Survey (5-year estimates).Then, I interpolated values for the years in between.

Fig 5 )
, the opposite of what I found with the NCHS data (panel (a) in Fig 1).

Figure A :
Figure A: Lorenz curves for selected years and variables (a) Homicide rates (NCHS)

Figure A: Continued
Figure A: Continued

(d )
Imprisonment rate (e) Share of the population 15 to 29 (f ) Share of the population that is African American Data sources are discussed in S1 Appendix A1.
number is obtained from the United States Department of Justice, Federal Bureau of Investigation, Uniform Crime Reporting Program Data: Offenses Known and Clearances by Arrest, years 1960-2020, Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].The same data sets also provide the number of cleared homicides, where homicides are cleared if at least one person was arrested, charged for the crime, and remanded to court, or alternatively, if some force outside the agency prevented arrest, for instance, the death of the suspect.I link law enforcement agency codes to counties using the National Archive of Criminal Justice Data, Law Enforcement Agency Identifiers Crosswalk [United States],