Skip to main content
SearchLoginLogin or Signup

Variability in Violent Crime Rate Trends Among a Sample of U.S. Cities

Published onDec 15, 2023
Variability in Violent Crime Rate Trends Among a Sample of U.S. Cities
·

Abstract

In 2020, the United States experienced a historic increase in some severe violent crimes like murder. This national spike in violence is often applied universally to U.S. cities and is typically associated with the COVID-19 pandemic and simultaneous social upheavals.  Importantly, average national violent crime rate trends may mask substantial variation in trends across U.S. cities, which could inform crime prevention policy. In this study, violent crime data from 590 municipal law enforcement agencies that reported complete crime data to the National Incident-based Reporting System for 2018–22 were analyzed using group-based trajectory modeling. Results show that (1) much of the national change in violent crime appears to be attributable to a minority of cities that reported relatively high crime rates in prior years and (2) analytic methods used to predict variation around a single average trend appear to be appropriate for explaining changes in violence over this period.

Keywords: violent crime, crime rates, trend analysis, group-based trajectory modeling

Introduction

In 2020, the United States experienced its greatest single year increase in the murder rate in over a century (Howard, 2021). Although other types of crime, such as robbery, decreased during this time, the historic increase in murders shocked the nation. The impact of COVID-19 and the resulting policies enacted to prevent its spread, such as stay-at-home orders and a diminishment of the criminal justice system are obvious culprits to explain this increase, but 2020 also included the murder of George Floyd and resulting social unrest, a large spike in gun purchases, and domestic migration. Previous spikes or drops in the national crime rate have been researched extensively to understand why certain trends occurred (e.g., large decline in crime rates in the 1990s) and have proven difficult to explain. To advance research aimed at explaining variation in violent crime trends over COVID-19 and the other events of 2020, including social unrest following the murder of George Floyd, this study describes variation in trends in violent crime rates among a large sample of U.S. cities from 2018–22.

Figure 1 shows recent national trends in violent crime for all violent crime, homicide, aggravated assault, and robbery, which are estimated based on police-reported crime data submitted to the Federal Bureau of Investigation’s (FBI) Uniform Crime Reporting Program (UCR) and available from the FBI’s Crime Data Explorer (CDE) website.1 Crime rate estimates for 2010–20 and 2022 are based on data from the UCR’s Summary Reporting System (SRS) and National Incident-based Reporting System (NIBRS) while the 2021 crime rate estimates are based on data from the UCR’s NIBRS (Berzofsky, Liao, Couzens, Smith, & Barnett-Ryan, 2022).2 From this figure, the historic spike in the murder and aggravated assault rate in 2020 is evident, along with a continuation of the downward trend in the robbery rate. Evidence suggests that the increase in the murder rate was nearly ubiquitous in the U.S., as the size of the increase was similar in both urban and rural cities across the country (Frosch, Maher, & Elinson, 2022). Importantly, these national average trends hide the fact that there are places in the country that did not experience a large spike in violent crime rates, and some may have even experienced decreases in violence over this time. Understanding whether the policies and practices in some U.S. cities prevented the spike in violence that much of the country experienced is of vital importance to inform future violence prevention efforts. The current study tests for the presence of different trajectories in recent violent crime trends among a large sample of U.S. cities. Our hypothesis is that the national trends shown in Figure 1 conceal subgroups of cities that experienced distinct trends in violence between 2018 and 2022. The findings from this study will benefit future research aimed at explaining variation in violent crime trends over this period.

It is possible that the national trajectories for different violent crime outcomes represent a collection of groups (or a “mixture”) of locations that experienced distinct trajectories not accurately represented by the national average. An example of this can be found in Figure 2 from a study conducted by Scott, Wellford, Lum, and Vovak (2019), where the authors found that the average decline in the homicide clearance rate between 1981 and 2013 among the 100 largest law enforcement agencies was formed by 60% of agencies decreasing over time and 40% of agencies increasing over time. By simply examining the average trend or by conducting a statistical analysis like random-effects or growth curve modeling that predict variation around the average trend, one would miss the important finding that many police departments were improving in their clearance rates at the time of a national decline. It would also limit one’s ability to understand which factors distinguish agencies that improved over time from those that became worse off. Simply put, aggregated average trends can hide important sub-group variation.

The same phenomena of subsets experiencing trajectories distinct from the overall trend could be said for recent violent crime rate trends. Clearly, many cities experienced large increases in violent crimes such as murder and aggravated assault between 2018 and 2022. However, it could be that groups of cities avoided this surge in violence, and it is important to understand which factors distinguish these groups from cities that experienced the surge to inform future crime prevention policies. Likewise, among cities that experienced a spike in violence in 2020, some cities may have been able to greatly reduce their crime levels in 2021 and 2022 whereas other cities might have remained at elevated levels compared with pre-2020 levels. The current study uses group-based trajectory modeling (GBTM) to identify whether national trends for multiple types of violent crime for 2018–22 comprise unique trends from distinct groups of cities. In doing so, it adds a richness of information to our understanding of how violent crime changed over the COVID-19 pandemic and the other events of 2020 in the U.S. and can inform future studies aimed at explaining between-city variation in violent crime trends over this period.

Methods

The current study uses NIBRS victimization data for 2018–22 to examine changes in police-reported violent crime before, during, and after the events of 2020. Specifically, we use a statistical method known as GBTM and crime victim data from a sample of municipal police departments that reported these data to NIBRS during each month of this period to test whether groups of cities experienced unique crime trajectories between 2018–22 for the offenses of murder, aggravated assault, robbery, and carjacking. Due to limited variation in crime trajectories within cities of fewer than 25,000 persons, we limit our analyses to a sample of cities with 25,000 residents or more as of the year 2018. To conduct the analyses, we (1) combined data from cities with 25,000 or more residents into a single sample and (2) stratified the sample into population groups, including 250,000 or more residents, between 100,000 and 249,999 residents, between 50,000 and 99,999 residents, and between 25,000 and 49,999 residents. In the Results Section, we describe findings from analyses with the combined sample and discuss unique findings from the stratified sample results, which are provided in the Appendix. Next, we discuss our data, measures, and statistical analyses.

Data

The NIBRS, operated by the FBI and implemented in 1991 as a crime reporting program within the FBI’s UCR Program, is a crime reporting standard that compiles incident-based data on all crimes voluntarily reported by participating state, local, and federal law enforcement agencies (Addington, 2019; Akiyama & Nolan, 1999; Dunn & Zelenock, 1999; FBI, 2020). NIBRS is incident-based in that a criminal incident, rather than a single criminal offense, is the unit of measurement with each crime incident comprising data about all offenses, property, victims, offenders, and arrestees in that incident. The UCR Program became a NIBRS-only collection on January 1, 2021 (FBI, 2018); therefore, NIBRS is now the single national standard for law enforcement-based crime data (Addington, 2019). According to the FBI’s CDE website, all 50 state UCR programs are NIBRS certified as of 2022, and over 80 percent of the more than 18,000 law enforcement agencies in the United States are now contributing detailed data inputs as required by NIBRS.

Using NIBRS data to study changes in crime over time poses some challenges to researchers because of missing data, particularly during the study period as many agencies were in the process of transitioning to NIBRS during this timeframe. Additionally, some law enforcement agencies do not consistently report from year to year or even from month to month within a given year. Because of inconsistency in reporting patterns, it can be difficult to distinguish between true missing values and zero crimes. It is reasonable to assume that some agencies will have zero homicides or aggravated assaults in any given month or even year in smaller jurisdictions. The extract files themselves do not contain information that distinguishes true missing from zero crimes in any given month. Using data from the FBI for NIBRS agencies, we identified which months an agency did and did not report data during our observation period. We combined this information with data on the date each agency started reporting to NIBRS, which is available in the NIBRS extract files, to create a sample of law enforcement agencies that started reporting to NIBRS on or before January 1, 2018, and that submitted data to the UCR each month from 2018–22. We call this sample of agencies “full reporters” given their lack of missing data over the observation period. Although restricting our sample of municipal police departments to full reporters limits the number of agencies in our sample, we felt it was more important to provide accurate data for a smaller sample than to impute missing months of crime data and report potentially inaccurate crime trends for a larger sample.

The only exception made to our full reporting status sample inclusion criterion was to impute 2 months of missing data for Dallas (TX) Police Department, which started reporting to NIBRS on March 1, 2018. We made an exception for this city given its limited amount of missing data (2 months) and its large population size. The largest agencies have historically been under-represented in NIBRS, as such, including Dallas improves the external validity of findings. Because our analytic procedure uses an agency’s entire 20-quarter crime rate trajectory as data, 2 months of data is unlikely to have an impact on model estimates. Therefore, we applied a simple method of imputation by replacing Dallas Police Department’s missing January and February 2018 crime rate values with their March 2018 crime rate values for each of our offense outcomes.

Because we were unsure of how completely our FBI-provided indicators of monthly UCR reporting status could identify agencies with no missing data, we conducted an outlier detection analysis. Specifically, the authors were concerned that agencies may have failed to submit complete crime incident data to NIBRS for some time during our observation period and these partial or missing data may lead to the inaccurate assignment of municipalities into crime rate trajectories. There are several reasons why agencies could fail to report crime incident data to NIBRS for some period despite actively participating in the data collection, including if their record management system was not working properly. To evaluate this concern, we conducted a two-step outlier detection procedure that first used single-linkage hierarchical clustering of agency crime data to split each agency’s data into high and low clusters and then used a median + median absolute deviation (MAD) method to detect outliers by calculating how many MADs apart each value was from the median of each cluster and labeling crime counts with a value of 4.5 MADs or greater as outliers (Bunker, Liao, & Berzofsky, 2023). The authors examined the results for agencies with low outliers to review whether the low values appeared to be naturally low values or whether they might reflect missing or partial data for 1 or more months of crime data. We found no instances of outliers that reflected partial/missing data in a month giving us confidence that every agency in our sample reported full crime data to NIBRS for each month between 2018–22. Therefore, we proceeded to conduct analyses using crime data from our sample of 590 municipal police departments.

It is important to note that while the number of agencies reporting to NIBRS continues to increase, the available data are not representative of the United States. Table 1 compares our sample of cities to remaining places in the U.S. on 5-year 2018 estimates from the U.S. Census. Some of the cities included in our sample were not included in the U.S. Census place data. For example, North Andover, MA, and North Attleboro, MA, were part of Andover and Attleboro in the U.S. Census. As a result, the total number of places included in our sample is smaller in Table 1. One can see that our sample of cities represented by full-reporter municipal police departments in NIBRS are less populous and wealthy compared to other Census-designated places of 25,000 residents or more. They tended to be more white and had higher proportions of residents with at least a bachelor’s degree. Areas with higher rates of disability, and lower rates of unemployment, and Northeastern and Midwestern agencies were overrepresented in the sample. Clearly, our sample is not nationally representative, so results should not be generalized to all U.S. cities.

Measures

Our outcomes of interest include crime rates for the offenses of murder and non-negligent manslaughter (“murder” going forward), aggravated assault, robbery, and carjacking. We define carjacking as any victimization that involved a robbery offense where a motor vehicle (e.g., car; bus) was stolen or recovered.3 After creating these offense type indicators, we summed victimizations for each outcome to the quarter-year and agency levels and generated quarterly victimization rates with the population variables in NIBRS by dividing each agency’s quarter-year crime counts by their population that year and multiplying by 100,000. Table 2 describes the annual crime rates per 100,000 people among the 590 full NIBRS reporting cities that were included for analysis for 2018–22. One can see that 2022 rates were higher for all types of crime except robbery.

Statistical Analyses

The unit of analysis in the current study is the quarter-year and the observation period includes 2018–22. This study uses a semi-parametric, finite mixture modeling analytic method called GBTM to test for the presence of unique crime rate trajectory groups among our sample of municipal police departments representing cities with populations of 25,000 or more residents for each of our offense outcomes over the 20 quarter-year period. Given that our crime rate outcomes are continuous, we model these data using the censored normal distribution. GBTM is valuable if one believes that an average trend of an outcome may conceal a mixture of unique groups of units that followed qualitatively distinct trajectories over time in such a way that the average is not an accurate summary measure of unit trends (Nagin, 2005; Nagin & Tremblay, 2005; Scott et al., 2019). We hypothesize that this is the case for changes in crime over the observation period and therefore use GBTM to test whether the distribution of crime rate trends in the aggregated data reflect a mixture of city trends that represent groups of cities that are similar enough to some city trends and distinct enough from other city trends to represent unique groups of cities that experienced qualitatively distinct changes in violent crime during the 2018–22 period.

In this study, we adhere to the recommendations described by the developer of the method (Nagin, 2005; Nagin & Odgers, 2010) when selecting final models and base our modeling decisions on several measures of model fit, including comparing information criteria between models, examining the posterior probabilities of group membership, examining the plots of each model, comparing observed to predicted group means, and reviewing model output to determine whether the shape of each trajectory (i.e., flat, linear, quadratic, or cubic) is statistically significant. Importantly, sometimes a better fitting model might exist that disaggregates one group of cities into multiple similar groups. For example, one group of cities that maintained a relatively constant crime rate of 10 crimes per 100,000 persons per year over time might better be represented by two groups that maintained a relatively constant crime rate of eight crimes and 12 crimes per 100,000 persons per year over time, respectively. Depending on the outcome, this difference might not be substantively meaningful. In these cases, we prioritized the simpler model to make our results easier to interpret. Likewise, since the purpose of our application of GBTM is to identify unique groups of cities that followed similar trends and not to identify outliers, we viewed groups with less than 5% of city police departments assigned to them with extra scepticism and avoided models that included them unless they greatly improved the fit of the model based on the criteria discussed above. Finally, extreme values created issues with estimating GBTMs for aggravated assault with the combined sample of all cities with 25,000 or more residents and with the sample of cities with a resident population of between 25,000 and 49,999. Therefore, for these analyses we removed seven and four agencies, respectively, that were above the 99th percentile in this outcome in either the first or last quarter of our observation period.

Importantly, because deciding on final models can often be as much art as science, two of the study authors with experience using GBTM independently selected final models and compared results for the combined sample of agencies. They initially disagreed on the final model for two of the four outcomes. In both instances, the difference came down to the presence of a relatively small group that only one analyst included. After discussion and a review of the predicted posterior probabilities between models with and without the small group, the analysts agreed on presenting the model without the smaller group in each case. We selected simpler models based on the purpose of our analyses, which is to identify groups of meaningful variation within the sample of city police departments and not to identify small sets of outliers when more complex models do not greatly improve model fit.4 All models were estimated using the “traj” program in Stata, and our data and code are publicly available here: [REMOVED FOR BLIND REVIEW].

Results

Results below are presented for the combined sample of 590 U.S. cities with a population of 25,000 persons or more whose police departments reported data to NIBRS each month between 2018–22. We present trajectory group plots for each of our four crime rate outcomes along with a description of model diagnostics for each outcome. The Appendix presents additional results for each of the four crime rate outcomes that are specific to each resident population size strata. For each crime type, average probability of group membership for each group exceeded 0.97 and within each identified group, very few agencies were assigned a probability less than 0.80 (i.e., typically fewer than 5% of the agencies assigned to that group). This suggests very good model fit across the offense type models.

Combined Sample Results, by Crime Outcomes

Murder

For the overall sample, the trajectory analysis for murder revealed that the 590 cities could be grouped into two trajectories, as displayed in Figure 3. Four hundred and 46 cities (76%) can be characterized as “Low stable” and 144 cities (24%) as “High Increasers” based on reported murders from 2018–22 (group numbers 1 and 2, respectively). Cities in the Low Stable group averaged a quarterly murder rate of approximately 0.40 per 100,000 people in years 2018 and 2019 and only slightly increased to an average of about 0.50 from 2020–22. Alternatively, the trajectory for the High Increasers group began at a relatively higher quarterly murder rate (a rate of approximately 2.5 per 100,000 people), reached an average rate of 3.8 per 100,000 in 2020 and 2021, and then declined slightly to 3.5 per 100,000 in 2022.

Aggravated Assault

The GBTM analysis for aggravated assault (Figure 4) identified three trajectories for the combined sample. Three hundred and ninety-five cities (68%) are described as “Low Stable,” 150 cities (26%) as “Middle Increasers,” and 38 cities (7%) as “High Increasers” (group numbers 1, 2, and 3, respectively). Most cities (“Low Stable”) began with very low quarterly aggravated assault rates, which remained relatively stable throughout the study period (ranging from an average quarterly rate of 27 per 100,000 in 2018-2019 to about 31 per 100,000 from 2020 to 2022), whereas about one-quarter of cities began at much higher quarterly rates (approximately 91 per 100,000) and experienced a small increase from 2020–22 (reaching approximately 105 per 100,000) (“Middle Increasers”). High Increasers began with the highest quarterly aggravated assault rate in 2018 (approximately 193 per 100,000) and experienced a notable increase from 2020-2022 (approximately 233 per 100,000).

Robbery

Three trajectories were identified for the robbery rate outcome (Figure 5). Three hundred and sixty cities (61%) can be described as “Low Stable,” as they began and ended with a relatively low quarterly robbery rate (starting at an average of 8.3 per 100,000 in 2018 and declining slightly to 6.5 per 100,000 by 2022) (group 1). One hundred and ninety-one cities (32%) can be described as “Middle Decreasers Stable” who began at relatively higher quarterly robbery rates than the Low Stable group (approximately 32.1 per 100,000 in 2018) and experienced a small but steady decline to 25.5 per 100,000 by 2022 (group 2). Finally, 39 cities (7%) began at a relatively high quarterly robbery rate (an average of 86.9 per 100,000 in 2018) and experienced a steady decline across the study period to an average quarterly rate of 68.4 per 100,000 in 2022. This group of cities could be described as “High Decreasers” (group 3).

Carjacking

Two trajectories were identified for carjacking rates (Figure 6). Four hundred and fifty-five cities (77%) can be characterized as “Low Increasers,” as they began at low quarterly carjacking rates of approximately 0.37 per 100,000 and that increased slightly to about 0.57 per 100,000 in 2021-2022 (group 1). One hundred and thirty-five cities (23%) can be described as “High Increasers” (group 2). High Increasers began at a relatively higher quarterly rate than Low Increasers (approximately 3.3 per 100,000 in 2018) and experienced multiple changes in the direction of their carjacking trajectory. Specifically, High Increasers declined to a quarterly average of 3.1 per 100,000 in 2019 before increasing to 3.7 in 2020 and 4.2 per 100,000 in 2021 but then declining slightly to 4.1 per 100,000 in 2022. Despite the small decline from 2021–22, the 2022 rate for High Increasers remained at a higher level than where it started in 2018.

Population-stratified Results

Stratified analysis of the sample was performed for four groups: cities with a population of 25,000 to 49,999 residents (N=323); cities with a population of 50,000 to 99,999 residents (N=165), cities with a population of 100,000 to 249,999 residents (N=78), and cities with a population of 250,000 or more (N=24). Here, we point out several interesting results that were not found in the combined sample analyses. The full set of population-stratified results can be found in the Appendix. Compared with the combined sample results, analyses conducted with the largest set of cities (those with populations of 250,000 residents or more) revealed the presence of groups of cities with much higher crime rates than those identified in the combined sample analyses. For example, nine of the 24 largest cities had carjacking rates at or above 10 per 100,000 over the observation period, compared with around four per 100,000 in the High Increaser group in the combined sample analysis. Compared with analyses with the other samples where the highest robbery trajectory group declined over time, the highest robbery trajectory group in the sample of cities with a population between 100,000 and 249,999 remained stable throughout the study period at a quarterly robbery rate of around 75 robberies per 100,000 residents. Among cities with between 50,000 and 99,999 residents, there were more trajectories for robbery identified than in the combined sample, totaling five and adding a “Middle Increasers” and a “Middle Decreasers” trajectory group.

Discussion

There are two key findings in this study. First, the large spike in national rates of some forms of severe violence in 2020, most notably for the offense of homicide, appear to be driven by a small percentage of U.S. cities that had relatively high crime rates prior to 2020. Second, among a sample of U.S. cities that consistently reported to NIBRS between 2018–22, there do not appear to be large groups of cities that followed qualitatively distinct violent crime rate trends such as experiencing a decrease in homicides over this time or immediately returning to pre-2020 crime levels in 2021 or 2022. This suggests that analytic methods that predict variation around a single average trend such as growth curve modeling are appropriate for research aimed at explaining between-city variation in crime trends over this period. This study also builds on prior reports (Asher, 2023; Rosenfeld, Boxerman, & Lopez, 2023) by leveraging the detailed crime incident data available in NIBRS to describe how the rate of carjacking victimizations changed over this period among a large sample of cities. We find that carjackings increased in 2020, despite an overall decreasing rate of robbery. Again, this increase is largely driven by a small subset of cities that had relatively high rates of carjacking prior to 2020. Additionally, this increase in carjackings appears to be national in scope (Harrell, 2022).

The current study is descriptive and does not attempt to explain these crime trends or differences between them such as why carjackings increased in 2020 and remained at elevated levels between 2021–22 compared with 2018-19 levels while robberies decreased during this time. Due to a lack of data on key constructs, it is difficult to identify the causes of variation in crime trends. Much occurred in 2020, including the COVID-19 pandemic and its impacts such as domestic migration, a diminishment of the criminal justice system through de-policing and court delays, a break from social controls such as school and work, and emotional and financial strain, social unrest following the murder of George Floyd and other police killings, an increase in firearm sales, and more. This descriptive study is an important first step toward understanding how and why violent crime changed in the U.S. over this period.

One interesting finding that future research should explore is the increase in carjackings, which is a form of robbery, during this period despite a simultaneous decrease in overall robberies. It is difficult to think of why proposed explanatory factors like a greater availability of guns, increased drug use, or de-policing would predict an increase in carjackings but not an increase in overall robbery. It may be that a cultural change took place that made carjacking a more attractive crime (Jacobs & Cherbonneau, 2023; Jacobs, Topalli, & Wright, 2003), possibly because of a lack of social controls and the greater availability of peers in unstructured settings because of stay-at-home orders or increased frustration and hopelessness felt by residents because of COVID-19 and the high-profile killings of Black citizens by police. To improve our understanding of crime and inform criminal justice policy, researchers should work to understand these nuances in crime trends.

In sum, our analyses show that between 2018 and 2022 in the United States, the extent of the increase in the murder, aggravated assault, and carjacking rate and decrease in the robbery rate depended on a city’s pre-existing crime levels. Cities with the highest crime rates prior to 2020 saw the greatest amount of change in crime between 2018 and 2022. In fact, cities with relatively low violent crime rates, which represent the majority of this study’s sample, saw little to no change in violent crime rates over this period. Therefore, the national narrative that severe forms of violence such as murder and aggravated assault increased drastically in the United States in 2020 seems to be driven by changes among a small subset of cities with the highest rates of violent crime. Importantly, this finding should be confirmed based on a nationally representative sample because our sample is not nationally representative.

Limitations

As mentioned, the sample used in this study is not nationally representative. Although NIBRS allows for a description and analysis of crime trends based on a much larger sample than has been used previously to describe U.S. changes in crime over COVID-19 (Arthur & Asher, 2021; Major Cities Chiefs Association, 2022; Rosenfeld & Lopez, 2022; Rosenfeld, Boxerman, & Lopez, 2023), many large cities are not represented in NIBRS and the densities of cities across U.S. states and regions are not representative of the nation. Although one can obtain nationally representative data on homicides at the county level using the Centers for Disease Control and Prevention’s (CDC) Multiple Causes of Death data, we chose to use NIBRS data for our analysis to examine variation in crime rates at a smaller geographic unit of analysis, among a larger number of offense types, and for a longer period (CDC data for 2022 was not available at the time of publication). Although this study’s results cannot speak to variation in national crime trends, they do represent a large sample of U.S. cities and should inform future research aimed at explaining violent crime trends during this period.

Additionally, there are key aspects of GBTM one must be aware of when interpreting the results of this study. Although GBTM has great value in allowing one to statistically test for subgroups in trends among a population, it is not without limitations. It is important to emphasize that the number of groups, their size, and unit group assignments are all sensitive to changes in the sample and observation period, and there can be much variation around group averages among units in a group. The group-based results presented here are a simplification of a much more complex reality. This simplification is valuable for describing variation in crime rate trends among U.S. cities, but readers should not treat a given city’s group assignment as a statistical fact or consider the estimated crime rate groups as existing in real life.

Conclusion

Findings from this study should inform violence prevention policy, as they suggest that to best prevent trauma and the loss of life following an event that is likely to increase crime, the greatest number of resources should be directed to cities with the highest violent crime rates. Future research should explain variation in crime rate trends during this time so that more targeted interventions can be developed to best prevent violence in the face of events such as those encountered in 2020. Additionally, identifying correlates that are associated with trajectory group membership can inform practice and policy in cities that reported greater changes in violence during this time. Expanding this type of analysis to include county law enforcement agencies is another opportunity to expand upon these findings. Finally, although researchers will need to overcome the typical challenges inherent to explaining crime trends like endogeneity and omitted variable bias, it appears that 2018–22 violent crime rate trends can be modeled as being generated from a single population rather than from a mixture of subgroups, which should simplify analyses. Unfortunately, the biggest obstacle researchers will face in explaining crime trends over this period is the lack of national crime data due to the FBI’s transition to NIBRS in 2021.

References

Addington LA. NIBRS as the new normal: What fully incident-based crime data mean for researchers. In: Krohn M, Hendrix N, Hall GP, Lizotte A, editors. Handbook on crime and deviance Handbooks of sociology and social research. Switzerland: Springer; 2019.

Akiyama, Y., & Nolan, J. (1999). Methods for understanding and analyzing NIBRS data. Journal of Quantitative Criminology, 15(2):225-38. Available from: https://doi.org/10.1023/A:1007531023247. DOI: 10.1023/A:1007531023247

Arthur, R., & Asher, J. (2021, February). What drove the historically large murder spike in 2020? The Intercept. Available from: https://theintercept.com/2021/02/21/2020-murder-homicide-rate-causes/

Asher, J. (2023, April). Are carjackings falling because cars are easier to steal? Jeff-alytics. Available from: https://jasher.substack.com/p/are-carjackings-falling-because-cars

Berzofsky, M., Liao, D., Couzens, G.L., Smith, E.L., & Barnett-Ryan, C. (2022). Estimation procedures for crimes in the United States based on NIBRS data. Available from: https://bjs.ojp.gov/content/pub/pdf/epcusbnibrsd.pdf

Bunker, J., Liao, D., & Berzofsky, M. (2023). Outlier Detection for Administrative Data. Presentation at the 2023 Joint Statistical Meetings of the American Statistical Association

Criminal Justice Information Services (CJIS) Division's Uniform Crime Reporting (UCR) Program. (2020). 2019.2.1 National Incident-Based Reporting System Technical Specifications. Federal Bureau of Investigation (FBI). https://le.fbi.gov/file-repository/nibrs-technical-specification.pdf/view

Dunn C. S., & Zelenock, T., J. (1999). NIBRS data available for secondary analysis. Journal of Quantitative Criminology, 15(2):239-48. Available from: https://doi.org/10.1023/A:1007583007317. DOI: 10.1023/A:1007583007317

Federal Bureau of Investigation (2018). About the UCR program. Available from: https://www.fbi.gov/file-repository/ucr/about-the-ucr-program.pdf

Federal Bureau of Investigation (2020). UCR focuses on NIBRS and other tools to make more relevant data available to users. Available from: https://ucr.fbi.gov/nibrs/2019/resource-pages/ucr-focuses-on-nibrs-and-other-tools.pdf.

Frosch, D., Maher, K., & Elinson, Z. (2022, June). Rural America reels from violent crime. ‘People lost their ever-lovin’ minds’. The Wall Street Journal. Available from https://www.wsj.com/articles/violent-crime-rural-america-homicides-pandemic-increase-11654864251.

Harrell, E. (2022). Carjacking Victimization, 1995-2021. U.S. Department of Justice, Office of Justice Programs, Bureau of Justice Statistics. Available from: https://bjs.ojp.gov/sites/g/files/xyckuh236/files/media/document/cv9521B.pdf

Howard, J. (2021). US records highest increase in nation’s homicide rate in modern history, CDC says. CNN. Available from https://www.cnn.com/2021/10/06/health/us-homicide-rate-increase-nchs-study/index.html.

Jacobs, B. A., & Cherbonneau, M. (2023). Carjacking: Scope, Structure, Process, and Prevention. Annual Review of Criminology6, 155-179.

Jacobs, B. A., Topalli, V., & Wright, R. (2003). Carjacking, streetlife and offender motivation. British Journal of Criminology43(4), 673-688.

Major Cities Chiefs Association (2022, February). Violent Crime Survey – National Totals Year End Comparison January 1 to December 31, 2021, and 2020. Available from: https://majorcitieschiefs.com/wp-content/uploads/2022/08/MCCA-Violent-Crime-Report-2021-and-2020-Year-End.pdf

Nagin, D. (2005). Group-based modeling of development. Harvard University Press.

Nagin, D. S., & Odgers, C. L. (2010). Group-based trajectory modeling in clinical research. Annual review of clinical psychology6, 109-138.

Nagin, D. S., & Tremblay, R. E. (2005). Developmental trajectory groups: Fact or a useful statistical fiction?. Criminology43(4), 873-904.

Rosenfeld, R., & Lopez, E. (2020). Pandemic, Social Unrest, and Crime in U.S. Cities. 2021 Year-end Update. Washington, D.C.: Council on Criminal Justice. Available from https://counciloncj.org/crime-trends-yearend-2021-update/

Rosenfeld, R., Boxerman, B., & Lopez, E. (2023). Pandemic, Social Unrest, and Crime in U.S. Cities. Year-end 2022 Update. Washington, D.C.: Council on Criminal Justice. Available from https://counciloncj.org/pandemic-social-unrest-and-crime-in-u-s-cities-year-end-2022-update/.

Scott, T. L., Wellford, C., Lum, C., & Vovak, H. (2019). Variability of crime clearance among police agencies. Police Quarterly22(1), 82-111.

United States. Bureau of Justice Statistics. National Incident-Based Reporting System, 2018: Extract Files. Inter-university Consortium for Political and Social Research [distributor], 2022-11-22. https://doi.org/10.3886/ICPSR37649.v2

United States. Bureau of Justice Statistics. National Incident-Based Reporting System, 2019: Extract Files. Inter-university Consortium for Political and Social Research [distributor], 2022-11-22. https://doi.org/10.3886/ICPSR38565.v2

United States. Bureau of Justice Statistics. National Incident-Based Reporting System, 2020: Extract Files. Inter-university Consortium for Political and Social Research [distributor], 2022-11-22. https://doi.org/10.3886/ICPSR38566.v2

United States. Bureau of Justice Statistics. National Incident-Based Reporting System, 2021: Extract Files. Inter-university Consortium for Political and Social Research [distributor], 2023-12-14. https://doi.org/10.3886/ICPSR38807.v1

United States. Bureau of Justice Statistics. National Incident-Based Reporting System, 2022: Extract Files. Inter-university Consortium for Political and Social Research [distributor], 2023-12-14. https://doi.org/10.3886/ICPSR38925.v1

U.S. Census. (n.d.). 5-year Data Profiles estimates, Place. U.S. Department of Commerce. Retrieved September 28, 2023 from https://data.census.gov/

U.S. Census. (2023). Gazetteer Files – Places. Retrieved from https://www.census.gov/geographies/reference-files/time-series/geo/gazetteer-files.html.

Appendix

Population-stratified Sample Results

Cities with 250,000 or more residents

Murder. Three trajectories were identified for murder rates for the subsample of 24 cities with a population of 250,000 or more residents as displayed in Figure 7. These trajectory groups can be described as “Low Increasers” (N=13, 54%), “Middle Increasers” (N=6, 25%), and “High Increasers” (N=5, 21%) (groups 1, 2, and 3, respectively). Both Low Increasers and Middle Increasers began at an average quarterly murder rate less than five per 100,000 residents in 2018 and had experienced small increases by 2022, whereas High Increasers began with a quarterly murder rate of over five per 100,000 residents, which nearly doubled throughout the study period. Average probabilities of group membership exceeded 0.96 for each group. The minimum probability for group membership in the Low Increasers group was 0.56 (1 agency had a probability less than 0.99) and 0.99 for the Middle Increasers group, whereas all five cities assigned to the High Increasers group had a probability of 1.0.

Aggravated Assault. Four trajectories were identified for aggravated assault rates. These groups can be characterized as “Low Stable” (N=6, 25%), “Low Increasers” (N=9, 38%), “Middle Increasers” (N=7, 29%), and “High Increasers” (N=2, 8%) (groups 1, 2, 3, and 4, respectively). Low Stable cities began and ended with a very low quarterly aggravated assault rate, whereas Low Increasers began with a relatively low rate but experienced a small increase by 2022. Middle increasers began at a quarterly aggravated assault rate of approximately 200 per 100,000 residents and increased slightly throughout the study period. Finally, High Increasers began at a relatively high quarterly aggravated assault rate (>250) and experienced a significant increase through 2022. Average probabilities of group membership exceeded 0.99 for each group. The minimum probability for group membership was at least 0.99 for each group.

Robbery. Four trajectories were identified for robbery rates. These groups can be characterized as “Low Stable” (N=8, 33%), “Middle Increasers” (N=7, 29%), “Middle Decreasers” (N=6, 25%), and “High Decreasers” (N=3, 13%) (groups 1, 2, 3, and 4, respectively). Low Stable cities began and ended with a very low quarterly robbery rate (less than 50 per 100,000 residents), whereas Middle Increasers began with a moderate rate but experienced a slight increase by 2022. Middle Decreasers began at a quarterly aggravated assault rate of approximately 110 per 100,000 residents and declined to a quarterly rate of approximately 60 by 2022. Finally, High Decreasers began at a relatively high quarterly aggravated assault rate (>150) and experienced a decline to approximately 125 by 2021 before increasing again in 2022 (though not to 2018 levels). Average probabilities of group membership exceeded 0.99 for each group. The minimum probability for group membership was at least 0.99 for each group.

Carjacking. Four trajectories were identified for carjacking rates. These groups can be characterized as “Low Stable” (N=10, 42%), “Low Increasers” (N=5, 21%), “Middle Stable” (N=7, 29%), and “High Increasers” (N=2, 8%) (groups 1, 2, 3, and 4, respectively). Both Low Stable and Low Increasers began at very low quarterly carjacking rates in 2018; however, the Low Stable did not experience notable changes, whereas the Low Increasers group experienced a nearly two-fold increase in carjacking rates by 2022. The Middle Stable group began and ended the study period with a quarterly carjacking rate of approximately 10 per 100,000 residents, whereas High Increasers began at a relatively high level of carjackings and experienced a small but notable incline by 2022. Average probabilities of group membership exceeded 0.97 for each group. The minimum probability for group membership was at least 0.78 for each group.

Cities with between 100,000 and 249,999 residents

Murder. Four trajectory groups were identified for the murder rate outcome for the subsample of 78 cities with a population of 100,000 to 249,999 residents as displayed in Figure 8. These groups can be characterized as “Low Stable” (N=26, 33%), “Low Increasers” (N=34, 44%), “Middle Increasers” (N=14, 18%), and “High Increasers” (N=4, 5%) (groups 1, 2, 3, and 4, respectively). Average probabilities of group membership exceeded 0.96 for each group. The minimum probability for group membership was at least 0.78 for each group. Probabilities for group membership for Middle and High Increasers exceeded 0.98 in each group. Two cities were assigned probabilities less than 0.73 (one in the Low Stable, and one in the Low Increasers group).

Aggravated Assault. Three trajectories were identified for aggravated assault rates. These groups can be characterized as “Low Stable” (N=49, 63%), “Middle Increasers” (N=23, 29%), and “High Increasers” (N=6, 8%) (groups 1, 2, and 3, respectively). Cities assigned to the Low Stable group began and ended at a quarterly rate of approximately 50 per 100,000, whereas the Middle Increasers began at a quarterly rate closer to 100 and escalated slightly by 2022. The High Increasers began closer to a rate of 200, which increased to approximately 300 by 2022. Average probabilities of group membership exceeded 0.99 for each group. The minimum probability for group membership was at least 0.78 for each group. The minimum probability for group membership was at least 0.86 for each group.

Robbery. Three trajectories were identified for robbery rates. These groups can be characterized as “Low Decreasers” (N=40, 51%), “Middle Decreasers” (N=30, 38%), and “High Stable” (N=8, 10%) (groups 1, 2, and 3, respectively). Both Low and Middle Decreasers experiences minor declines in quarterly robbery rates throughout the study period, whereas High Stable cities began at relatively higher rates but remained stable through 2022. Average probabilities of group membership exceeded 0.99 for each group. The minimum probability for group membership was at least 0.92 for each group.

Carjacking. Two trajectories were identified for carjacking rates. These groups can be characterized as “Low Increasers” (N=57, 73%) and “High Increasers” (N=21, 27%) (groups 1 and 2, respectively). Although Low Increasers began at a low quarterly rate and experienced a modest increase, quarterly rates for High Increasers nearly doubled across the study period. Average probabilities of group membership exceeded 0.98 for each group. The minimum probability for group membership was at least 0.87 for each group.

Cities with between 50,000 and 99,999 residents

Murder. Two trajectories were identified for murder rates for the subsample of 165 cities with a population of 50,000 to 99,999 as displayed in Figure 9. These groups can be characterized as “Low, Small Increasers” (N=138, 84%) and “High Increasers” (N=27, 16%) (groups 1 and 2, respectively). Average probabilities of group membership exceeded 0.97 for each group. The minimum probability for group membership was at least 0.78 for each group. Probabilities for group membership for exceeded 0.79 in each group, except for two cities with probabilities below 0.60 (one in each group).

Aggravated Assault. Five trajectory groups were identified for aggravated assault. These groups can be characterized as “Low, Stable” (N=46, 28%), “Low Increasers” (N=50, 30%), “Middle Stable” (N=37, 22%), “Middle Increasers” (N=23,14%), and “High Increasers” (N=9, 5%) (groups 1, 2, 3, 4, and 5, respectively). Average probabilities of group membership exceeded 0.94 for each group. All but six cities were assigned probabilities greater than 0.70 (three in the Low Stable and 3 in the Low Increasers groups).

Robbery. Five trajectories were identified for robbery. These groups can be characterized as “Low Stable” (N=67, 41%), “Middle Stable” (N=46, 28%), “Middle Increasers” (N=30, 18%), “Middle Decreasers” (N=13, 8%), and “High Decreasers” (N=9, 5%) (groups 1, 2, 3, 4, and 5, respectively). Average probabilities of group membership exceeded 0.92 for each group. All but eight cities were assigned probabilities greater than 0.70 (all from the Low Stable, Middle Stable, and Middle Increasers groups).

Carjacking. Two trajectories were identified for carjacking rates. These groups can be characterized as “Low Increasers” (N=125, 76%) and “High Increasers” (N=40, 24%) (groups 1 and 2, respectively). Both groups experienced modest increases in their quarterly rates. Average probabilities of group membership exceeded 0.97 for each group. All but one agency in the High Increasers group was assigned a probability of 0.70 or higher.

Cities with between 25,000 and 49,999 residents

Murder. Two trajectories were identified for murder rates for the subsample of 323 cities with a population of 25,000 to 49,999 residents as displayed in Figure 10. These groups can be described as “Low Stable” (N=281, 87%) and “High Increasers” (N=42, 13%) (groups 1 and 2, respectively). The High Increasers group doubled in its quarterly murder rates by 2022. Average probabilities of group membership exceeded 0.96 for each group. All but four cities were assigned a probability of 0.70 or higher.

Aggravated Assault. Four trajectories were identified for aggravated assault: “Low Stable” (n=147, 46%), “Middle Stable” (N=110, 43%), “Middle Increasers” (n=47, 15%), and “High Increasers” (N=15, 5%) (groups 1, 2, 3, and 4, respectively). Both the Middle and High Increasers groups experienced only minor increases in aggravated assault. Average probabilities of group membership exceeded 0.98 for each group. All but nine cities across the Low Stable, Middle Stable, and Middle Increasers groups were assigned a probability of 0.70 or higher.

Robbery. Three trajectory groups were identified for robbery: “Low Stable” (N=184, 57%) and “Middle Decreasers” (N=112, 35%), and “High Decreasers” (N=27, 8%) (groups 1, 2, and 3, respectively). The High Decreases group experienced an especially notable decline in its quarterly robbery rates by 2022. Average probabilities of group membership exceeded 0.97 for each group. All but six cities in the Low Stable and Middle Decreasers groups were assigned a probability of 0.70 or higher.

Carjacking. Two trajectories were identified for carjacking: “Low Stable” (Group 1, N=281, 87%) and “High Increasers” (N=42, 13%) (groups 1 and 2, respectively). The High Increasers group doubled in its quarterly carjacking rates by the end of the study period. Average probabilities of group membership exceeded 0.94 for each group. All but 12 cities (seven in the Low Stable and five in the High Increasers groups) were assigned a probability of 0.70 or higher.

Table 1. Description of sample of cities with law enforcement agencies that fully reported to NIBRS from 2018–22 and remaining places in the United States with a population of 25,000 or more and with police agencies that did not fully report to NIBRS during this timeα

Mean (s.d.)

Full-period NIBRS Reporters

Non-Full-period NIBRS Reportersβ

Population

82,261.79 (124,275.60)

93,320 (302,081.30)

Income

83,711.31 (32,955.14)

89,894.55 (35,693.46)

% female

51.13 (1.99)

51.08 (2.03)

% male

48.87 (1.99)

48.92 (2.03)

% 18 years or under

22.55 (4.65)

23.42 (4.65)

% White

77.62 (14.77)

67.67 (19.37)

% Black or African American

10.41 (13.60)

13.47 (16.14)

% Asian

4.54 (5.32)

7.67 (10.21)

% American Indian/Alaska Native

0.61 (1.11)

0.57 (1.17)

% Native Hawaiian/Pacific Islander

0.18 (0.53)

0.25 (0.88)

% Hispanic or Latino

12.15 (12.52)

23.93 (22.24)

% with a bachelor’s degree or higher

35.18 (15.64)

33.88 (16.25)

% living with a disability

12.61 (3.81)

11.37 (3.63)

Unemployment Rate

5.61 (2.36)

6.06 (2.53)

% of households living below the poverty line in the last year

10.14 (6.26)

10.17 (6.53)

Total (column %)

Total (column %)

Region

Northeast

114 (21.47%)

137 (10.72%)

Midwest

146 (27.50%)

261 (20.42%)

South

151 (28.44%)

460 (35.99%)

West

120 (22.60%)

420 (32.86%)

Total

531

1,278

α Data compiled using 2018 5-year place estimates provided by the U.S. Census Bureau.

β Includes all places designated by the U.S. Census as cities, towns, Census-designated places, and boroughs.

Table 2. Annual crime rates per 100,000 among cities with law enforcement agencies that fully reported to NIBRS from 2018 and 2022, by offense type

Mean (s.d.)

Murder

Aggravated Assault

Robbery

Carjacking

2018

5.08 (8.68)

269.64 (269.02)

93.74 (103.44)

5.08 (9.96)

2019

5.38 (9.51)

268.61 (271.03)

84.16 (93.45)

4.65 (9.37)

2020

7.73 (13.78)

312.99 (344.43)

81.77 (86.07)

5.55 (10.67)

2021

7.96 (13.76)

320.41 (347.38)

75.77 (80.67)

6.65 (11.43)

2022

7.50 (12.37)

319.05 (330.68)

75.30 (85.81)

6.88 (13.18)

Figure 1. Change in violent crime, by offense type, 2010-2022

Figure 2. An example of several distinct group trends creating a misleading average trend. Reprinted with author’s permission from Scott, T., Wellford, C., Lum, C., & Vovak, H. (2019). Variability of crime clearance among police agencies. Police Quarterly, 22(1), 82-111

Figure 3. Estimated trajectory groups, murder rate

Figure 4. Estimated trajectory groups, aggravated assault rate

Figure 5. Estimated trajectory groups, robbery rate

Figure 6. Estimated trajectory groups, carjacking rate

Figure 7. Crime rate trajectory groups, by offense type, cities with 250,000 or more residents

Figure 8. Crime rate trajectory groups, by offense type, cities with between 100,000 and 249,999 residents

Figure 9. Crime rate trajectory groups, by offense type, cities with between 50,000 and 99,999 residents

Figure 10. Crime rate trajectory groups, by offense type, cities with between 25,000 and 49,999 residents

Comments
0
comment
No comments here
Why not start the discussion?