1.1 Contents of the Element

Crime Dynamics considers many of the major theoretical and empirical contributions of the criminologists, sociologists, historians, and economists who have sought to explain the sources of change in crime rates. This study, like most crime trends research, is avowedly macroscopic in orientation. The focus is on crime trends in nations, cities, regions, and other large subnational units during the past several decades. Some potentially relevant topics, developmental and life-course research on individual change in criminal behavior, for example, are not covered comprehensively or are omitted. The Element also focuses on trends in street crime, primarily homicide, in the United States. Comparable data on trends in corporate, white-collar, and online crime do not exist, and homicide is the most serious, best measured, and most frequently studied criminal offense in the crime trends research literature. Comparable studies of crime trends in other nations are left to others.

The sections in this Element cover the major data sources in the study of crime trends; the impact on violent crime of property crime, the prevalence of firearms, and imprisonment; the demography of crime trends; the relationship between crime and the economy; crime trends and institutional change; exogenous shocks that produce large and unexpected changes in crime rates; forecasting future crime rates; and where crime trends theory and research should go from here. Section 2 describes the most widely used sources of data in the study of crime trends. Some data sources are based on offenses recorded by police departments. Others are from public health sources and surveys of crime victims. Most of the data are collected and disseminated by government agencies, but data compiled by private entities play an increasingly prominent role in recent crime trends research. Each data source has strengths and limitations, and the different sources are best viewed as complementary rather than incompatible.

Section 3 discusses three sources of change in violent crime: property crime, firearms, and imprisonment. While some prominent criminologists have argued that property crime has little or no effect on violent crime, and homicide in particular, the section subjects this contention to critical scrutiny and offers reasons why the sheer volume of property crimes should be expected to exert a sizable influence on criminal violence. There is little dispute that homicide is related to the availability of firearms. More controversial, inside and outside of criminology, is whether widespread access to firearms increases or reduces rates of homicide and other violent crimes. The relationship between crime and imprisonment trends is also controversial. Recent research suggests that increases in imprisonment probably result in decreases in crime rates, but the effects generally are small and diminish even further at high levels of imprisonment.

Section 4 examines differences in crime trends by race, ethnicity, and gender. It may come as some surprise that the crime trends among males and females and in the Black, White, and Hispanic populations are quite similar. What distinguishes these groups is their level of crime, not the change in their crime rates over time. This could mean that the common group trends are a consequence of common causes. Not all group-specific crime trends are the same, however. For example, this section presents evidence that intimate-partner homicide has declined more rapidly over time among males than females.

Section 5 considers the relationship between crime trends and changes in the age composition of the population. Street crime rates peak in the late teens and early twenties. As the adolescent and young adult segment of the population grows, as it did during the 1960s and 1970s resulting from the “baby boom” after the Second World War, crime rates turn up. As the baby boom cohorts were replaced by smaller age cohorts in the 1980s, crime rates came down. But not for long. Crime rates rose again beginning in the late 1980s, even as the size of the youthful population continued to decline. Changes in the size of this population place upward or downward pressure on crime rates, but other factors often outweigh the effect on crime trends of changes in the age composition of the population. Just as the size of age cohorts matters for crime trends, so does when the cohorts were born. Cohorts born at different times can differ in their current crime rates, quite apart from the effects of age or other current conditions. The influence on crime rates of the circumstances of birth and early development is an indispensable part of the story of why crime rates move up and down over time.

Research on the impact of economic conditions on crime rates is as old as the study of crime trends itself. Section 6 discusses the evolution of this research from early studies of the relationship between crime trends and the business cycle, to investigations of the effects of unemployment on crime rates, to recent research on how crime rates respond to changes in consumer sentiment and inflation. This is one of the most stimulating and important areas of inquiry in the study of crime trends because it bridges the theoretical interests and methodological tools of economists and criminologists, and it necessarily directs attention to policy and institutional realms well beyond the criminal justice system.

To this point, Crime Dynamics covers the data and much of the research on crime trends, but stops short of asking what it all means. What are the underlying structures, processes, and mechanisms that help to make sense of the disparate demographic, social, and economic influences on crime trends discussed thus far? In short, is there a theory of crime trends? Section 7 presents the outlines of such a theory that is rooted in what has been termed the “new institutionalism” in criminology. Social institutions are the guideposts of society. Institutional structure, regulation, and performance shape the incentives, opportunities, and constraints that result in both long- and short-run changes in crime rates over time.

Not all sources of change in crime rates are knowable in advance. On occasion crime rates change abruptly without prior warning. No one to my knowledge predicted the Covid-19 pandemic or the effects, which turned out to be quite complex, it would have on crime rates. Section 8 discusses the impact on crime rates of such “exogenous shocks” and how they elude the conventional explanatory tools of crime trends research. Exogenous shocks not only upset “normal science” studies of past crime trends, they pose a significant challenge to forecasting future crime rates. Forecasting has all but disappeared from criminology, in no small part because of embarrassingly erroneous claims of an impending crime boom by a few prominent criminologists just as crime rates were beginning their historic decline in the 1990s. Section 9 contends that, when carefully done, crime forecasting can benefit both policymaking and theory testing and is a natural and needed extension of macrolevel research on crime. The final section of the Element points to improvements in theory, data, and research methods that portend a bright future for the study of crime trends.

This Element is written in nontechnical language and statistics are kept to a minimum. Where statistical terms and procedures are used, they are described in plain language in the text. Interested readers are directed to an appendix for supporting technical material on crime forecasting in Section 9. Much of the story is told in time-series graphs of crime rates and related phenomena (e.g., the age composition of the population, firearm prevalence, inflation, imprisonment, confidence in the police). The story begins in the following section with a description of the major sources of data on crime trends.

2.1 Uniform Crime Reports

The nation’s major source of crime data is the Uniform Crime Reports (UCR). The UCR program began in 1930 and is housed at the FBI.Footnote ² The data consist of eight major violent and property offenses and are based on crimes reported to and recorded by local law enforcement agencies. The violent crimes are criminal homicide, rape, robbery, and aggravated assault and the property crimes are burglary, larceny, motor vehicle theft, and arson. Other generally less serious crimes such as forgery, drug law violations, and simple assaults are also included in the FBI’s annual series Crime in the United States.

In 2021 the FBI transitioned from the UCR summary system to the National Incident-Based Reporting System (NIBRS), a far more detailed compilation of crime data that is based on individual crime incidents. Many agencies did not report NIBRS data for 2021, however, and the FBI did not include the 2021 data in its multiyear trend presentations (https://crime-data-explorer.fr.cloud.gov/pages/explorer/crime/crime-trend). The problematic transition to NIBRS reveals a deeper problem in the nation’s crime statistics based on law enforcement data: participation in the system is voluntary.

2.2 The NIBRS Transition

The FBI released its annual report on crime in the United States for the year 2021 on October 5, 2022. This was the first report under NIBRS, a new and far more detailed data format. NIBRS replaced the UCR “summary system” the FBI has used since the 1930s that includes major felony offenses and arrests recorded by local law enforcement agencies. NIBRS counts many more offenses and provides much greater detail about them, such as the age, sex, and race of victims and the circumstances of the crimes. NIBRS had been in the works since the 1980s, and so full conversion, even if it took nearly forty years to accomplish, has to be counted as good news.

The bad news is that the conversion to NIBRS was far from complete. Only 63% of law enforcement agencies, covering about 65% of the US population, had made the switch to NIBRS by the FBI’s deadline of January 1, 2022. And many other agencies submitted NIBRS data that covered only part of the year – just 52% submitted data for all twelve months of 2021. The police departments of some of the nation’s largest cities submitted no data at all, including the departments in New York, Los Angeles, Phoenix, and San Francisco. And some states were barely covered by the NIBRS data. Only 15 of California’s 740 law enforcement agencies, 40 of Pennsylvania’s 1,504 agencies, and 2 of Florida’s 757 law enforcement agencies sent in data. The crime data for nonparticipating agencies had to be estimated based on data for prior years and comparisons with agencies of similar population size and composition.

With participation rates this low and uneven, and the need to estimate such a large number of unknowns, the FBI itself cautioned against comparing the 2021 data with data from previous years.Footnote ³ That meant that the nation’s official crime statistics could not answer the most basic question about crime, whether it is going up or down. That would be a problem in any year. It was an especially serious problem in 2022, when crime had again moved to the forefront of public concerns and loomed as a leading issue in the fall midterm elections. The uncertainties surrounding the crime data were grist for the political mill. If your position is that crime increases are exaggerated, just cite the FBI’s estimate that robbery went down by 9%. If you want voters to believe crime is out of control, cite the estimate that murder went up over 4%. Either way, who’s to say you’re right or wrong?

It did not have to be this way. The FBI should not be faulted for pushing hard on NIBRS. It is a much better statistical system and should have been fully implemented long before 2022. Moreover, law enforcement agencies were not caught unawares that NIBRS was coming. The Department of Justice announced the conversion to NIBRS in 2015 and distributed over $120 million to prepare for the transition.Footnote ⁴ But the FBI knew well ahead of the January 2022 deadline that NIBRS participation would be much lower than the 85–95% participation rate in the former summary system and would therefore require far more estimation than needed in the past.Footnote ⁵ At that point, in the fall of 2021, the FBI could have decided to allow agencies that would not be able to meet the upcoming deadline to submit summary data in lieu of NIBRS. The FBI chose instead to require full compliance with NIBRS by the deadline with no exceptions.

The FBI knew that the NIBRS conversion would be a technical challenge for many agencies. That is why they were given fair warning years ago and funds to support the transition. But technical issues were not the primary stumbling block that slowed the transition. The major obstacle was that law enforcement agencies are not required to submit crime data to the FBI. Voluntary participation in the nation’s crime reporting system might have made sense in 1930 when the FBI’s Uniform Crime Reporting Program was established. Many law enforcement agencies, with venerable traditions of independence and local control, would have resisted a mandate to send sensitive information to Washington that would be made public and could be used to criticize their performance. But the days when crime data were treated as the personal property of the local sheriff or police chief are long gone. If the FBI cannot or will not require local law enforcement to submit their crime data, Congress can, and should.

A federal mandate to submit crime data to the FBI would not have guaranteed full participation in NIBRS, but it probably would have increased participation, reducing the need for extensive estimation. Meanwhile, the FBI put itself in the odd position of mandating that local agencies submit crime data under the NIBRS system while not requiring that they submit any crime data at all.Footnote ⁶ We are left with the hope that 2021 was a one-off anomaly in the nation’s ninety-year-old crime data infrastructure. A serious glitch to be sure, and one that could have been avoided, but a teachable moment that offers important lessons for how to operate a bona fide federal statistical system.

2.3 National Crime Victimization Survey

A strength of the UCR-NIBRS data is that they are available for counties, cities, metropolitan areas, and census regions as well as the nation as a whole. The data are subject to crime classification errors, however, and the county-level data are often incomplete (Reference Maltz and TargonskiMaltz and Targonski 2002; Reference Nolan, Haas and NapierNolan et al. 2011). The chief drawback of the UCR data, however, is that they exclude crimes that are not reported to the police.

The second major source of US crime data and statistics is the National Crime Victimization Survey (NCVS). The NCVS is an annual survey since 1973 of the US household population that asks whether individuals age twelve and older have been the victim of a property crime or violent crime, excluding homicide, during the last six months. Respondents are also asked whether the crime was reported to the police. The extent of unreported crimes varies substantially across offense types. For example, in 2019 victims or others reported 79.5% of motor vehicle thefts to the police, compared with just 33.9% of rapes or other sexual assaults (Reference Morgan and ThompsonMorgan and Thompson 2021).

NCVS data are currently available for the nation as a whole and a subset of states (Reference Grace and MorganKena and Morgan 2023). One limitation is that persons who reside in institutional settings such as jails or nursing homes are not included in the survey. The NCVS does provide a more complete picture of crime than the UCR, however, and the two crime data systems should be viewed as complementing one another, with one filling in the gaps left by the other (Reference Lynch and AddingtonLynch and Addington 2007; Reference Morgan and ThompsonMorgan and Thompson 2022).

2.4 Other Homicide Data Sources

Three additional US data sources are available for homicide. One is the FBI’s Supplementary Homicide Reports (SHR), which provides data on homicide incidents by victim and offender (when known) age, race, and sex; weapon type; victim–offender relationship (e.g., family member, intimate partner, acquaintance, stranger); and attributes of the incident (e.g., drug-related, gang-related, argument). A second source is the Fatal Injury Reports from the National Vital Statistics System (NVSS), which compiles homicide and other data on cause of death from local coroners and medical examiners. Homicide counts and rates are typically somewhat higher than those from the UCR and SHR, in part because reporting to the NVSS is mandatory while reporting to the UCR is voluntary. Nonetheless, time trends derived from the two homicide data sources correspond closely (see https://bjs.ojp.gov/content/pub/pdf/ntmh.pdf). The National Violent Death Reporting System (NVDRS), begun in 2002, compiles data on violent deaths from multiple sources and has been implemented in all fifty states, the District of Columbia, and Puerto Rico.

2.5 Nongovernmental and European Sources

Other nongovernmental crime data sources are also available for particular jurisdictions, offense types, and time periods in the United States. For example, the Gun Violence Archive compiles data from multiple sources on gun-related homicides, suicides, and injuries. The Council on Criminal Justice publishes monthly trend data from the police departments of several dozen large cities for ten different offenses. AH Datalytics provides year-to-date homicide data from the police departments of about eighty large cities.

In addition to US sources, data sources for international crime data include the European Sourcebook of Crime and Criminal Justice Statistics, the International Crime Victimization Survey (see Van Reference Dijk, Jan and Joudo LarsenDijk et al. 2022), the United Nations Surveys on Crime Trends and the Operations of Criminal Justice Systems, and, for homicide, the World Health Organization. Descriptions and data bases for these and the US sources are listed in Appendix I.

Criminology does not suffer from a lack of data to track crime trends in the United States and elsewhere, although each source contains strengths and limitations that should always be examined and compared with those of other sources. Moreover, it should be kept in mind that, with the exception of the SHR, all of the crime data sources used in this Element are limited to counts and rates of criminal victimization and not offending. Even the SHR data on homicide offenders must be viewed with caution because of the large amount of missing data, which has prompted researchers to impute estimates for the unobserved offender attributes (Reference Fox and SwattFox and Swatt 2009). Finally, as discussed in the final section, crime trends may differ in important ways across areas within cities, and the data needed to track these subunit trends are not available in standardized form. With these caveats in mind, we now turn to how the crime data have been used to develop and test hypotheses on the sources of change in crime rates over time.

3.1 Are Homicide and Property Crime Causally Related?

Throughout the Element data from multiple sources are presented to illustrate trends in crime and other conditions. As a starting point consider the relationship between trends in homicide and property crime in the United States. Figure 1 displays UCR homicide and property crime rates per 100,000 US population between 1960 and 2020. Because the property crime rate is so much higher than the homicide rate, to reveal the relationship between them over time they have been scaled on different axes, with property crime on the left and homicide on the right. Both homicide and property crime increased from 1960 to 1980, fell for a few years, increased again through the early 1990s, and then dropped more or less continuously for the next two and a half decades during what has been called the “Great American Crime Decline” (Reference ZimringZimring 2007). Up until that point, property crime and homicide rates moved together over time. Almost without fail, whenever property crime increased or decreased, so did homicide. The correlation between the two series over the sixty-year period is an impressive .864.Footnote ⁷

Figure 1 US homicide and property crime rate, 1960–2020.

Source: Uniform Crime Reports

The strong correspondence between the homicide and property crime trends was interrupted in 2015, when the homicide rate began to rise while the property crime rate continued to fall. The homicide rate in 2016 was 21% higher than in 2014. Homicide subsided during the next few years, only to shoot up again in 2020, this time by 30% over the rate in 2019. Such abrupt and substantial increases in homicide are unlikely to have been caused by typically slow-moving changes in demographic and economic conditions. They coincided with widespread social unrest following controversial incidents of police violence in Ferguson, Missouri, in late 2014 and Minneapolis in May 2020. The homicide spike in 2020 also occurred during the first wave of the Covid-19 pandemic. These conditions constituted “exogenous shocks” that altered the normal trajectory of homicide (Reference RosenfeldRosenfeld 2018). This issue is taken up in Section 8.

The relationship between homicide and property crime trends, at least until the break in 2015, raises the question of whether they are causally related. If we accept the answer by Reference Zimring and HawkinsZimring and Hawkins (1997) that “crime is not the problem” then the apparent relationship between changes over time in homicide and property crime is probably spurious. The sheer magnitude of the long-run relationship, however, and the fact that the year-to-year changes are generally similarFootnote ⁸ suggest that a causal association between homicide and property crime should not be discounted without further investigation.

Involvement in property crime exposes offenders to dangerous persons and situations and is a risk factor for violence. Criminal activity typically occurs during the evening hours, away from the home, and in the company of likeminded others, especially among youthful offenders. These activity patterns and associated proximity to offenders raise the probability of violent victimization (Reference Hindelang, Gottfredson and GarofaloHindelang et al. 1978). Violent crime may result directly from property crime, as when a burglar assaults a homeowner who interrupts his plans. Co-offenders assault and rob one another; criminals sometimes attack or intimidate witnesses, suspected “snitches,” or others with incriminating information. In these and other ways, violent crime “feeds off” property crime (Reference FelsonFelson 2002: 105–119). Reference DobrinDobrin (2001) found that individuals who have been arrested for property crimes are significantly more likely than others to become the victim of a homicide, controlling for age, race, sex, and local socioeconomic conditions.

Moreover, the huge difference between the volume of property crimes and homicides means that it takes a very small proportion of property crimes to generate a sizable proportion of homicides. In 2019, for example, there were 16,425 homicides and 6,925,677 property crimes recorded by the UCR in the United States, a ratio of 422 property crimes for every homicide. If just one tenth of 1% of the property crimes ended in a homicide, that would generate nearly 7,000 homicides, or about 42% of the total number of homicides in 2019. Property crime does not have to be very violent in the aggregate, therefore, to produce a notable increase in the homicide rate.

3.2 Firearms and Homicide

Exposure to other offenders armed with a gun is one way involvement in property crime can contribute to homicide. A controversial issue in criminology is the impact of firearms on the homicide rate. Some analysts claim that an armed society is a safer society. As Reference LottLott (1998, Reference Lott2010) put it, more guns result in less crime. Lott argued that “right to carry” gun laws that expand access to firearms reduce violent crime by making offenders think twice about attacking persons who could be armed. Lott’s research has been heavily criticized. A study commissioned by the National Research Council reviewed the available research on the impact of gun laws on violent crime, conducted its own analysis, and concluded that “with the current evidence it is not possible to determine that there is a causal link between the passage of right-to-carry laws and crime rates” (National Research Council 2005: 150).Footnote ⁹ This conclusion was controversial even within the committee itself. One member dissented, arguing that “the evidence presented by Lott and his supporters suggests that RTC laws do in fact help drive down the murder rate, though their effect on other crimes is ambiguous” (271).

The research on the relationship between the prevalence of firearms and the homicide rate indicates that the relationship is probably reciprocal. Increases in firearm availability spur increases in homicide, and increases in homicide lead to an upswing in firearm acquisition (RAND 2018). There is no disputing that firearms are plentiful in the United States and that the use of a firearm elevates the fatality rate in assaults (Reference Braga, Griffiths, Sheppard and DouglasBraga et al. 2021; Reference Kleck and McElrathKleck and McElrath 1991). A key methodological question in research on the relationship between firearm prevalence and homicide is how to measure firearm prevalence. Most of the studies reviewed in RAND (2018) rely on a proxy measure known as “Cook’s Index,” the percentage of suicides committed with a firearm. This measure has been validated in comparisons with more direct measures of firearm prevalence from surveys (Reference Deborah, Cook and MillerAzrael et al. 2004; but see National Research Council 2005). Figure 2 displays the year-over-year percentage change in the US homicide rate and Cook’s Index between 1982 and 2020. The two series are modestly correlated (r=.334), such that increases in firearm prevalence as proxied by the percentage of suicides by firearm tend to be associated with increases in homicide rates. But the correlation between gun prevalence and homicide does not settle debate over whether more guns lead to more homicides or more homicides lead to more guns.

Figure 2 Year-over-year percentage change in US homicide rate and percentage of suicides committed with a firearm, 1982–2020.

Source: Uniform Crime Reports; National Center for Injury Prevention and Control

Efforts to identify the causal direction in the firearms-homicide relationship and, by extension, the precise role of firearm prevalence in the study of homicide trends remain inconclusive (RAND 2018).

3.3 Crime and Imprisonment

Another controversial issue in criminology and more broadly is whether imprisonment reduces crime rates and, if so, by how much. For most of the twentieth century, US imprisonment rates were on par with those in most of the developed world, about 100 prisoners per 100,000 population. That began to change in the mid-1970s, when the US prison population started a 30-year rise, peaking at about 500 prisoners per 100,000 population in 2008 and then falling to under 400 per 100,000 in 2020. The past half century has come to be known as the era of “mass incarceration” in the United States.

The logic behind an imprisonment-crime connection is based on assumptions about the deterrent effects and incapacitation effects of imprisonment. In theory, the experience of imprisonment could reduce future offending by deterring former prisoners from committing new crimes (specific deterrence). The threat of imprisonment could reduce crime by deterring the general population from committing crime (general deterrence). Imprisonment could also reduce crime by preventing incarcerated persons from committing crimes against the general public (incapacitation). In practice, it has proven virtually impossible to reliably isolate each of these hypothesized crime-reduction mechanisms from the others in research on the impact of imprisonment on crime rates.

A research review by the National Research Council (2014) concluded that increases in imprisonment rates over time have a modest effect, at best, on crime rates. This conclusion is illustrated in Figure 3, which presents the year-over-year percentage change in the US violent crime rate and imprisonment rate (the number of state and federal prisoners per 100,000) between 1961 and 2020. A small negativeFootnote ¹⁰ correlation exists between the two series (r=-.179), and they frequently move in the same direction, such as during the 1990s crime drop. Most studies report a negative effect of imprisonment on crime – more imprisonment, less crime – although estimates of the magnitude of the imprisonment effect vary so widely that the National Research Council was unable to arrive at a common estimate.

Figure 3 Year-over-year percentage change in US violent crime rate and imprisonment rate, 1961–2020.

Source: Uniform Crime Reports; Bureau of Justice Statistics

Part of the uncertainty over the impact of imprisonment on crime stems from the fact that the rate of imprisonment is endogenous – that is, it depends on the crime rate. Other things equal, when the crime rate goes up, so does the imprisonment rate. The relatively few studies that have sought to untangle the mutual causation inherent in the imprisonment–crime relationship find somewhat stronger imprisonment effects on crime than studies that do not adjust the estimates for endogeneity (e.g., Reference LevittLevitt 1996). A second difficulty in precisely determining the effect of changes in the imprisonment rate on changes in the crime rate is that the effect is not constant over the scale of imprisonment: As the imprisonment rate rises, its effect on the crime rate becomes smaller (Reference Johnson and RaphaelJohnson and Raphael 2012). Perhaps the best one can say about the effect of the imprisonment rate on the crime rate is that it probably exists, it is probably negative, and its magnitude probably diminishes at higher rates of imprisonment.

4.1 Black, White, and Hispanic Homicide Trends

It is widely recognized that homicide rates are higher in the Black population than in the White population in the United States. Less commonly understood is that it has not always been this way. Prior to the twentieth century, White homicide rates often exceeded those in the Black population. Black rates began to rise with increasing urbanization and Jim Crow discrimination at the turn of the century (Reference DuBoisDuBois 1996[1899]; Reference RothRoth 2009). The racial gap in homicide rates has varied somewhat in recent decades, but it remains an enduring form of social inequality in the United States.

Figure 4 displays NVSS homicide rates by race and ethnicity between 1990 and 2020. The homicide series begin in 1990 because that is the first year the NVSS partitioned homicide rates by Hispanic ethnicity. The main takeaway from Figure 4 is the marked difference in homicide levels across the three groups over the entire thirty-year period. The non-Hispanic Black homicide rate was roughly ten times greater than the non-Hispanic White rate in the early 1990s. From the late 1990s into the current century the Black–White homicide ratio decreased somewhat before returning to ten-to-one in 2020. Throughout the period Hispanic homicide rates moved between the rates for the non-Hispanic Black and White populations. During the early 1990s, the Hispanic rates were about four times higher than the non-Hispanic White rates. More recently, the Hispanic–White homicide ratio has fallen to about two-to-one.

Figure 4 Non-Hispanic White, non-Hispanic Black, and Hispanic homicide rate, 1990–2020.

Source: National Vital Statistics System

The large difference in the homicide rates of the three race-ethnic groups shown in Figure 4 makes it difficult to determine whether their year-to-year trends are similar or diverge from one another. Figure 5 places the non-Hispanic Black and White homicide rates on separate vertical axes and reveals very similar patterns of change in the two trends. Both homicide rates fell during the 1990s, flattened during the first years of the current century, rose in 2015 and 2016, and rose again in 2020. Hispanic homicide rates, not shown in Figure 5, follow the same pattern. The only notable divergence between the non-Hispanic Black and White homicide rates occurred in 2001 when White homicides increased as a consequence of the September 11 terrorist attack. The nearly 3,000 deaths from the attack are counted as homicides in the NVSS but not in the UCR. In general, however, the Black and White homicide series rise and fall together, which suggests that, despite the difference in their levels, their similar time trends result from common causes.

Figure 5 Non-Hispanic White and non-Hispanic Black homicide rate, 1990–2020.

Source: National Vital Statistics System

Extensive research has been conducted on the relationship between race and violent crime in the United States. Early studies examined the effect of race, typically percent Black, on total homicide and violent crime rates (e.g., Reference Blau and BlauBlau and Blau 1982). More recent research has focused on race-specific rates of homicide and, on occasion, other violent offenses (Reference Steffensmeier, Ulmer, Feldmeyer and HarrisSteffensmeier et al. 2010). Much of this research has addressed the “racial invariance” thesis, which holds, in general terms, that violence in both the Black and White populations is rooted in the same conditions of structural disadvantage (see Reference Sampson, Julius Wilson and KatzSampson et al. 2018, for a comprehensive review). A few studies have included the Hispanic population in assessments of the invariance thesis (Reference Light and UlmerLight and Ulmer 2016; Reference PhillipsPhillips 2002; Reference Steffensmeier, Ulmer, Feldmeyer and HarrisSteffensmeier et al. 2010; Reference Ulmer, Harris and SteffensmeierUlmer et al. 2012). Other recent research has investigated levels and time trends in the Black–White homicide gap (e.g., Reference Gary, Baumer and O’BrienLaFree et al. 2010; Reference Light and UlmerLight and Ulmer 2016; Reference Lo, Howell and ChengLo et al. 2012; Reference Ulmer, Harris and SteffensmeierUlmer et al. 2012; Reference Velez, Krivo and PetersonVelez et al. 2003).

Results of evaluations of the racial invariance thesis depend on how “invariance” is defined (Reference Steffensmeier, Ulmer, Feldmeyer and HarrisSteffensmeier et al. 2010). If invariance means that similar conditions of structural disadvantage are significantly associated in the same direction with Black and White homicide rates, research results generally support the thesis (Reference Light and UlmerLight and Ulmer 2016; Reference Sampson, Julius Wilson, Hagan and PetersonSampson and Wilson 1995; Reference Sampson, Julius Wilson and KatzSampson et al. 2018). But if the invariance thesis also requires that the magnitude of the effects of structural disadvantage not differ significantlyFootnote ¹¹ for Black and White homicide, the research is less supportive (Reference OuseyOusey 1999; Reference PhillipsPhillips 2002; Reference Steffensmeier, Ulmer, Feldmeyer and HarrisSteffensmeier et al. 2010). Even the less restrictive conception of invariance has been challenged, for example by research showing that Black and White homicide rates are uncorrelated across metropolitan areas, which implies that the sources of the race-specific rates are different (Reference Feld and BauldryFeld and Bauldry 2018). A further complication arises from the problem of “restricted distributions,” or non-overlapping race-specific measures of structural disadvantage (Reference McNultyMcNulty 2001; Reference Ulmer, Harris and SteffensmeierUlmer et al. 2012). Finally, mixed results regarding the effect of racial segregation on homicide rates is an additional source of uncertainty in research on the invariance thesis.

Sound theoretical reasons exist for positing a positive effect of residential racial segregation on homicide in the Black population. Reference MasseyMassey (1995) points out that because segregation concentrates the correlates of violence in Black communities, it also concentrates high rates of violence in these geographic areas. Not all research finds a significant effect of segregation on the racial gap in homicide (e.g., Reference Gary, Baumer and O’BrienLaFree et al. 2010; Reference PhillipsPhillips 2002). Studies that do find a segregation effect on Black homicide rates – which is consistently positive – report a null or negative effect on White rates, especially in the context of “hyper” or “macro” segregation that crosses municipal boundaries (Reference Light and ThomasLight and Thomas 2019; Reference MasseyMassey 1995; Reference O’Flaherty and SethiO’Flaherty and Sethi 2010; but see Reference Peterson and KrivoPeterson and Krivo 2010). The contrasting effects of racial segregation on Black and White homicide rates could be interpreted as contradicting the racial invariance thesis advanced by Reference Sampson, Julius Wilson and KatzSampson et al. (2018), who argue that if one factor consistently reduces homicide in one community and increases it in another “that would be evidence against the theory” (p. 17).

Though Latinos are the largest ethnic minority in the United States, fewer studies of their crime patterns exist than those of non-Latino Black and White Americans. As shown in Figure 4, nationwide Latino homicide rates are somewhat higher than those in the non-Latino White population and much lower than in the non-Latino Black population. Between 2016 and 2020, the yearly average Latino homicide rate was 5.5 per 100,000 population, the non-Latino White rate was 2.8 per 100,000, and the non-Latino Black rate was 24.7 per 100,000.Footnote ¹² The research on homicide that includes the Latino population has found that Latino immigration into the United States either is unrelated to or reduces crime and violence and that Latino homicide rates are related to the same conditions of structural disadvantage as the homicide rates in the non-Latino Black and White populations.

There is a strong research consensus that immigration generally and Latino immigration in particular are unrelated to or negatively associated with crime and violence in the United States (see the extensive research review in Reference Ousey and KubrinOusey and Kubrin 2018). Non-Latino White, non-Latino Black, and Latino homicide rates all are related to levels of poverty, female headship, and other indicators of structural disadvantage (Reference Steffensmeier, Ulmer, Feldmeyer and HarrisSteffensmeier et al. 2010; Reference Ulmer, Harris and SteffensmeierUlmer et al. 2012). Reference PhillipsPhillips (2002) estimated that about half of the Black–White homicide gap would be eliminated if the Black and White population had the same demographic and socioeconomic characteristics. Over 100% of the homicide gap between non-Latino Whites and Latinos would be eliminated – the homicide rate of the Latino population, in other words, would be lower than that of the non-Latino White population – if they experienced identical structural conditions (see Reference Lauritsen and LangLauritsen et al. 2018, for a similar result with respect to the nonlethal violence gap).

The idea that violent crime in the Latino population is lower than would be expected based on indicators of structural disadvantage has been termed the “Latino paradox” and “immigrant revitalization perspective” (Reference Ben, Sun, Harris and CullenFeldmeyer et al. 2022; Reference MartinezMartinez 2002; Reference Martinez and LeeMartinez and Lee 2000; Reference SampsonSampson 2008; Reference Ulmer, Harris and SteffensmeierUlmer et al. 2012). According to this thesis, violence is buffered by chain migration into cohesive immigrant communities with strong family bonds and comparatively high levels of economic and social capital. The bulk of the research supporting this perspective, however, is based on one or a few traditional Latino migration destinations situated along the Mexican border. Whether emerging Latino settlement areas away from the border offer the same protections against violence remains an open question (Reference Harris and FeldmeyerHarris and Feldmeyer 2013).

The diffusion of Latino immigrants from traditional settlement areas to nontraditional destinations further inland has been described as “perhaps the most significant trend in U.S. population redistribution over the past quarter century” (Reference Lichter and JohnsonLichter and Johnson 2009: 497). Immigrant growth rates in the newer destinations in the Midwest and South are three to five times greater than in the traditional settlement areas of the Southwest (Reference Lichter and JohnsonLichter and Johnson 2009). Both push and pull factors have fueled growth in these newer settlement areas: economic stagnation in traditional settlement areas of the Southwest and greater economic opportunities in the newer ones (Reference Ludwig-Dehm and IcelandLudwig-Dehm and Iceland 2017). Another possible push factor is stricter enforcement of immigration laws and border patrol scrutiny in the traditional destinations (Reference Harris and FeldmeyerHarris and Feldmeyer 2013).

Traditional and emerging Latino migration destinations differ in ways that are potentially relevant to the study of crime trends. Education levels are higher and the degree of concentrated poverty among Latinos is lower in nontraditional areas (Reference Lichter and JohnsonLichter and Johnson 2009; Reference Ludwig-Dehm and IcelandLudwig-Dehm and Iceland 2017). These and other differences have prompted empirical examinations of violent crime in traditional and emerging Latino settlement areas directed at the question of whether the protective effects of strong community and family bonds found in the traditional locales extend to the newer migration destinations. The results of this research are mixed. Cross-sectional investigations generally find that protection against violent crime is greater in traditional settlement areas (e.g., Reference Ben, Sun, Harris and CullenFeldmeyer et al. 2022; Shihadeh and Barranco 2010; Reference Min and BaumerXie and Baumer 2018), whereas a recent longitudinal study finds a negative relationship or no relationship between immigration and Latino, non-Latino Black, and non-Latino White homicide rates in both traditional and newer settlement areas (Reference LightLight 2017).

In summary, extensive prior research has found that the same conditions of structural disadvantage affect both non-Latino Black and White homicide rates, although the magnitude of the effects may differ. An exception is racial segregation, which elevates Black homicide and reduces or is unrelated to White homicide. More limited research has disclosed significant effects of structural disadvantage and significant or null effects of immigration on Latino homicide. Poverty and family disruption tend to increase Latino homicide rates and immigration is unrelated to or reduces them. The terms “elevates,” “increase” and “reduces” are used advisedly because most of the relevant research is cross-sectional and does not reveal movement in homicide and other conditions over time. That said, the research literature tends to favor the idea that the common trends in race and ethnic homicide rates are related to common causes.

4.2 Male and Female Homicide Trends

A simple, if exaggerated, way of summarizing gender patterns in homicide is that men kill men, women kill men, and men kill women. There is no disputing that men do most of the killing and are killed the most in nearly every society for which we have records. But are changes over time in male and female homicide victimization rates similar or different?

Figure 6 displays male and female homicide trends from 1981 to 2020. Because the male rates are several times higher than the female rates, the two series are scaled on separate axes.

Figure 6 Male and female homicide rate, 1981–2020.

Source: National Vital Statistics System

The male and female trends closely track one another. They rise and fall together in a pattern that resembles the overall homicide trend shown in Figure 1. As with the race-ethnic trends it appears that many of the same underlying conditions may help to explain the parallel trajectories of male and female homicide. Recent research has identified some of the explanatory factors related to trends in homicide and nonfatal violence among males and females in the United States.

4.2.2 Intimate Partner Violence

Criminologists and other researchers have devoted substantial attention to trends in violence among intimate partners (i.e., married, separated, divorced, never married couples). The research results differ depending on whether the focus is on fatal or nonfatal violence.

Homicide

Studies of intimate partner homicide trends have uncovered two robust facts about this type of killing: (1) Women’s victimization rates are several times higher than men’s, and (2) men’s victimization rates have declined more than women’s over time (Reference Angela and WilliamsBrowne and Williams 1993; Reference Angela, Williams, Dutton and ZahnBrowne et al. 1999; Reference Fridel and Alan FoxFridel and Fox 2019; Reference Carole, Puzone, Linda, Kresnow, Thompson and MercyPuzone et al. 2000). The first fact is not surprising. Victims’ advocates have effectively defined domestic violence as violence against women.Footnote ¹³ The second fact, at first glance, is more puzzling. Given the widespread public attention to and growth in protective services and resources for female victims of intimate partner violence, how can we explain the greater decline in male victimization rates? Part of the answer appears to be that the increasing availability of assistance for female victims of intimate partner violence has helped to save men’s lives.

Researchers have relied on two key concepts to explain this seeming paradox in intimate partner homicide trends: victim precipitation and exposure reduction. A sizable number of homicides in general, and most intimate partner homicides committed by women, are precipitated by the victim (Reference Campbell, Glass, Sharps, Laughon and BloomCampbell et al. 2007; Reference WolfgangWolfgang 1958). Women kill in these situations to stop violence or abuse by their partners or ex-partners. It follows that interventions that reduce their exposure to violent intimates may obviate the need, put bluntly, to kill their way out of the relationship.

One such mechanism of exposure reduction is the liberalization of divorce laws. When access to divorce is made easier for women and “unilateral” (i.e., one spouse can initiate the divorce without the consent of the other), spousal homicide, especially against men, should decline. Consistent with this hypothesis, studies have linked decreasing rates of male victimization in spousal homicide to rising divorce rates (Reference Laura, Nagin and RosenfeldDugan et al. 1999; Reference RosenfeldRosenfeld 1997). Some research has found that unilateral divorce also reduces nonfatal spousal violence (Reference Stevenson and WolfersStevenson and Wolfers 2006).

Another means of exposure reduction is to provide assistance in the form of domestic violence hotlines, shelters, and legal aid to women, and often their children, to facilitate exiting violent or abusive relationships. Here again, the research is supportive. Studies have shown greater declines in male intimate partner homicide rates in cities and states where such domestic violence resources are more plentiful (Reference Angela and WilliamsBrowne and Williams 1989; Reference Laura, Nagin and RosenfeldDugan et al. 1999). In a subsequent study, Reference Laura, Nagin and RosenfeldDugan et al. (2003) found an exception that, in effect, proves the rule of exposure reduction. Not all kinds of domestic violence resources, they discovered, necessarily reduce intimate partner homicide. Some in fact may increase it “suggesting a retaliation effect when interventions stimulate increased aggression without adequately reducing exposure” (169).

Nonfatal Violence

Trends in nonfatal intimate partner violence are not as easily summarized or explained as the homicide trends. Figure 7, from Reference Lauritsen and LangLauritsen et al. (2022), displays trends in rates of nonfatal serious violence victimization (sexual assault and rape, aggravated assault, and robbery) among male and female intimate partners based on NCVS data from 1973 to 2020.Footnote ¹⁴ We see that, like the gender difference in levels of intimate partner homicide, rates of nonfatal intimate partner violence are much higher among females than males. The female and male rates both rose from the early 1970s through the early 1990s and then declined through the beginning of the current century. The gender gap in nonfatal intimate violence (the ratio of the female rate to the male rate) narrowed when the female and male rates were rising, because the increase in the male rates was greater than the increase in the female rates. It continued to narrow when both rates were falling, because the decrease in the female rates exceeded the decrease in the male rates. The female rates remained essentially flat during the first two decades of the current century, while the male rates decreased from 2014 to 2020, which widened the gender gap in nonfatal intimate violence, returning it to where it was in the late 1970s.

Figure 7 Gender gap in serious violence by intimate partners: 1973–2020.

Source: Reference Lauritsen and LangLauritsen et al. (2022)

This is a more complicated pattern than we saw in the intimate partner homicide trends. Ideally, both fatal and nonfatal intimate violence should be explained by the same theory. Feminist theories that highlight the role of patriarchy are able to explain gender differences in the levels of fatal and nonfatal intimate violent but do not as readily accommodate differences in the trends, especially in nonfatal violence. Incorporating movement over time into theories of intimate partner violence is a major task for those committed to explaining and reducing violence against both women and men.

6.1 Crime, Inflation, and Underground Markets

Persons engaged in income-producing crime must have a way of disposing of the goods they steal that they do not consume. They can give them away, sell them for cash, or exchange them for something else of value. Gifts aside, property offenders perforce must become involved in underground or “off-the-books” transactions because it is illegal to buy and sell stolen goods. Applying standard economic theory, the greater the demand for stolen goods, the higher the returns to acquisitive crime (Reference BeckerBecker 1968; Reference EhrlichEhrlich 1973). The key issue in understanding the influence of underground markets on acquisitive crime, then, concerns the conditions that shape the demand for stolen goods.

Research on illicit drug markets has considered their demand characteristics in some detail (Reference ReuterReuter 2010). Research on markets for stolen goods is more limited, especially with respect to how changes in the formal economy affect their expansion and contraction over time. Obtaining systematic time-series data on such markets is difficult, of course, because detailed records are rarely kept, but relevant insights can be drawn from observations of change in the retail sector of the formal economy. Generally speaking, when aggregate incomes fall or prices rise, consumers search for cheaper goods and services, a phenomenon economists refer to as “trading down.” For example, as prices increase, midlevel retail outlets typically lose customers to those selling the same or similar goods at discount prices, such as Walmart and other “big-box” discount outlets, “dollar” stores, and the retail shops operated by Goodwill Industries and the Salvation Army.

But where do those who had been shopping at Dollar General or Goodwill trade down when the economy sours? Although the evidence trail leaves off at the bottom rungs of the formal retail market, a reasonable inference from the foregoing is that some enter the underground economy, including the market for stolen goods, in search of lower prices.

The cardinal quality of stolen merchandise is that it is cheap. If it were not, it would attract few consumers away from legal markets that sell the same products with purity and quality guarantees, return and replacement policies, peaceful procedures for resolving disputes, and no risk of arrest.

Inflation is not the only condition that influences the demand for stolen goods, but its effects are likely to be stronger and more direct than those of unemployment, economic growth, and income, especially when wages fail to keep pace with prices. In contrast with unemployment, the crime-relevant effects of inflation are widespread, instantaneous, and direct. The official unemployment rate, which excludes discouraged workers and others who have dropped out of the labor force, is a narrow and incomplete measure of joblessness.

But even broader measures such as labor-force participation necessarily apply only to persons who are out of work (and their families) and do not capture the effects of changing economic conditions on those who remain employed, who constitute the large majority of the working-age population even under the worst economic circumstances. In addition, the full effects of job loss typically develop over a period of months, as savings are drawn down and unemployment insurance benefits are exhausted (Reference David and LandCantor and Land 1985). By contrast, although its effects are greatest for the poor, inflation touches all consumers.

In a national-level US study, Reference Rosenfeld and LevinRosenfeld and Levin (2016) found strong support for the effect of inflation on change over time in “acquisitive” crime, offenses committed for monetary gain (robbery and property offenses). The higher the inflation rate, the higher the acquisitive crime rate. As the inflation rate decreased, so did acquisitive crime. The inflation effects were more robust than the effects of economic growth and unemployment and they persisted over several years. Only two other variables rivaled the impact of inflation on acquisitive crime: consumer sentiment and the imprisonment rate. Other things equal, declining acquisitive crime rates were associated with increases in consumers’ optimism about the state of the economy and their own economic fortunes, and they were related to growth in the prison population.

Recent research has also revealed significant effects of inflation on trends in acquisitive crime and homicide in a sample of US cities during the past several decades (Reference Rosenfeld and VogelRosenfeld and Vogel 2023; Reference Rosenfeld, Vogel and McCuddyRosenfeld et al. 2019). The researchers found that the effect of inflation on homicide was partially mediated by acquisitive crime. In other words, inflation had a significant effect on acquisitive crime which, in turn, had a significant effect on homicide. The reasons why robbery and property crimes can influence the homicide rate were discussed in Section 2. But the effect of inflation on homicide was only partially mediated by acquisitive crime; a significant and sizable direct effect of inflation on homicide remained after taking account of its indirect effect through acquisitive crime. Reference Rosenfeld and VogelRosenfeld and Vogel (2023) speculated that inflation also affected homicide through its impact on institutional legitimacy.

6.2 Crime, Inflation, and Institutional Legitimacy

The criminologist Gary LaFree maintained that rates of homicide and other crimes will increase when social institutions lose their legitimacy (Reference LaFreeLaFree 1998). Strong institutions discourage crime by guiding behavior into socially approved channels. When institutional supports for normative behavior and controls over disapproved behavior weaken, crime rates go up. Institutions rely for their effective performance on public confidence. If confidence wanes, behavior becomes unmoored from its institutional guideposts. Reference LaFreeLaFree (1998) observed that public confidence in political and economic institutions plummeted during the 1960s and 1970s amid racial tensions and disorder in US cities, widespread protests against the Vietnam War, and the Watergate crisis that brought down a president.Footnote ¹⁷ As a consequence, crime rates climbed.

Another major culprit was rampant inflation which, according to Reference LaFreeLaFree (1998), undermined public confidence in both the economy and government during the 1960s and 1970s. Surveys conducted during the 1990s in the United States and Germany found that respondents in both countries believed that inflation is both a consequence and a cause of political instability and that it damages national cohesion and prestige, in addition to reducing living standards (Reference Shiller, Christina and DavidShiller 1997). The research on crime, inflation, and institutional legitimacy makes an important contribution to the theory of crime trends discussed in the following section.

In summary, inflation affects crime trends in at least two ways: by increasing the demand for and supply of cheap stolen goods and by destabilizing social institutions. The small, but growing, literature on crime trends and inflation (see Reference Rosenfeld and LevinRosenfeld and Levin, 2016, for a review) offers strong support for including inflation in future research on why crime rates change over time.

7.1 Institutional Analysis of Crime Trends

An exciting development in criminological theory over the past few decades has been the renewed attention devoted to the role of social institutions, a development that has been termed the “new institutionalism” in criminology (Reference Karstedt, McLaughlin and NewburnKarstedt 2010). This new institutionalism has been manifested most prominently in recent research on criminal punishment (e.g., Reference GarlandGarland 1990, Reference Garland2001). Institutional analysis also has an important role to play in the study of crime trends.

Three dimensions of social institutions are particularly important when analyzing the implications of the institutional order for change over time in crime rates: institutional structure, institutional regulation, and institutional performance (see Reference Messner, Rosenfeld, Karstedt, Francis and WilcoxMessner et al. 2011). Institutional structure encompasses the content of the institutional rules and their internal consistency. An example is the emphasis on individual economic achievement under conditions of open, competitive mobility inherent in the value complex known as the “American Dream.” A competitive, profit-oriented capitalist economy channels social action in accordance with this value complex (Reference Messner and RosenfeldMessner and Rosenfeld 2013).

Institutional regulation refers to the basis of compliance with the rules. Not all action in conformity with institutional norms is “institutionalized” in the formal sense of the term. Individuals may align their behavior with the rules of the game based on strictly utilitarian calculations of self-interest or in response to coercive pressures by the more powerful. The distinctive feature of institutionalized social action is that it is governed by a sense of obligation; individuals act in accordance with the rules because they believe it is the right thing to do. When institutional regulation is strong, the rules of the game are granted considerable legitimacy.

The procedural justice literature in criminology provides a good example of institutional regulation (Reference TylerTyler 1990; Reference Tyler, Atiba Goff and MacCounTyler et al. 2015). When individuals perceive their treatment by the police to be fair, impartial, respectful, and just, they are more likely to willingly comply with the law. When people believe treatment is unjust, compliance is less likely or occurs under conditions of duress. Confidence in the police fell among Black Americans after two incidents of police violence that were widely perceived to be unjust: the killings of Michael Brown in Ferguson, Missouri, in 2014 and George Floyd in Minneapolis in 2020 (see Figure 10).Footnote ¹⁸ Confidence in the police among Black Americans rose in 2021, but not to the level that prevailed prior to the Ferguson killing. Meanwhile, confidence in the police among White Americans dropped only slightly after these incidents. Homicide rates increased in both the Black and White population in 2015 and 2020 (see Figure 5), but any causal connection between the decrease in confidence and the increase in homicide remains uncertain.Footnote ¹⁹

Figure 10 Percentage of White and Black adults with a great deal or quite a lot of confidence in the police, 1993–2021.

Source: Gallup Poll

It is also apparent from Figure 10 that confidence in the police among Black Americans has been much lower than in the White population for at least the past three decades and undoubtedly longer. Lack of confidence in the police is related to a pervasive and long-standing “legal cynicism” in highly disadvantaged Black communities (Reference Hagan, McCarthy and HerdaHagan et al. 2020; Reference Kirk, Papachristos, Cullen, Wilcox, Robert and BrendanKirk and Papachristos 2015; Reference Sampson and Jeglum BartuschSampson and Bartusch 1998). According to this research, legal cynicism and lack of confidence in the police are not an outgrowth of the “culture” of these communities, the values and beliefs that constitute their institutional structure in the terms of the current discussion. Rather, they reflect institutional deregulation, a loss of the fervor and force of the cultural rules such that compliance, if it occurs at all, is the product of strictly utilitarian considerations or has to be compelled by force.

Institutional performance concerns the extent to which commitment to and faithful enactment of institutional roles result in the expected institutional outcomes. Performance suffers when institutions fail to “deliver the goods.” Generally speaking, institutional performance is most relevant for understanding short-run changes in crime rates, as when an increase in inflation or consumer pessimism results in an increase in crime (Reference Rosenfeld and LevinRosenfeld and Levin 2016). But if disruptions in institutional performance persist over many years or decades, they can weaken institutional regulation by reducing confidence in existing institutional arrangements and, on occasion, fomenting a legitimacy crisis, as was the case during the 1960s and 1970s in the United States (Reference LaFreeLaFree 1998). Reference RothRoth (2009) presents several examples in American history of crime increases associated with declining confidence in the federal government. Finally, severe institutional crises can lead to fundamental alterations in the institutional order and ignite crime increases during the transition to a new set of institutional arrangements, as occurred after the fall of the Soviet Union (Reference Pridemore and KimPridemore and Kim 2006).

Institutional structure, regulation, and performance are distal sources of change in crime rates. Further theoretical development will require specification of more proximate conditions. Reference Baumer, Velez and RosenfeldBaumer et al. (2018) offered a “blueprint” that outlines several more proximate sources of change in crime rates and that can be situated in the institutional framework outlined here. Their blueprint also would bring a theory of crime trends to the center of mainstream criminological theory.

Reference Baumer, Velez and RosenfeldBaumer et al. (2018) proposed that crime trends can be understood on the basis of three related concepts: social controls, individual criminal propensities and motivations, and criminal opportunities. Each concept is rooted in one or another mainstream criminological perspective. Social control theory holds that delinquent and criminal behavior results when individuals’ attachment and commitment to, and involvement and belief in, conventional groups and institutions are weakened or absent (Reference HirschiHirschi 1969; Reference KornhauserKornhauser 1978). Anomie and strain theories locate the motivations for crime in blocked goals and negative emotions (Reference Adler and Robert.Agnew 2005; Reference MertonMerton 1938). Self-control theory stipulates that criminal propensities are a product of impulsivity, weak social attachments, and poor parenting (Reference Gottfredson and HirschiGottfredson and Hirschi 1990). Rational choice and routine activity theories explain criminal activity in terms of rational cost-benefit calculations by motivated offenders in “criminogenic settings” where suitable victims are present and capable guardians are absent (Reference Cohen and FelsonCohen and Felson 1979; Reference Cornish and ClarkeCornish and Clarke 1986).

Centering crime trends theory in mainstream criminological theories helps to reveal the institutional moorings of each of the three proximate sources of change in crime rates. Social control, both formal and informal, is a basic function of institutions and is the primary means by which behavior is channeled into normative pathways. Families exert normative controls through the socialization process. The free market economy exerts remunerative controls through the wage and profit system, although this type of control is prone to institutional deregulation by fostering compliance based on self-serving utilitarian grounds (Reference Messner and RosenfeldMessner and Rosenfeld 2013). The criminal justice system exerts normative or coercive controls, depending on the legitimacy accorded to its procedures.

Anomie theories link the motivation to commit or refrain from crime both to the structure of social institutions and to institutional performance. Reference Messner and RosenfeldMessner and Rosenfeld (2013) maintained that commitment to the values inherent in the American Dream promotes crime when institutional controls are weakened. Reference MertonMerton (1938) famously proposed that criminal adaptations can result when adherence to the dominant rules of the game does not lead to expected rewards. Reference Cohen and FelsonCohen and Felson (1979) argued that the institutional changes that emptied homes by bringing women into the paid labor force altered the opportunity structure for crime and led to increases in residential burglaries.

Other changes in criminal opportunities, however, are less closely tied to the structure, regulation, or performance of social institutions. Modifications to the physical environment and consumer products, such as shielding shopkeepers behind heavy glass and installing electronic locking and ignition systems in motor vehicles, can reduce opportunities for crime in the absence of institutional change. That said, a broad-scale “safety consciousness” has taken off in the United States and elsewhere in recent decades, propelled in no small measure by the public health profession. Safety consciousness involves inspecting every aspect of the physical and social environment for potential hazards and eliminating them through environmental or product redesign – what criminologists term “target hardening.” An example is the secure storage of firearms so that children cannot gain access to them (Reference Tamera, McGee, Johnson and BordleyCoyne-Beasley et al. 2002). To the extent that the reduction of criminal opportunities is part of this broader shift in institutional structure, it too belongs in a theory of crime trends and the institutional order.

The “new institutionalism” in criminology serves as a sturdy basis for the development of a macrolevel theory of crime trends. Crime rates move up and down over time as a consequence of changes in the institutional order of societies – most of the time. Sometimes, however, crime rates undergo sharp and abrupt shifts for reasons that cannot be anticipated by even the most comprehensive theory or the most reliable research methods. These changes are produced by “exogenous shocks” to the social order.

9.1 Background

A literature search produced just eight macrolevel studies that include forecasts of crime rates in the United States and none that were published in the last fifteen years.Footnote ²⁵ The oldest study was published almost fifty years ago (Reference FoxFox 1978), and the most recent studies appeared in 2008 (Reference Baumer, Arthur and RosenfeldBaumer 2008; Reference Pepper, Arthur and RosenfeldPepper 2008). The strong focus of most of the studies is on how projected changes in the age composition of the population are expected to influence future crime rates, although some investigations incorporate additional factors, such as unemployment, imprisonment, and policing (Reference Cohen and LandCohen and Land 1987; Reference Cohen, Felson and LandCohen et al. 1980; Reference FoxFox 1978). A study by Reference Land and McCallLand and McCall (2001) is particularly noteworthy because it devotes explicit and sustained attention to the assumptions and challenges of forecasting crime rates.

Reference Land and McCallLand and McCall (2001) pointed out that all crime forecasts are prone to uncertainty owing to the simple fact that the future is unknowable until it arrives. Crime forecasters, therefore, should avoid “point forecasts” that predict a single future crime rate. Instead, researchers should employ a range of forecasts or “forecast cones” bounded by upper and lower limits that can be estimated in different ways and usually involve expert judgment regarding the future conditions likely to affect crime rates. They also cautioned against “continuity bias” in crime forecasts that are “heavily influenced by recent trends in crime rates just prior to the period for which the forecasts are made” (344). They recommended that forecast periods should be as short as possible, no more than a few years ahead, to reduce forecasting error and in recognition of the abbreviated budget and planning horizons of most policymakers, especially at the local level. Their final recommendation is to update forecasts as often as possible as new data become available (345).

Reference Land and McCallLand and McCall (2001) engaged in a forecasting exercise based on these suggestions. They forecasted the number of 14- to 17-year-old Black male homicide offenders and the number of same-age White male homicide offenders in the United States for each year from 1998 to 2007. The forecasts were based on the homicide offending rates of the two groups of adolescents from 1980 to 1987 and Census Bureau projections of the size of each group through the forecast period. They set the lower bound of forecasting cones for each group at 25% of their lowest homicide offending rate between 1980 and 1987 and the upper bound at 125% of their highest 1980–1987 homicide offending rate. The midrange forecast assumed their average 1980–1987 offending rate would remain constant throughout the forecast period.

The resulting forecasts cover a very wide range of estimated outcomes. For example, by 2002, midway through the forecast period, the estimated number of White homicide offenders ranged from fewer than 100 at the lower limit of the forecast cone to about 3,800 at the upper limit. The midlevel estimate was about 500 (341, Figure B). Reference Land and McCallLand and McCall (2001) did not offer strong reasons why the midlevel forecast should be preferred over the upper and lower limits of the forecast cone; the key point of the exercise is to avoid single-point forecasting. Their forecast cones for adolescent male homicide offenders are likely to be very accurate but so imprecise they would have little practical utility. The solution is not to favor point forecasts over a range of estimates, but to create forecast cones with boundaries that are narrow enough for policy purposes but wide enough to produce accurate forecasts. Useful and reliable forecasting, in other words, always involves a tradeoff between precision and accuracy.

This section presents a forecasting exercise using time series data on violent and property crime rates in New York City. The main point of the exercise is not to correctly predict New York crime rates, although the forecasts are evaluated for their accuracy and precision. Rather, the primary purpose of the exercise is to explore the feasibility of and challenges to crime forecasting. The section concludes that, while the challenges are considerable, the benefits of crime forecasting, for both researchers and policymakers, outweigh them. It is time for a revival of crime forecasting in criminology.

9.2 Data and Methods

The forecasting exercise is carried out with Reference Land and McCallLand’s and McCall’s (2001) thoughtful discussion of the dos and don’ts of crime forecasting in mind. As they point out, because criminal justice policymaking is largely a local matter in the United States, crime forecasting is better done at the local than the national level. This exercise, therefore, is based on data for the city of New York. The outcomes are New York’s violent crime rate (murder, rape, robbery, and aggravated assault) and property crime rate (burglary, larceny, and motor vehicle theft). The sample data span the period 1980 to 2016.

Two out-of-sample forecast periods are examined. The first is the period between 2017 and 2021. This five-year out-of-sample period, during which New York’s violent and property crime rates are known, is used to validate the forecasts derived from a model based on the 1980–2016 data. The violent and property crime rates for 2022 to 2024 are then forecasted. The crime rates for this period were unknown when these analyses were carried out. The forecasting exercise is summarized in the text, and technical details can be found in Appendix II.

9.3 Results

Descriptive statistics for the variables (in original metric) used in this analysis are shown in Table 1. The violent and property crime rates exhibit substantial variation over the 1980–2016 observation period. Both trended downward over time. The highest violent crime rate during the period, 2383.6 violent crimes per 100,000 population in 1990, was more than four times greater than the lowest rate, 568.2 per 100,000 in 2009. The highest property crime rate, 8007.3 per 100,000 in 1981, was over five times greater than the lowest rate, 1467.4 per 100,000 in 2016. The explanatory variables exhibit comparable or even greater variation during the observation period.

New York’s violent and property crime rates between 1980 and 2016 are displayed in Figure 11. To eliminate scale differences and reveal the relationship between the trends in the two crime types, they are scaled on separate axes, with violent crime on the left-hand axis and property crime on the right-hand axis of the figure. Both violent and property crime rates fell during the early 1980s and rose during the next few years, before falling continuously well into the current century during what has been termed the “Great American Crime Decline” (Reference ZimringZimring 2007). What is also striking about the violent and property crime time trends is how closely they correspond (the Pearson’s correlation (r) between the two series is .98). This high degree of convergence suggests that the same sources of change over time in New York’s violent crime rate also underlie the change in the property crime rate.

Figure 11 New York violent and property crime rates, 1980–2016.

Source: New York State Division of Criminal Justice Services; Uniform Crime Reports

Figure 11 provides clear visual confirmation that New York’s violent and property crime rates between 1980 and 2016 are nonstationary; the mean rates decrease over time. The forecast models were therefore fit to the first-differenced violent and property crime rates between 1980 and 2016. The years 2017 to 2021 were “held back” from the models so they could be used to validate the forecasts from the 1980–2015 baseline period. The closer the forecasted crime rates are to the observed rates during the validation period, the greater our confidence in the forecasts for 2022 to 2024 when the crime rates are unknown.

The forecasting results are presented in Figures 12 and 13 (and Table A-1 in Appendix II). The figures display the observed and forecasted values of the crime series from 2000 to 2024. The solid line denotes the observed values from 2000 to 2021. The dashed line denotes the forecasts for 2000 to 2016, the dotted line indicates the forecasts for 2017 to 2021, the validation period, and the gray-shaded dotted line represents the forecasts for 2022 to 2024. In 2017, the first year of the forecast validation period, the New York’s observed violent crime rate dropped by 37 violent crimes per 100,000 population, while the forecasted rate shows little change. The changes in the observed and forecasted violent crime rates are fairly close during the next few years until 2021, when the observed rate increased by more than 80 violent crimes per 100,000, and the forecasted rate drops by about 25 violent crimes per 100,000. The forecasts indicate that New York’s violent crime rate should rise in 2022 and 2023. The rate in 2024 should exhibit little change over the previous year.

Figure 12 Observed and forecasted year-over-year change in New York violent crime rate, 2000–2024.

Source for observed data: New York State Division of Criminal Justice Services; Uniform Crime Reports

Figure 13 Observed and forecasted year-over-year change New York property crime rate, 2000–2024.

Source for observed data: New York State Division of Criminal Justice Services; Uniform Crime Reports

As shown in Figure 13, New York’s observed and forecasted first-differenced property crime rates are very similar between 2017 and 2021. In 2017, they diverge by 66 property crimes per 100,000, by 34 in 2018 and 2019, and by just 19 in 2020. As with violent crime, the difference between the changes in the observed and forecasted property crime rates is larger in 2021, when the observed rate grew by nearly 100 property crimes per 100,000, and the forecasted rate falls slightly. The forecasts suggest that New York’s property crime rate should rise in 2022 and 2023 before falling in 2024.

Contrary to Reference Land and McCallLand’s and McCall’s (2001) advice, this exercise yields point forecasts of New York’s crime rates and not the kind of forecast cones they prefer. Nonetheless, it is important to create boundaries around the point forecasts to evaluate their utility for both policymaking and theory testing. This means that the policymaker or researcher will have to decide how much forecast error is tolerable, which is a substantive and not a statistical decision. We will assume for current purposes that forecasted crime rates that diverge from the observed rates by no more than 10% are sufficiently accurate and precise for both policy and theory evaluation. Forecasts that fall outside of these limits would be uninformative, although they do suggest that the forecast model probably needs to be revised.

The forecast errors for violent crime fall outside the 10% limits in 2021 and are inside the limits for the other years of the validation period (see Table A-1 in Appendix II). These results suggest that the forecast model for violent crime should be revisited to determine the source of the forecast error in 2021. One likely candidate is the exogenous shock to crime rates from the Covid-19 pandemic, which no forecast model could have predicted. In the meantime, a conservative approach would be to rely on the five-year mean forecasts instead of the yearly estimates for forecasting purposes. This would reduce the average forecasting errors during the validation period, but it would not capture the forecasted rise in violent crime rates from 2022 to 2024 estimated from the ARIMA model. A fair conclusion is that the forecasting results for violent crime are acceptably precise on average and that the large error during 2021 was unavoidable.

The story for property crime is somewhat different. The forecasted property crime rates during the validation period are well within the 10% tolerance limits. These limits were arbitrarily drawn, but even if the 10% divergence threshold were cut in half, the forecasted property crime rates would have been accurate and precise enough for both policy and research purposes between 2017 and 2020 and nearly so in 2021.

The forecasting results for New York’s violent and property crime rates inspire guarded optimism about the prospects for forecasting in criminology. The forecasted property crime rates are very close to the observed rates during the validation period and provide a reliable basis for forecasting property crime rates three years ahead. The same is generally true for the violent crime results. The forecast error for violent crime in 2021, however, is outside the 10% tolerance limits placed around the estimates. More permissive limits could have been chosen, of course, but the optimal standard – how much error is acceptable – is not a statistical question but a matter of judgment and the preferred tradeoff between forecast accuracy and precision. The forecast error is also an important reminder of the challenge of unanticipated shocks to forecasting crime rates, or anything else.

9.4 Conclusion

Forecasting future crime rates, when done carefully on the basis of a credible forecasting model, is a natural and needed extension of the study of crime trends. Forecasting provides data to test theory that were not in the sample of observations used to develop the theory. Forecasting answers the perennial plea of policymakers, the press, and the public: You’ve told me what happened yesterday; now tell me what will happen tomorrow. The answers will not always be accurate or precise, but they will come from an explicable set of methods and decisions that assume that the probable future of crime rates is related to their past behavior and to expected changes in the conditions known to influence it.

Modesty is the best policy when forecasting crime rates. Forecasts will almost always be off the mark in the presence of exogenous shocks that sever the future from the past and that no forecasting model could have predicted. A recent example is the Covid-19 pandemic, which affected crime rates in complex ways (Reference Lopez and RosenfeldLopez and Rosenfeld 2021). Forecasts will also be incorrect when the conditions known to influence the crime rate change in unexpected ways, such as the abrupt increase in inflation beginning in 2021 (Reference Banerjee, Hall, Kouretas and TavlasBanerjee et al. 2023). That is why forecasts should always be tested against out-of-sample conditions that are known, the approach taken here, before taking on the unknowable future. A useful way of presenting the forecasts is in the form of conditional probabilities: If these conditions hold, then the crime rate will be X. When uncertainty exists regarding the future state of crime-generating conditions, as it usually does, a range of probabilities can be estimated. For example, all else equal, if the inflation rate is 3%, then the crime rate will be Y. If the inflation rate increases to 5% or 7%, the crime rate will be Z. These conditional probabilities offer another way to construct forecast cones around the point forecasts.

The forecasting exercise conducted here is intended to revive interest in crime forecasting in criminology. I have tried to be explicit about the reasoning behind each of the steps taken to (1) ready the data for reliable forecasting, (2) specify an explanatory model to be used in multivariate forecasting, (3) choose a forecasting model, and (4) interpret the results. Each of these decisions is open to criticism and alternative approaches. While I believe the data, methods, and models used here are useful for reviving discussion and analysis of crime forecasting, they can be augmented in several ways, such as by increasing sample size and incorporating additional covariates to reduce sampling error.Footnote ²⁷ In addition, other types of forecasting models (e.g., exponential smoothing models, which give greater weight to more recent observations) should be applied to the study of crime trends and their performance should be assessed against the ARIMA methods used here.

Crime forecasting has fallen on hard times in criminology, but past mistakes should not prevent renewed attention to this important endeavor. There is much work to do, both technical and political. This study has barely scratched the surface of the rich forecasting literature in other fields.Footnote ²⁸ Preparing the policy environment to make optimal use of crime forecasting will take years. Meanwhile, there are grounds for optimism. Enough is now known about the behavior of crime rates to support reliable short-run forecasts of the future. And testing our theories against future crime rates is the best way to improve our explanations of the past.

10.1 Summary of Crime Trends and Explanations

10.2 Theory

There is no comprehensive macrolevel theory of change in crime rates. Mainstream criminological theories such as the anomie-strain, control, cultural, and routine activity perspectives are certainly pertinent, but mainly as second-order suppliers of ideas and empirical findings. They do not constitute the core of a theory of change, either separately or in combination. New theoretical ground will have to be broken.

Reference Baumer, Velez and RosenfeldBaumer et al. (2018) offer valuable ideas for consolidating existing criminological theories in a conceptual framework that would facilitate the development of a theory of crime trends and, in the process, move the study of crime trends closer to mainstream criminology. As indicated earlier, in my view this framework should be informed by the central propositions of the new institutionalism in criminology:

• Crime is a social fact to be explained by other social facts.

• Crime is a normal feature of the institutional order.

• The form and frequency of crime will change with the regular rhythms of institutional performance, the sway of institutional regulation, and transformations in institutional structure.

Institutional-anomie theory (Reference Messner and RosenfeldMessner and Rosenfeld 2013) comes close to the kind of institutional approach that is needed, but it was developed primarily to explain differences across societies in crime rates and not trends in crime within societies.

10.3 Data

Theory development and testing will also require improvements to existing data and methods. The FBI’s Uniform Crime Reports constitute the main data source in contemporary crime trends research. Other key sources are the Bureau of Justice Statistics’ National Crime Victimization Survey and homicide data from the FBI’s Supplementary Homicide Reports and the National Vital Statistics System. A significant gap in each of these systems is coverage of crime changes in real time. The UCR and NIBRS data are disseminated nine to ten months after the collection year, and the victimization and vital statistics data are even less timely. For some studies and when crime rates are changing relatively slowly, a lack of timeliness in the release of crime data is not a problem. But when crime levels appear to be changing very rapidly, such as during the recent homicide spikes in the United States, up-to-date data are clearly needed to dispel misconceptions (e.g., homicide rates are higher than ever) and anchor public debate in evidence rather than anecdote or groundless innuendo. There is no technical reason why the FBI crime data cannot be disseminated a month or two after collection, as was done in the quarterly crime reports issued during the 1930s and again beginning in 2020 amid concerns of a major crime rise. Private entities such as the Council on Criminal Justice (Reference Rosenfeld and LopezRosenfeld and Lopez 2022) and AH Datalytics (www.ahdatalytics.com/about-us/) should not have to fill the data void left by the nation’s major governmental sources of crime data.

Reference Baumer, Velez and RosenfeldBaumer et al. (2018) offer several useful recommendations for strengthening the nation’s crime data infrastructure, in addition to speeding up the dissemination of the data. They propose that quarterly or semiannual surveys compile data on crime, the criminal justice system, and demographic and socioeconomic conditions from representative samples jurisdictions across the country. The data should be sufficiently granular to track crime trends across areas within cities and should be supplemented by qualitative studies of communities that differ in crime trends and “criminogenic events” such as drug markets and plant closings. Moreover, Lauritsen (2023) recommends establishing a new crime data infrastructure that includes offenses beyond those currently reported to and recorded by local law enforcement agencies. These are tall orders and the multifaceted system would take years and substantial funding to implement. And if recent history is any indication, as discussed in Section 2, the road to a new system for compiling crime data could be quite bumpy.

10.4 Methods

Methodological advances are needed to better capture the nature of change in crime rates. New methods and ways of thinking about crime changes have emerged in microspatial research that can be adopted in macrolevel studies (see Reference Hipp and Iris LuoHipp and Luo 2022). The microspatial study of crime has important, if underexplored, implications for crime trends theory and research. Crime is heavily concentrated in a small number of urban and, according to some studies, suburban spaces (Reference Gill, Wooditch and WeisburdGill et al. 2017; Reference David, Groff and YangWeisburd et al. 2012). Reference WeisburdWeisburd (2015) has argued that crime concentration is so pronounced, pervasive, and consistent across jurisdictions and over time it qualifies as a criminological “law.” Strong research support exists for the microspatial concentration of crime (Reference Braga, Andresen and LawtonBraga et al. 2017). Where disagreement exists, it mainly involves questions about the best ways to measure crime concentration, the most appropriate microspatial unit of analysis, and the stability of crime concentration over time.

The macrolevel study of crime trends has much to learn from the microspatial study of crime concentration. The most important lesson is to decompose the units of analysis. National-level crime trends consist of the trends in subnational units such as cities and states, and the crime rates of different places do not necessarily travel at the same speed. Some places may have a disproportionate influence on the overall trends, and a critically important research task is to identify the sources of spatial variability in the movement of crime rates over time.

Group-based trajectory methods, originally developed to analyze individual developmental differences in antisocial behavior, have been applied in microspatial crime research and hold great promise for advancing the macrolevel study of crime trends (Reference NaginNagin 2005; Reference Weisburd, Bushway, Lum and YangWeisburd et al. 2004). These methods identify distinct clusters of persons or places on the basis of the trajectory over time of some outcome (e.g., antisocial behavior, crime rate). One question to which these methods can be productively applied is whether city crime rates follow a common national trend. Reference McDowall and LoftinMcDowall and Lofin (2009) found this to be true in the United States. Reference Baumer and WolffBaumer and Wolfe (2014) essentially confirmed this result, but their study uncovered somewhat greater heterogeneity in the movement of city crime rates over time and across different crime types. Group-based trajectory methods can be used to identify coherent clusters of cities that differ in their trends and in the demographic, socioeconomic, and policy attributes that underlie their distinct trajectories.

Finally, the trajectory methods can also enrich crime forecasting by identifying subtypes of places for which forecasts are more or less accurate and precise. It bears repeating that forecasting crime rates is a logical and, in my view, necessary next step in the macrolevel study of crime trends. Crime forecasting and trajectory analysis, of course, will always have to contend with exogenous shocks that can upset predictions from the most reliable models. In this respect, the criminologist is in a situation similar to British Prime Minister Harold Macmillan’s who, when asked by a journalist what was the greatest challenge for a statesman, is reputed to have replied: “Events, dear boy, events.”Footnote ²⁹ With credible forecasting methods and models in place, however, and mindful of the uncertainties thrown up by unforeseen events, the study of crime trends can secure its future.