Arrest rates by race reflect crime rates by race

Last Updated on May 19, 2024

There is no dispute that there are large racial disparities in arrest rates in the United States, particularly between blacks and whites. The most natural inference to draw from this data is that these racial disparities in arrest rates correspond to racial disparities in crime rates. However, some argue that this natural inference is either incorrect or unjustified. They argue that racial differences in arrest rates may not reflect racial differences in crime rates. For example, perhaps arrests are biased due to biased police practices or perhaps the arrest data reported by law enforcement agencies is biased.

In this post, I will provide data showing that the natural inference is in fact correct and justified. There is ample independent data showing that FBI-reported arrest rates by race are accurate reflections of true crime rates by race. The main focus will be on crime and arrest differences between blacks and whites, but some of the sources include data on other racial groups as well.

Two Hypotheses

According to the 2019 arrest statistics [archived] published by the FBI’s Crime in the United States, the latest year of this publication, the black percentage of arrestees greatly exceeds the black percentage of the general population. For example, despite being roughly 14% of the U.S. population (Census), black people account for 27% of total arrestees, 30% of arrestees for property crime, and 36% of arrestees for violent crime. The percentage of arrestees by race for all violent crimes and property crimes are as follows:

There is not any dispute about the existence of this data. However, there is some dispute about how these reported racial differences in arrests relate to racial differences in crimes. Specifically, there are two competing broad hypotheses to explain relationship between the FBI-reported arrest rates by race and the true crime rates by race:

  • The Parity Hypothesis: FBI-reported arrest rates by race accurately reflect true crime rates by race.
  • The Bias Hypothesis: FBI-reported arrest rates by race do not accurately reflect true crime rates by race. Specifically, racial differences in FBI-reported arrest rates are greater than the true racial differences in crime rates. In other words, FBI-reported arrest rates over-estimate racial differences in crime rates.

(technically, there is also a third hypothesis which posits that FBI-reported arrest rates under-estimate racial differences in crime rates. However, I’ll ignore this hypothesis for the sake of this post. Among those who are skeptical of the FBI arrest data, they usually are worried that the data over-estimates true racial differences in crime. If I come across any findings suggesting that FBI arrest data under-estimates racial differences in crime, then I will consider such findings as evidence for the Parity Hypothesis over the Bias Hypothesis).

The Bias Hypothesis can posit a number of explanations for why racial differences in FBI-reported arrest rates are greater than the true racial differences in crime rates. These explanations can fall into 2 broad categories:

  1. FBI-reported arrest rates by race do not reflect true arrest rates by race due to biases in the FBI data. For example, law enforcement agencies might report biased arrest data to the FBI, or there may be biases in which law enforcement agencies decide to report data.
  2. FBI-reported arrest rates by race do reflect true arrest rates by race, but arrest rates by race do not reflect true crime rates by race. For example, black offenders may be more likely to be arrested than black offenders, or innocent blacks may be more likely to be unjustly arrested than innocent whites.

In this post, I argue that the Parity Hypothesis is true. That is, the evidence suggests that FBI-reported arrest rates by race accurately reflect true crime rates by race. For example, I will argue that, just as blacks account for 50% of FBI-reported arrests for murder, blacks actually account for 50% (or more) of murder offenders. The same is true for violent crimes generally. When comparing these two hypotheses, my primary focus will be crime and arrest rates for blacks and whites, however other groups will sometimes be considered when the relevant data is available.

In order to test the Parity Hypothesis vs the Bias Hypothesis, we need some independent sources that proxy the true crime rates by race. In this post, these sources will include data on homicide victimization rates, assault victimization rates, criminal victimization surveys, and police reports. As I shall show below, the findings from each of these sources provides evidence for the Parity Hypothesis over the Bias Hypothesis. Each of these findings are predicted by the Parity Hypothesis without the need for any ad-hoc assumptions. That is, the findings from each of these sources are exactly what we would expect if the Parity Hypothesis was true.

It is important to note what this post is not concerned with. My post is not concerned with whether race is associated with arrests after controlling for other confounding variables. In other words, my post is not concerned with whether race per se is independently associated with arrest rates (however, I do briefly address this concern in a section on arrest bias near the end of this post). The Parity Hypothesis and Bias Hypothesis only make claims about whether arrest rates by race reflect true crime rates by race. Neither of these hypotheses make claims about whether race per se is independently associated with arrests. For example, a defender of the Bias Hypothesis might think that being black per se raises one’s odds of being arrested, or such a defender might think that black people are more likely to be poor and being poor (not being black per se) are more likely to be arrested.

Homicide victimization rates


The FBI arrest data shows that blacks constitute about 51% of arrestees for murder. To test whether the homicide arrest rates by race reflect true homicide rates by race, we can examine rates of homicide victimization by race.

One might worry that criminal offending rates by race don’t reflect criminal victimization by race. However, this worry is unlikely to be a problem when specifically examining homicide, because the vast majority of homicides are intra-racial rather than inter-racial. For example, see the following data [archived] published by the FBI showing the race of homicide victims and offenders in 2019:

As you can see, this data indicates that the vast majority (89%) of homicides against black victims were committed by black offenders (2,574/2,906). If we exclude cases where the offender race is unknown, then the percentage increases to 90%. Now, this table does not include data on all homicides, since it only includes data on cases when some information about the offender (e.g., age, race, sex, etc.) is known. However, there’s no reason to believe that the complete data would indicate higher rates of non-black homicide offenders against black victims. We have no reason to believe that rates of interracial homicide are much greater among cases where the homicide offender is unknown. Therefore, I will use black homicide victimization rate as a proxy for black homicide commission rate.

Likewise, the data also suggests that most homicides (60%) against Hispanic victims were committed by Hispanic offenders (595/987). However, there was a much larger percentage of cases where the ethnicity of the offender was unknown. If we exclude cases where the offender ethnicity is unknown, then 71% of homicides against Hispanic victims were committed by Hispanic offenders (595/836). This suggests that Hispanic homicide victimization rates can be used as a proxy for Hispanic homicide commission rate, although it will be a less reliable proxy for Hispanics compared to blacks.

In the rest of this section, I will document racial disparities in homicide victimization. There are two main sources [archived] of data to measure homicide victimization rates by race. The first source comes from the FBI. The second source comes from the CDC.

FBI reported deaths

In a report conducted by Cooper and Smith (2011) [archived] published by the Bureau of Justice Statistics, the black share of homicide victims has reflected the black share of homicide offenders from 1980 to 2008. Over this time period, blacks constituted about 47% of homicide victims and about 53% of homicide offenders (Table 1):

More recent data shows similar patterns. In 2019, the latest year for the Crime in the United States publications, the FBI reports [archived] that blacks constituted about 54% of homicide victims (7,484/13,927), which is only slightly greater than the black share of arrestees for murder in 2019 (51%, see above):

This also suggests that Hispanics account for about 18.1% of homicide victims when excluding victims of unknown Hispanic origin (2193/(13927-1808)). This is a bit lower than the FBI-reported Hispanic share of homicide arrestees (20.7%, see table 43).

The latest year of data for Crime in the United States reports was 2019 (see here for list of all publications). More recent FBI crime data is published in the Crime Data Explorer (CDE) tool. This tool shows that blacks constituted about 60% of both known offenders and known victims of homicides in 2022:

CDC reported deaths

The next source of information on homicide death rates by race comes from the CDC. Every year, the National Center for Health Statistics (NCHS) processes information from all death certificates filed in each of the 50 states and the District of Columbia. The CDC publishes reports using this data to present detailed information on deaths and death rates according to various demographic and medical characteristics. The most recent report by Xu et al. (2021) [archived] reports on deaths in 2019. The relevant information for this post is Table 10, which reports age-adjusted death rates by race, sex, Hispanic origin, and cause of death.

The table shows that the homicide death rate for males (9.6 per 100k) is 4.0 times the rate for females (2.4 per 100k). The homicide death rate for blacks (23.7 per 100k) is about 8.8 times the rate for whites (2.7 per 100k), and about 15.8 times the rate for Asians (1.5 per 100k). Hispanics also have relatively large homicide death rates, however the Hispanic homicide rate is still much lower than the black homicide rate. The homicide death rate for Hispanics (5.0 per 100k) is about 1.9 times as large as the rate for whites, and about 3.3 times the rate for Asians.

Table 8 also shows the total number of deaths due to homicide by race/ethnicity:

This data shows that non-Hispanic blacks accounted for about 51.9% (9,951/19,141) of homicide victims in 2019. This nearly perfectly matches the FBI-reported black percentage of murder arrestees in 2019 (51.2%).

The data for Hispanic victims is also moderately corroborated by the CDC and FBI data. The CDC data shows that 16.3% (3,122/19,141) of homicide victims were Hispanic in 2019. This is similar to the FBI-reported Hispanic percentage of murder victims (18.1%). Now, the Hispanic percentage of homicide arrestees (20.7%) is slightly greater than the Hispanic percentage of homicide victims from both datasets, but this discrepancy does not distort the overall findings on crime rates by race.

Conclusion

Overall, the FBI-reported homicide arrest rates by race is corroborated by the homicide victimization rates reported by both the FBI and the CDC. All three sources of data reinforce the hypothesis that blacks constitute about 50% of homicide offenders and victims in the United States. Likewise, Hispanics appear to constitute about 16% to 20% of homicide offenders and victims.

Assault injury rates


The FBI arrest data shows that blacks constitute about 33.2% of arrestees for aggravated assault in 2019. By contrast, whites (including Hispanics) accounted for 61.8% of arrestees for aggravated assault. Thus, the number of black arrestees for aggravated assault is 54% of the number for white arrestees. Given that the white population size (including Hispanics) is 5.55 times as large as the black population size (75.5% vs 13.6%, Census [archived]), this implies that the aggravated assault arrest rate for blacks is 0.54 * 5.55 = 2.98 times the rate for whites (a similar figure can be calculated using this source showing arrest rates by race in 2020, using a slightly different dataset).

To test whether the assault arrest rates by race reflect true assault rates by race, we can examine assault victimization rates by race. In this section,  I consider data on rates of hospitalizations and emergency department visits due to injuries caused by assault. This includes both fatal and non-fatal injuries, though I’ll focus on non-fatal injuries as fatal injuries were considered in the previous section. I will then check whether racial differences in rates of injuries due to assault are in line with the FBI-reported racial differences in rates of arrests for assaults.

According to an analysis of National Crime Victimization Survey data by Planty et al. (2013) [archived], only a portion of incidents (25.1%) of violence resulted in an injury, and only a portion of those incidents resulted in treatment at a hospital. Thus, the data here can only provide information on racial disparities in assaults that were serious enough to result in injuries requiring medical treatment. Regardless, this data is still useful because it provides an additional independent proxy for assault rates by race (albeit relatively severe assaults).

It should be noted that, unlike homicide, I don’t have direct data showing that most assaults are intra-racial. However, there is some indirect reason to believe this is the case. According to the latest iteration of the National Crime Victimization Survey (NCVS), about 72% of violent offenses against black victims involved a black offender, among cases where the race of the offender was known (Thompson and Tapp 2023, Table 13). Importantly, assaults accounted for the vast majority (81%) of the reported violent crimes in 2022 (Table 1). So black assault victimization rates should be a good proxy for black assault crime rates.

By contrast, only about 40% of violent offenses against Hispanic victims involved a Hispanic offender, among cases where the race was known. Thus, it’s possible that Hispanic assault victimization rates are not a good proxy for Hispanic assault crime rates (another possibility is that NCVS respondents are less able to correctly identify their offenders as Hispanic).

Firearm assaults

First, I will review racial disparities in firearm injuries due to assault. About 28% of aggravated assaults in 2019 were committed with firearms. Across many different datasets, studies consistently find large racial disparities in firearm assault injury rates, with the disparities typically being much greater than the FBI-reported racial disparities in assault arrest rates.

Nationwide Inpatient Sample (NIS)

Salemi et al. (2015) [archived] examined inpatient hospitalizations for firearm injuries in the United States between 1998 and 2011 using the Nationwide Inpatient Sample (NIS). The authors describe the NIS as “the largest all-payer, publicly-available inpatient database in the US”. The NIS collects data on a random 20% sample of all non-federal community hospitals from participating states (which has grown from 22 states in 1998 to 46 states in 2011). The NIS contains approximately 7 million inpatient hospitalizations each year, which provides estimates of 36 million hospitalizations when weighted.

The authors found that firearm hospitalizations, particularly for injuries due to assault, were much more common among non-Hispanic black patients. The following table shows the number of hospitalizations by race and injury intent:

The findings were described as follows by the authors:

Table 2 describes the distribution of inpatient hospitalizations for FRIs by the reported manner or intent of injury and the geographic location in which the injury occurred. Assault was the intent of injury in 6 of every 10 FRI hospitalizations overall, and was more likely the reason for the FRIs among Hispanics (72.8%), NH blacks (70.8%), patients with lower household incomes, patients without private insurance, and patients 14–34 years of age.

Note that the vast majority of these hospitalizations did not result in an immediate death. Only 5.7% of hospitalizations due to assault resulted in death before being discharged. Thus, the numbers presented here are likely good estimates of frequency of non-fatal firearm assault injuries by race.

The data shows that blacks accounted for 44% (113,812/257,675) of all hospitalizations for firearm assault injuries. Note that this is a lower bound, as the denominator includes patients with unknown or unreported race who will contain some number of black patients. When excluding patients with unknown/unreported race, blacks accounted for 57% (113812/199738) of patients hospitalized for firearm assault injuries. These findings clearly support the Parity Hypothesis, as these percentages are much greater than the FBI-reported percentage of assault arrestees (~30%).

Furthermore, there were nearly 4 times more firearm assault injuries to black patients than  to white patients, despite the fact that the white population is nearly 5 times larger (excluding Hispanics). Hispanics also have much higher rates of firearm assault injuries than non-Hispanic whites in this dataset. There were about 50% more firearm assault injuries to Hispanic patients than to non-Hispanic white patients, despite the non-Hispanic white population being several times as large as the Hispanic population.

Another interesting finding is that while blacks and Hispanics are more likely to suffer injuries due to assault, non-Hispanic whites are more likely to suffer self-inflicted injuries. Non-Hispanic whites accounted for 61% (80% when excluding patients with unknown race) of patients hospitalized for intentional or self-inflicted injuries. This pattern will be replicated in other datasets.

Other studies finding similar racial disparities in firearm injuries in the NIS include Lal et al. (2023), Schnippel et al. (2021), Cook et al. (2017), and Kalesan et al. (2016), and Agarwal (2014). The findings from these studies are mostly identical to the findings already reviewed, except that some of these studies provide data on injury rates for Asian Americans. Those studies find that Asian Americans have the lowest firearm injury rates of any racial group.

National Trauma Data Bank (NTDB)

Foote et al. (2022) [archived] examined patterns of firearm injuries between January 2017 to December 2019 in the National Trauma Data Bank (NTDB). The NTDB contains data from millions of electronic records in hundreds of trauma centers across the United States. The authors describe the NTDB as “the world’s largest trauma data repository”. All level I and level II trauma centers present data to the NTDB whereas level III and IV trauma centers submit data voluntarily.

The authors split firearm injuries into two categories, “homicides” and “suicides”. There were 100,031 homicides and 11,714 suicides across the study period. It should be noted that the “homicide” category was defined somewhat unintuitively: the “homicide” group included firearm injuries due to assault (N=97,639), legal intervention (N=2,303), military operations (N=9), terrorism (N=76), and war operations (N=4). Thus, the “homicide” category can be considered an estimate of injuries due to assault, since assaults account for about 98% of the category. Moreover, despite the names “homicide” and “suicide”, the majority of the firearm injuries for both categories did not result in death. It appears the category names were meant to describe intent rather than outcome. In-hospital mortality rate was about 12% for homicide firearm injuries and 45% for suicide firearm injuries.

The authors found large racial disparities in firearm injuries, with the exact split depending on the firearm injury category. Specifically, the majority of homicide firearm injury victims were black, whereas the majority of suicide firearm injury victims were white.

Demographics between the two groups differed in several ways. Median age for suicide subjects (36, Interquartile range (IQR): 19, 54) was 16 years older than for homicide subjects (20, IQR: 14, 30) (Table 1). African–American subjects made up 62% of the homicide group and only 10% of the suicide group. Asian, American Indian, and White subjects had higher percentages in the suicide group than the homicide group (Table 1). Males comprised the vast majority of both the homicide and suicide groups, 88% and 83%, respectively.

The percentage of firearm injuries for each racial group are presented in the following table:

Thus, blacks accounted for 62.3% of those with assault firearm injuries. These findings support the Parity Hypothesis, as this percentage is far greater than the black share of FBI-reported assault arrestees (~30%).

Other studies of the same dataset have produced similar findings. For example, Olufajo et al. (2020) examined firearm injuries specifically among youth in the National Trauma Data Bank (NTDB). They find that “while Blacks accounted for most (68.3%) deaths from pediatric firearm assault, 71.3% of the victims of pediatric firearm suicide were White individuals.”

National Electronic Injury Surveillance System (NEISS)

Kalesan et al. (2017) [archived] analyzed firearm injuries in the United States during 2001 to 2013 based on data from the CDC’s Web-based Injury Statistics Query and Reporting System (WISQARS). Fatal firearm injury data are collected by the National Center for Health Statistics (the same source cited earlier on racial disparities in homicide death rates). Non-fatal firearm injury data are collected by the National Electronic Injury Surveillance System – All Injury Program (NEISS-AIP), a collaborative program between the CDC and the Consumer Product Safety Commission. The NEISS-AIP collects data on non-fatal injuries from 66 of the 100 NEISS-AIP designed hospital emergency departments (EDs). These hospitals are a nationally representative, stratified probability sample of all US hospitals with at least 6 beds and 24-hour emergency services.

During the study period, there were about 1.3 million firearm-related injuries, with about 400k of those injuries being fatal. Table 2 shows rates of fatal and non-fatal firearm injuries by race/ethnicity and by intent. I already covered racial disparities in fatal assaults earlier (see homicide death rates), so we should focus on racial disparities in non-fatal firearm injuries. The data is as follows: 

The data here is consistent with all of the other data presented thus far. For each year of data, black persons were roughly 20 times more likely than white persons to be hospitalized due to a non-fatal firearm assault injury. Hispanic persons were about 6 times more likely than white persons to be hospitalized for similar reasons.

Comparing the relative rates of injury (i.e. rate per 100,000) to the relative arrest rates in the FBI will be somewhat tricky because this dataset excludes Hispanics from the “white” category unlike the FBI. However, we can safely estimate that the non-fatal firearm injury rate for blacks would be >3 times the rate for whites even if Hispanics were included in the white category. That’s because the injury rate for blacks is about 20 times the rate for whites and about 3 times the rate for Hispanics. So if the “white” category included Hispanic whites (which would still be overwhelmingly non-Hispanic white), the black-white ratio in firearm injury rates would be considerably greater than the black-white ratio in FBI-reported arrest rates (which is about 3 to 1). Thus, these findings support the Parity Hypothesis.

Earlier reports of data from the NEISS have reported similar findings on racial disparities in fatal and non-fatal firearm injuries. For example, the CDC published a report by Gotsch et al. (2001) [archived] to conduct surveillance for fatal and non-fatal firearm injuries in the United States during 1993 to 1998 using NEISS data. The authors found that “Black males aged 20–24 years had the highest average annual fatal (166.7/100,000 population) and nonfatal (689.4/100,000 population) firearm-related injury rates during the 6-year period”. More detailed data on firearm injury rates by race/ethnicity were presented in Table 5:

Again, data on fatal assault injury rates have already been covered earlier, so we should focus on non-fatal assault injury rates. Here are what I believe are the most interesting findings:

  • The non-fatal firearm assault injury rate was about 17 times greater for blacks compared to whites (82.1 vs 4.7 per 100,000) over the study period. For comparison, the rate for males was about 8 times greater than the rate for females (34.3 vs 4.3 per 1000,000 see top of Table 5). Thus, the black-white disparity was over twice as large as the male-female disparity.
  • Perhaps more surprisingly, the non-fatal firearm assault injury rate for black females was over twice the rate for white males (16.8 vs 8.1 per 100,000).
  • Hispanics also experienced relatively high rates of non-fatal firearm assault injury compared to whites. The rate for Hispanics was about 8 times greater than the rate for whites (37.7 vs 4.7 per 100,000), which is about the same as the male vs female disparity.
  • Also, Hispanic females had about the same rate of non-fatal firearm assault injury as white males over the study period (8.0 vs 8.1 per 100,000).

Plenty of other studies have found similar racial disparities in firearm injury rates in the NEISS (Eber et al. 2004, Parker 2020).

Other datasets

Spitzer et al. (2020) [archived] examined firearm injuries in California from 2005 to 2015. Non-fatal firearm data was gathered from databases containing all emergency department and inpatient records from California-licensed hospitals. A total of 81,085 non-fatal firearm injuries were identified from 2005 to 2015, with most of the injuries being due to assault (69.7%). Consistent with all other data, the authors found that black men had the highest rate of firearm assault injury, followed by Hispanic men: “Overall, Black men had an annual firearm assault injury rate of 126.5 per 100 000 people, 4-fold that of Hispanic men, the racial/ethnic group with the next highest rate (30.6 per 100 000 people)”. The rank-order rates of non-fatal firearm assault injury are consistent with the prior data: black males tend to have rates of firearm assault that are about 15 times the rate for white males, Hispanic males tend to have rates that are about 3 to 4 times the rate for white males, and Asian males had rates that were similar or lower than white males.

Another interesting finding of the California data is that the non-fatal firearm assault injury for black females is greater than the rate for white males, consistent with data presented earlier. The rate of non-fatal firearm assault injuries for women was presented in Figure 3 in the supplementary material. The figure shows that, from 2005 to 2015, the rate for black females decreased from about 17 to 12 injuries per 100,000 women. By contrast, Figure 2 shows that the rate for white males decreased from about 9 to 7 injuries per 100,000 men over the same time period.

Racial disparities in firearm injuries have been observed in many other datasets (Goel et al. 2023, Blumberg et al. 2018, Kalesan et al. 2013). For example, Goel et al. (2013) found that black children accounted for the majority of emergency department visits and hospitalizations for firearm injury across two separate databases (NEDS and KID):

Racial disparity pertaining to gun violence is well established. Black and Hispanic students are disproportionately impacted by gun violence and associated trauma. In our study, Black children had both the highest overall prevalence of FI-associated admissions as well as the majority of unintentional and assault related admissions. Per the 2020 US census, 14% of the US population were Black, yet Black children accounted for 50.2% of ED visits and 52.6% of hospitalizations for FI in our study, thus offering another example of racial and socio-economic disparities in the US extending to the pediatric age group. By contrast, White children had the highest number of intentional FI, which is consistent with previous reports, and likely reflects suicide attempts.

There were also other studies I found reporting large racial disparities in firearm injury rates, but I did not cover them in detail because they were either restricted to certain regions or covered small samples (Rodriguez et al. 2017, Carter et al. 2017, Shaahinfar et al. 2018).

Knife-related assaults

In this section, I will consider race differences in injuries due to knife-related assaults. About 18% of aggravated assaults in 2019 involved “knives or cutting instruments”. Consistent with the previous data, blacks are much more likely to suffer injuries due to knife-related encounters or violence.

Vaughn et al. (2023) examined factors related to knife-related crimes using the 2019 Nationwide Emergency Department Sample (NEDS). NEDS represents 989 emergency departments from 40 states. The data set contained information on 33.1 unweighted million ED visits, corresponding to nearly 143 million weighted ED visits. Approximately 0.06% of all adult ED visits documented assault by a sharp object. When examining all ED visits due to a sharp object of any cause, blacks were somewhat overrepresented: whites were the most prevalent victims (58.9%), followed by blacks (21.2%), Hispanics (14%), and “other” (5.9%) (Table 1).

However, when focusing solely on ED visits due to assault with a sharp object, blacks were the most prevalent victims (46.7%), followed by whites (27.7%), Hispanics (18.6%), and other (7.1%). This provides support for the Parity Hypothesis, as the black share of assault victims (46.7%) greatly exceeds the black share of FBI-reported assault arrestees (~30%).

Another interesting finding from this study is the fact that blacks had much higher rates of assault injuries after adjusting for various control factors. Even when accounting for insurance status, region, urbanicity, zip code, etc. the authors found that blacks and Hispanics were more likely to visit the ED for knife-related assaults rather than other causes: “Compared to non-Hispanic white patients, patients who identified as black, Hispanic, or other races/ethnicities were at higher risk for victimization.”

Wolf et al. (2019) used the National Trauma Data Bank (NTDB) to analyze rates and costs associated with injuries due to firearms and cut or piercing encounters among children aged 17 years or younger. I’ll focus only on the findings regarding cuts and piercing as data on firearms has already been discussed thoroughly above. As mentioned above, the NTDB represents the largest collection of trauma data in the United States. Between 2007 and 2016, the authors identified 21,270 encounters for cut or pierce injuries.

The authors found that black children comprised about 31.1% of admitted patients due to cut or piercing injuries. This is already highly disproportionate relative to the black share of the population. However, the injury data was not disaggregated by both race and intent. Only about 41% of cut and pierce injuries were due to assault, whereas about 51% were due to unintentional injuries and 9% were self-inflicted. When focusing solely on assault encounters, it’s likely that an even greater proportion of victims would be black (given the previous data from previous studies).

Other assaults

One potential problem with the previous sections is that they only focus on rates of assault injuries using certain types of weapons (e.g., firearms, knives, etc.). For example, a defender of the Bias Hypothesis might object that firearm and knife assaults are only a portion of total assaults; if we instead looked at total assault injuries (including those without a firearm or knife), then the findings would not vindicate the Parity Hypothesis. The first thing to note about this objection is that nearly half (45.1%) of aggravated assaults were committed with firearms (27.6%) or knives or cutting instruments (17.5%) in 2019. So it’s unlikely that the main findings would be overturned after considering other injuries, unless the racial patterns for half of assaults are completely different from the other half.

Regardless, to address this objection, this section will consider rates of assault injuries by race that are not restricted to a particular weapon. This analysis reveals that, while it is true that racial disparities in assault injuries generally are not as great as racial disparities in, say, firearm assault injuries specifically, the disparities are nevertheless in line with the FBI-reported racial disparities in arrests. Thus, these assault injury rates by race also vindicate the Parity Hypothesis.

Barry et al. (2022) [archived] examined hospital costs using the 2016-2018 Nationwide Emergency Department Sample (NEDS) and the National Inpatient Sample (NIS). I will only focus on the results from the NIS dataset, because that is the only dataset that includes results broken out by race. The authors estimated results for 184,040 inpatient admissions from 2016 to 2018. The data indicated that black patients were disproportionately victims of assault, particular assaults involving sharp objects of firearms:

Of persons presenting to the ED due to assault, the largest proportion of females (NEDS, 47%; NIS, 26%) had experienced bodily force; 13% (NEDS) and 11% (NIS) had been assaulted with firearms. Individuals with assault due to firearms had the lowest mean age (NEDS, 29.2 years; NIS, 29.7 years) with most covered by Medicaid (NEDS, 39%; NIS, 52%) or having no insurance (NEDS, 41%; NIS, 28%). Across all assault mechanisms, most occurred in the Southern part of the US. The most common sites of injuries due to firearm assaults were the extremities (NEDS, 41%; NIS, 27%) and multiple locations (NEDS, 30%; NIS, 48%). In the inpatient setting, most bodily force (48%) and blunt object (41%) assaults involved patients who were White, and most sharp object (37%) and firearm (60%) assaults involved patients who were Black. Firearm assaults had the lowest ratio of ED to inpatient records (2:1); for every firearm assault recorded in the inpatient setting, there were 2 ED records, and bodily force had the highest ratio of ED to inpatient records (22:1).

The black patients were also disproportionately involved in assaults involving just bodily force or a blunt object:

Performing a weighted average across all mechanisms suggests that blacks constituted about 37% of all patients for assault. For comparison, whites constituted only about 36% of patients for assault, despite the white population being much larger than the black population. About 18% of patients victimized by assault were Hispanic. These findings provide support for the Parity Hypothesis. Black patients accounted for 37% of all patients admitted for assault, which is greater than the black share of FBI-reported assault arrestees (~30%).

Monuteaux et al. (2017) used the National Hospital Ambulatory Medical Care Survey to analyze a representative survey of emergency department visits from 2000 to 2010. The authors estimated rates of ED visits for violent injuries and calculated corresponding medical and work-loss costs accrued by these injuries. The authors found that about 1.4 million adults were treated for violent injuries, comprising 1.6% of all ED visits. Of the ED visits due to violent injuries, about 5% resulted in hospitalizations. The causes of violent-related injuries were fight, brawl, rape (52%), assault by cutting and piercing instrument (7.9%), assault by firearms (2.4%), other categories (2.7%), and assault by other and unspecified means (35%). The authors note that black men had the highest rates of violent-related injuries:

The rates of ED visits for violent-related injuries were highest for black males, ranging from 138 of 10,000 in 2003 to 240 of 10,000 in 2009 (rate difference, −102.1; 95% CI, −216.2 to 11.9). For every year from 2000 to 2010, the rates of violent-related injuries for black men were significantly greater than those of white men. Similarly, the rates for black females were significantly greater than that of white women every year. Furthermore, for every year except 2000, 2005, and 2010, the rates for black women were significantly higher than those of white men.

The rates of ED visits by race, sex, and year are presented in Figure 1:

This graph includes Hispanics in the “white” category, similar to the FBI arrest data. Nonetheless, we can see that blacks consistently have much greater rates of violent-related ED visits than whites. Unfortunately, the exact figures by year are not provided, so we cannot determine the exact ratio of the black to white rate of violent-related ED visits. However, from eyeballing the charts, it does seem that the rate for blacks is about 3 times the rate for whites. This provides support for the Parity Hypothesis, because the FBI data also suggests that blacks are about 3 times more likely than whites to be arrested for assault.

Rowe et al. (2021) [archived] examined non-fatal assault injuries in California from 2005 to 2015. The authors used hospitalization and emergency department discharge records from the California Office of Statewide Health Planning Department, which includes all inpatient and hospital discharge records in California during this time period. The authors excluded records that resulted in death so they could focus on non-fatal injuries. There were 1,494,739 assault injuries identified. The most common means of injury was “unarmed fight or brawl”, which on average resulted in about 140 injuries per 100,000 population per year (see Table 1), which is about 38% of the average total rate of injuries per year (364 per 100,000 population). The rates for the remaining means of injury were as follows: other means (77 per 100k), unspecified means (48), blunt object (33), sharp object (27), legal intervention (26), and firearm (13). Thus, the vast majority of these injuries can be considered to be non-firearm related injuries.

Consistent with prior data, the authors found that black people had the highest rate of assault injuries compared to all other racial groups. In fact, the injury rate for blacks significantly increased over the time period:

Age-adjusted annual rates among African Americans were consistently higher than among other racial/ethnic groups and increased 33% from 900 to 1,194 (Figure 1). In addition, the rate increased 35% from 423 to 572 among American Indian/Alaskan Natives, while remaining stable among other racial/ethnic groups.

The injury rates for other racial groups are shown in the following table:

This data provides some support for the Parity Hypothesis. The mean assault injury rate for blacks was 1,098 per 100,000 population, compared to 347 per 100,000 for whites and 326 per 100,000 for Hispanics (Table 1). Thus, the assault injury rate for blacks in California is about 3 times the rate for whites and Hispanics. This is consistent with the FBI-reported arrest rate, which shows that blacks have about 3 times the arrest rate for whites.

Interestingly, the injury rate for Hispanics was typically lower than that of non-Hispanic whites. This may be because Hispanics are less likely to engage in certain kinds of assaults included in the comprehensive data set (such as e.g. fights or brawls) or because Hispanics are less likely to receive medical treatment for their injuries. Regardless, this finding is in conflict with much of the other data cited above, which shows that Hispanics are more likely to be victims of assault.

Conclusion

The data in this section consistently provides support for the Parity Hypothesis. According to the FBI arrest data, blacks account for 31 to 33% of arrestees for assault, with an arrest rate that is 2.7 to 3.0 times the rate for whites (including Hispanics). Each study here found that blacks are overrepresented as assault victims to a similar or greater degree. The findings regarding Hispanics were a bit mixed. Hispanics were consistently overrepresented as victims of firearm assault injuries, but they were sometimes not overrepresented when including non-firearm assault injuries.

NCVS data: victim-reported race of offenders


In this section, I consider racial differences in rates of violence offending according to victim-reports from the National Crime Victimization Survey (NCVS). The BJS describes the NCVS as follows:

The BJS National Crime Victimization Survey (NCVS) is the nation’s primary source of information on criminal victimization. Each year, data are obtained from a nationally representative sample of about 240,000 persons in about 150,000 households. Persons are interviewed on the frequency, characteristics, and consequences of criminal victimization in the United States. The NCVS collects information on nonfatal personal crimes (i.e., rape or sexual assault, robbery, aggravated and simple assault, and personal larceny) and household property crimes (i.e., burglary/trespassing, motor vehicle theft, and other types of theft) both reported and not reported to the police. Survey respondents provide information about themselves (e.g., age, sex, race and Hispanic origin, marital status, education level, and income) and whether they experienced a victimization. For each victimization incident, the NCVS collects information about the offender (e.g., age, race and Hispanic origin, sex, and victim-offender relationship), characteristics of the crime (e.g., time and place of occurrence, use of weapons, nature of the injury, and economic consequences), whether the crime was reported to police, reasons the crime was or was not reported, and victim experiences with the criminal justice system.

So NCVS is the best available source for self-reported rates of victimization. That being said, there are a few problems with using self-reported rates of victimization to assess racial differences in criminality:

  • There may be racial differences in the likelihood of reporting crimes on the survey. There is some data indicating that black individuals may underreport their history of deviant behavior, crime, and/or arrest (Kleck 1982Fendrich and Johnson 2005Kirk 2006Ledgerwood et al. 2008). Similar underreporting could also be true regarding criminal victimization. I’m not saying such underreporting is occurring definitively. Rather, I’m just noting that it’s possible that self-reports are not reliable sources of racial differences in criminal victimization.
  • There may be biases in the reported race of the offenders. For example, victims may be more likely to recall a victimization when the offender was black rather than white (or vice versa). Or victims may be more likely to incorrectly identify their offender as black rather than white (or vice versa).

While these are two problems worth keeping in mind, the data nevertheless provides some additional independent evidence (albeit imperfect) on racial differences in criminality, so it should be taken into consideration. Ultimately, the findings from this data converges to the same conclusion as all the other data, showing that the black share of victim-reported offenders reflects the black share of FBI-reported arrestees. Therefore, this data qualifies as an additional point supporting the Parity Hypothesis.

2010 Data

Beck and Blumstein (2017) analyzes the degree to which racial disparities in imprisonment can be explained by racial disparities in arrests. As part of this analysis, the authors also examined the degree to which racial disparities in arrests can be explained by racial disparities in criminal involvement. Criminal involvement by race was proxied using the NCVS. The authors’ limited their analysis to rape, sexual assault, robbery, aggravated assault, and simple assault, as these are crimes in which the victims had contact with their offenders and could describe their offender’s race.

Before presenting their own analysis, the authors summarized the findings of a previous analysis which compared racial disparities in arrests to racial disparities in victim-perceived race of offenders in the 80s. The previous analysis found that the black share of arrestees reflected the black share of offenders according to victimization survey respondents:

Analysis of the NCVS data has been carried out by Tonry (1995). He examined the victims’ identification of blacks as their assailant in robberies and aggravated assaults for the years 1980–1991. He found that victim data on perceived race of assailants closely paralleled the black arrests among arrestees. For example, the percentages among robbery assailants in 1980, 1985 and 1990 were 54.8% (nonwhite), 55.5% (black), and 51.5% (black). The percentages of blacks among persons arrested for robbery were 57.7, 61.7, and 61.2%. He did not include victim reports in offenses involving multiple offenders.

The authors of the present study conducted the same analysis using data from the 2010 iteration of the NCVS and arrest data. Again, the authors replicated earlier findings, observing that black overrepresentation in arrests almost perfectly matched victim-perceived black overrepresentation of offenders:

We examined the issue using NCVS data on the violent crimes more broadly, including rape/sexual assault, robbery, aggravated assault, and simple assault. We also included data on race of the assailants for incidents involving multiple offenders. Table 10 presents the percentage of black offenders as reported by victims in the 2010 NCVS and the percentage black among adult arrestees as reported in the UCR. For all four crime types, the ratio of the black percentage of arrestees and of reported offenders are very close to 1.0, ranging from .91 to 1.10. The aggregate ratio for violent crimes was 0.97. These results provide a strong indication of support for the use of arrest as a proxy indicator of criminal involvement in non-fatal violent crime.

The black percentage of offenders and the black percentage of arrestees for each crime type are reported in Table 10:

As you can see, the black percentage of offenders according to victimization surveys aligns nearly perfectly with the black percentage of arrestees. The ratio of the former to the latter hovers between 90% and 110%.

2018 Data

A more recent comparison of racial disparities in NCVS vs arrest data was also conducted by Beck (2021) [archived] using 2018 data. Consistent with prior results, the author again found that black overrepresentation in arrests for violent crime closely matched black overrepresentation in victim-reported violent crime offenders, particularly when the analysis was limited to victim reported incidents that were reported to the police.

An examination of offenders’ characteristics, as reported by victims in the NCVS, provides information on racial and ethnic disparities beyond an arrestee and population-based comparison. Based on the 2018 NCVS and UCR, black people accounted for 29% of violent-crime offenders and 35% of violent-crime offenders in incidents reported to police, compared to 33% of all persons arrested for violent crimes (table 2).

The results for all racial groups were reported in Table 2:

Similar to the previous reports, violent crime includes rape, robbery, aggravated assault, and other assault. “Other assault” includes offenses such as stalking, intimidation, coercion, and hazing. When the analysis focuses on serious violent crimes (which excludes simple assault), there are similar results:

Among the most serious incidents of violent crime (rape or sexual assault, robbery, and aggravated assault), there were no statistically significant differences by race between offenders identified in the NCVS and persons arrested per the UCR (table 3). White and black people were arrested proportionate to their involvement in serious nonfatal violent crime overall and proportionate to their involvement in serious nonfatal violent crime reported to police. Hispanics accounted for 21% of persons arrested for serious nonfatal violent crime but 12% of persons involved in serious nonfatal violent crime reported to police. However, some of this difference may be due to victims not knowing the ethnicity of their assailants, even if they knew their race.

These findings suggest that blacks are not overrepresented in arrests relative to their proportion of perceived offenders according to victim surveys. Blacks constitute roughly 36% of offenders identified by victims in the NCVS and 36% of arrestees according to the FBI. On the other hand, Hispanics are overrepresented in arrests relative to their proportion of perceived offenders in victim surveys. This may suggest that Hispanic arrest rates are not reflective of Hispanic crime rates. However, another plausible explanation is that victims are less able to identify whether their offenders are Hispanic.

Conclusion

The respondents to the National Crime Victimization Survey (NCVS) reported rates of offending by race that perfectly lined up with FBI-reported rates of arrest by race. That is, black overrepresentation in FBI-reported arrestees nearly perfectly matches black overrepresentation in victim-reported offenders. This convergence is exactly what would be predicted by the Parity Hypothesis. Thus, the findings in this section provides evidence for the Parity Hypothesis.

A defender of the Bias Hypothesis might try to explain these findings by stipulating that the survey respondents are actually overrepresenting the black proportion of offenders. If that’s true, then the victim-reported race of the offenders would not actually reflect the race of violent criminals. However, I’m aware of no independent evidence for this stipulation, so it’s currently just an ad-hoc assumption to rescue the Bias Hypothesis.

NIBRS data: race of offenders in police reports


This section will consider the perceived racial background of suspects identified in police reports. The Parity Hypothesis predicts that perceived race of offenders in police reports would align with the race of arrestees, because we would expect racial disparities in crimes to be reflected in police reports. The Bias Hypothesis does not predict that perceived race offenders will converge with the race of arrestees, unless one makes the ad-hoc assumption that police reports are biased against black offenders in a way that coincidentally aligns with all of the independent data sources mentioned above.

The police report data comes from the FBI’s National Incident-Based Reporting System (NIBRS). The NIBRS is the FBI’s newer crime data collection program which retires [archived] the traditional Summary Reporting System (SRS) (See here for different data collections on the FBI’s UCR program). One of the most useful features of the NIBRS, as D’Alessio and Stolzenberg (2003) explains, is that it allows us to link a reported crime to an arrest. This allows us to estimate whether there is an association between race and the likelihood of arrest using official law enforcement data:

NIBRS represents the next generation of crime data and it is designed to replace the nearly 70-year-old UCR. The intent of NIBRS is “to enhance the quantity, quality, and timeliness of crime statistical data collected by the law enforcement community and to improve the methodology used for compiling, analyzing, auditing, and publishing the collected crime data” (Federal Bureau of Investigation 2000:1). NIBRS is unique because rather than being restricted to a group of eight Index crimes that the summary-based program uses, it gathers information from individual crime reports recorded by police officers at the time of the crime incident for 57 different criminal offenses. The information collected by police typically includes victim and offender demographics, victim/offender relationship, time and place of occurrence, weapon use, and victim injuries. Because NIBRS is capable of producing more detailed and meaningful data than that generated by the traditional UCR, it is a valuable tool in the study of crime.

NIBRS data are well suited for our intentions because it is possible to link a reported crime incident to a subsequent arrest that was heretofore not feasible with the UCR. The ability to merge crime incident data with arrest data enables researchers to calculate the actual probability of arrest by race for crimes communicated to the police where the victim is able to identify the race of the offender. This is the most appropriate strategy for evaluating the discriminatory use of the arrest sanction because the police can only act upon illegal behaviors that come to their attention. These data also afford us the opportunity to examine how the arrest sanction is influenced by a number of salient factors about which Hindelang lacked data, such as whether the victim was injured, the race of the victim, the victim/offender relationship, and weapon use.

Thus, the NIBRS police report data can also be used to estimate whether there are racial differences in the likelihood of arrest conditional on offender race being reported to police. If we assume that crimes are similarly likely to be reported to police regardless of offender race (NCVS provides some evidence for this, as shown below), then this data can also be used as a proxy for whether there are racial differences in the likelihood of arrest conditional on a crime being committed.

Recent analysis

Fogliato et al. (2021) [archived] analyzed racial disparities in arrest likelihood for white and black violent offenders using data collected from the NIBRS between 2007 and 2016. The NIBRS collects data from law enforcement agencies regarding the characteristics of known offenses and arrests, such as the demographics of the victims and offenders. Offense data is based on crimes reported to law enforcement, including crimes which did and did not result in an arrest. The authors limit the analysis to the 16 states which submitted all crime data through the NIBRS in 2014 (AR, CO, DE, ID, IA, DY, MI, MT, NH, ND, SC, SD, TN, VT, VA, and WV). They also focused only on incidents involving a single offender and single victim, where both the victim’s and offender’s races were recorded as either white or black (including Hispanics). In total, the authors collected data on millions of offenses for 5 types of violent crime: 9,181 instances of murder/non-negligent manslaughter, over 103,309 forcible rapes, over 101,133 robberies, nearly 596,324 aggravated assaults, and over 2,669,399 simple assaults.

The key variable of interest in this study was the arrest rate by race, which is “the share of offenses that result in arrests”. The authors found that white offenders had higher arrest rates than black offenders across all crime types:

Arrests are more frequent in case of incidents involving White offenders across all types of crimes (overall, 56% for White vs. 42% for Black offenders). However, there exist differences in the gap in arrest rates across types of offenses. The gap is largest in case of robbery (36% vs 19%) and aggravated assault (62% vs 44%), followed by simple assault (57% vs 43%), which constitute the majority of the offenses present in the data. In contrast, the gap is small for murder/non-negligent manslaughter (73% vs 67%) and rape (27% vs 25%). Lastly, we assess the racial composition of victims. Intraracial crimes constitute four-fifths or more of the offenses across all types of crimes, with the exception of rape (63%) and robbery (57%) offenses committed by Black offenders. Interestingly, arrest rates for interracial crimes are similar across racial groups (within 2%) for all types of crimes other than robbery. In summary, we observe that arrest rates are highest for intra-racial offenses among Whites, followed by interracial offenses, and last by intra-racial offenses among Black individuals.

These results were also presented in table 1:

As you can see, the arrest rate for black offenders is lower than the arrest rate for white offenders across every type of crime. Also interestingly, we see that arrest rates are relatively high for murder and relatively low for rape and robbery. Arrest rates for assault fall somewhere in between.

The authors next examined racial differences in arrest rates across each of the 16 examined states, to account for possible macro-level variations in arrest rates. The authors first analyze racial differences in arrest rates for simple assault, by far the most common type of violent crime committed. They find that white offenders have higher arrest rates than black offenders in nearly each state considered, however the gaps are not as large as the overall gap. The findings are discussed as follows:

We start by decomposing the gap in arrest rates between White and Black offenders (13%=57%-43%) into state-level differences, which are represented by the green stars in Figure 1. Two notable patterns are worth mentioning. First, there are quite large variations in arrest rates across states. While in Arkansas less than 35% of all offenses resulted in the arrest of the offender, in Delaware and Vermont 79% and 82% did respectively. Second, we observe that the gap in arrest rates between Black and White offenders varies substantially across states. Although White offenders were arrested at higher rates than Black offenders in almost all of the states considered, the pooled mean of the gap in arrest rates is 6% (std.dev.=6%) and thus it is smaller than the overall gap.

Thus, the overall gap in arrest rates (13 percentage points) is over twice as large as the pooled mean state-level gap in arrest rates (6 percentage points). These results for each state are illustrated in the following figure:

The authors then analyzed racial disparities in arrest rates for simple assault in select states. They similarly conclude that much of the racial disparity in arrest rates is due to the fact that black offenders tend to reside in cities with lower arrest rates in general. When comparing arrest rates for black and white offenders within the same city, racial disparities in arrest rates tend to be rather small. This explains why the middle of the boxplots above (representing median gap in arrest rates across police agencies) are typically lower in magnitude than the green star (representing the gap in state-level arrest rates).

When analyzing other types of violent crimes (not just simple assault), the authors reported similar findings for aggravated assault and robbery, where whites had higher arrest rates than blacks in almost all states considered. However, for rape and murder, there were rather small differences in arrest rates (the average state-level difference was only about 2 percentage points for both types of offenses). They describe the findings as follows:

Based on our analysis, we can categorize the offenses in two groups. The first includes aggravated assault and robbery. For these two types of crimes, the results are similar to those that we have described above and, more specifically, arrest rates are higher for White offenders in almost all of the states considered (Figure 2). The pooled means of the state-level gaps in arrest rates between White and Black offenders for aggravated assault and robbery are 6% and 11% respectively…For crimes of rape and murder, both the aggregate and state-level differences in arrest rates across racial groups are small (average state-level diff.=2% for both types of offenses).

These results are illustrated in Figure 2:

The authors then conducted regression analyses to examine whether these differences in arrest rates by race can be explained by various incident characteristics. Separate regression models were conducted for each of the five types of violent crimes. The independent variables in the regression model included victim and offender demographics, police agency information, presence of firearms, seriousness of injury, etc. The authors found that the gaps were reduced after controlling for these additional variables, however white offenders still had higher arrest rates for most assault and robbery. There was no statistically significant difference in arrest rates by race for rape and murder:

Consistent with the high-level findings of our macro-level analysis, our estimate of the coefficient of White offender race (vs. Black) is positive and statistically significant in case of aggravated assault, simple assault, and robbery (0.03, 0.04, 0.24 respectively), and not statistically significant for forcible rape and murder (0.04 and −0.2 respectively). However, note that the coefficients estimates in case of assault, despite being statistically significant, are close to zero.

The authors conclude from this that “if the model were correctly specified”, then “ceteris paribus, an arrest is more likely for a White offender compared to a Black offender for assaults and robbery, but not for murder and rape”. However, the authors point out that it’s more realistic that “our logistic regression model is misspecified and thus such interpretations of the coefficients estimates are not correct.” The researchers elaborate on why their model is likely misspecified, however those details are not relevant for the purposes of my post. My post is focused on determining whether arrest rates by race reflect crime rates by race, not whether race is associated with arrest after controlling for confounders. That is, my post is not focused on whether race per se (as opposed to confounding factors only) influences arrest. However I do briefly address this question in the section on arrest bias near the end of this post.

Accounting for unreported crimes

One shortcoming of the NIBRS analysis is that it measures rates of offenses based on offenses that are reported to law enforcement. The above analyses will be biased if offender race is associated with the probability of a crime being reported.

Fogliato et al. (2023) [archived] attempt to overcome this shortcoming by examining the relationship between offender race and arrest after accounting for unreported crimes. The authors analyze 2006-2015 data from NIBRS and 2003-2020 data from the NCVS (the same survey described earlier in this post). The NCVS asks participants whether a reported crime was reported to law enforcement. The authors use these answers to estimate the likelihood of a crime resulting in a police notification, while also considering other information such as the race, sex, age, etc. of the victims and offenders. As with Fogliato et al. (2021), the researchers in this study focus only on data from the 16 states that reported most of their crime through the NIBRS over the study time period. They also focus only on incidents involving black or white individuals (which includes Hispanics).

The authors estimate that reported offenses in the NIBRS only capture about 44% and 48% of violent offenses committed by white and black offenders, respectively. Thus, slightly more than half of offenses are not reported to law enforcement for both racial groups. According to the survey respondent data in the NCVS, black offenders are slightly more likely than white offenders to be reported to police. Thus, blacks should be slightly overrepresented among offenses known to law enforcement. However, this overrepresentation is very small. After accounting for racial differences in rates of police reporting, the gap in arrest rates between black and white offenders decreases, but white offenders still have higher arrest rates than black offenders, particularly for assault and robbery:

We now turn to the estimation of arrest rates. Overall, 49% (standard error<1%) of the offenses known to law enforcement involving white offenders resulted in arrest, compared to 37% (<1%) of those involving black offenders. Table 1 reveals that arrest rates are similar across racial groups for sex offenses, while robbery and assault incidents white offenders result in arrest considerably more often than those with black offenders. Past works on NIBRS have reached qualitatively similar conclusions (D’Alessio and Stolzenberg, 2003; Lantz and Wenger, 2019). Despite the lower crime reporting rates, crimes with white offenders remain more likely to result in arrests than those with black offenders once unreported crimes are accounted for. Overall, arrest rates for crimes are 21% (7%) for white offenders and 17% (5%) for black offenders. Table 1 shows that arrest rates are higher for white offenders in case of assaults and robbery, and are comparable across racial groups in case sex offenses.

The detailed data is presented in Table 1:

The table shows that black offenders are slightly more likely to be reported for crimes than are white offenders. However, these differences are very small. Even after accounting for unreported crimes, the data still shows that white offenders have higher arrest rates than black offenders for all crimes except for sex offenses, where arrest rates are very similar (and very low).

The authors then conducted regression analyses to estimate the probability of arrest by race after accounting for other crime characteristics such as firearm presence, injury severity, etc. Their analysis focused only on crimes involving a single offender. In multivariate regression models accounting for additional crime characteristics, white offenders had about 23% higher odds of arrest than black offenders for robbery, but there were no statistically significant differences in arrest rates for the other crime types (sex offenses and assault):

We estimate the likelihood of arrest conditional on crimes characteristics via the two-step logistic regression detailed in Section 4. The resulting odds ratios of the coefficient estimates are reported in Table 2. In case of robbery offenses, we find that there is a positive and statistically significant association between whether the offender is white and the likelihood that the incident results in arrest. Provided that our model is correctly specified, these results would indicate that white offenders are more likely to be arrested for robbery than black offenders, ceteris paribus. The estimates of this coefficient for the other types of crimes are close to zero and not statistically significant. Thus, the estimated disparities disappear once we account for crime characteristics other than the offender’s race.

A similar set of regression analyses were conducted for crimes involving one or more offenders. The results were mostly similar to the results mentioned earlier, with white offenders having higher arrest rates for robbery whereas there were no statistically significant differences in arrest rates for other crime types (see Table 5).

The authors concluded the study with the following statements:

In this work, we have proposed estimators of the rates of police notification and of arrest for nonfatal violent crime on NIBRS that leverage data of unreported crimes from NCVS. These estimators are consistent and asymptotically normal under some assumptions. Our empirical investigation of racial disparities revealed that incidents are marginally more likely to be reported to the police when the offender is black. However, in cases of assaults and robbery, crimes with black offenders are generally less likely to result in arrests. These differences are small after accounting for crime characteristics. Additionally, the model diagnostics showed that the direction of these disparities varies with crime characteristics.

Older analyses

Many older studies have analyzed arrest rates by race using NIBRS data which have reported similar findings, i.e. racial gaps in arrest rates are small for rape and homicide, but white offenders have higher arrest rates than black offenders for assault and robbery. For example, Fogliato et al. (2023) [archived] described previous studies of NIBRS data which have reached these conclusions on the relationship between race and arrest:

There is also mixed evidence concerning the magnitude of differences in arrests across racial groups for crimes known to law enforcement. While some works have concluded that crimes are more likely to result in arrest when the offender is black (Kochel, Wilson and Mastrofski, 2011; Lytle, 2014), multiple analyses focused on violent offenses on NIBRS data have reached a different conclusion (D’Alessio and Stolzenberg, 2003; Pope and Snyder, 2003; Roberts and Lyons, 2009). These works have found that, even after accounting for crime characteristics, black offenders are less likely to be arrested than white offenders for assault and robbery. Differences for rape and homicide were found to be negligible.

For example, D’Alessio and Stolzenberg (2003) [archived] used NIBRS data to analyze 335,619 instances of rape, robbery, and assault across 2,852 reporting jurisdictions in 17 states during 1999. The authors limited their analysis to incidents involving a single offender and a single victim (reducing the sample from about 424k incidents to about 336k incidents). The authors found that black offenders are less likely to be arrested than white offenders for robbery, simple assault, and aggravated assault. There was no statistically significant racial difference in arrest rates for rape:

We begin by examining the bivariate relationship between an offender’s race and the likelihood of arrest for all the forcible rapes, robberies, aggravated assaults, and simple assaults reported to the police. Crime incidents in which the victim was unable to identify the race of the offender were excluded from the analyses. The differential arrest hypothesis predicts that controlling for crimes reported to the police, black citizens have a greater chance of being subjected to arrest. Looking simply at the two-way relationships presented in Table 2, we see that there is a consequential association between the race of the offender and the prospect of arrest for robbery, as 807/2,620 (31%) of the robberies with white offenders and 1,132/5,278 (21%) of the robberies with black offenders are cleared by arrest (2 = 82.705, p < .001).

Inspection of this table also reveals that aggravated assaults and simple assaults involving white offenders are significantly more likely to be cleared by arrest. Whites have about a 10% greater chance of being arrested for both aggravated and simple assault. Only for the crime of rape do blacks have an enhanced proclivity to be arrested by police. There is about a 28% chance of arrest for blacks, whereas whites have a 27% probability of arrest. This 1% difference, however, is not statistically significant (2= .694, p = .405). These bivariate results are interesting because it appears that whites are more likely than blacks to be arrested by police. Such findings tend to cast doubt on the differential arrest thesis, which theorizes that black criminal offenders have a markedly higher prospect of arrest than do whites.

The data is shown more clearly in Table 2:

The authors then compared arrest rates after disaggregating by both the race of the offender and the race of the victim. For most crime types, the authors found that arrest rates were greater for white-on-white crimes than black-on-black crimes.

Table 3 shows the likelihood of arrest for white-on-white crimes, white-on-black crimes, black-on-black crimes, and black-on-white crimes. The results presented in this table indicate that the police are most disposed to effectuate an arrest for aggravated assaults and simple assaults involving white offenders and white victims. These findings run counter to much of the literature suggesting that blacks who victimize whites are more likely to be sanctioned severely by the state because of the elevated status of white victims in our society. In contrast, the odds of a black offender being arrested for raping a black is higher than for any other of the other victim/offender racial combinations. Visual inspection of Table 3 also reveals that white-on-black robberies have the greatest likelihood of arrest.

The percentage for each crime type and racial pair are presented in Table 3:

The authors next conducted various regression analyses to investigate whether offender race was associated with the probability of arrest in multivariate models that account for other factors such as victim race, sex, use of weapon, victim injury, etc. These analyses revealed that, compared to black offenders, the odds of arrest for white offenders was 22% greater for robbery, 13% greater for aggravated assault, and 9% greater for simple assault even in these multivariate models. Each of these findings were statistically significant at p < 0.01 (see Table 4). The results were different for rape. Black offenders had about 9% greater odds of arrest than white offenders for rape after accounting for other variables in the multivariate model, although this difference was not statistically significant.

Roberts and Lyons (2009) [archived] also examined arrest rates by race with NIBRS data across 30 states and the District of Columbia. These authors examined 2,798 homicide incidents from 2000 to 2005 NIBRS data, and 44,667 aggravated assault incidents from the 2005 NIBRS data. Like the previous studies, these authors exclude incidents involving multiple victims and/or offenders.

The main focus of the authors’ analysis is to consider the association between victim and offender race on probability of race in multivariate models that control for other characteristics. Because the purpose of my post is to examine whether arrest rates by race reflect crime rates by race (not whether race is associated with arrest rates after controlling for other characteristics), I will focus on the descriptive statistics reported by the authors on the arrest rates by race. The following table shows the arrest rates by crime and by race of the victim and offender:

Note that Hispanics are counted as “white” in NIBRS analysis, so the non-white categories are likely almost entirely black, as mentioned in previous studies. In fact, the authors note that their analyses are virtually unchanged when substituting non-whites with blacks:

Sensitivity analyses, available on request, using dyads constructed with Whites and African Americans only (with other racial groups excluded) produced virtually identical results to those reported below (using White and non-White) for both homicide and aggravated assault clearance.

The table above shows that the arrest rates are fairly similar across all racial dyads. The homicide arrest rates are around 66% for all racial dyads except for non-white on white dyads, which has an arrest rate of about 70%. The weighted average arrest rate for non-white offenders is about 67.0%, whereas the same figure for white offenders is about 65.1%. The findings are different regarding aggravated assaults. The arrest rates for aggravated assault are nearly twice as large for white on white incidents (34%) compared to non-white on non-white incidents (19%). The weighted average arrest rate for non-white offenders is about 20.3%, whereas the same figure for white offenders is about 33.2%.

As stated above, the authors’ main analysis focused on analyzing the effect of race on arrest rates in multivariate regression models that control for other factors such as sex, victim injury, weapon use, etc. The regression analyses mostly reinforce the findings of the descriptive statistics, which is that non-whites have higher odds of arrest for homicides, whereas whites have higher odds of arrest for aggravated assaults. Specifically, controlling for other variables, non-whites have about 20% higher odds of arrest for homicides, but have about 20% lower odds of arrest for aggravated assaults (see Hazard Ratio column in Table 3).

Pope and Snyder (2003) also performed a similar analysis on NIBRS data, but they focused solely on arrests of juveniles.

Comparing raw arrests and offenses

The FBI has published overall data on offenders by race [archived] and arrestees by race [archived] according to 2019 NIBRS data. I will use this data to compare whether arrest rates by race reflect offense rates by race. Unlike the data used in the previous studies, this overall data is not limited to states that have reported all crime data to the FBI, so there is more potential for the data here to be unrepresentative due to selection in the data that is reported. However, the overall data can still be useful to analyze for a number of reasons:

  1. The overall data is not limited to a small selection of states. Thus, if findings from the overall data diverge sharply from the findings shown earlier, this may provide some (albeit very limited) indication that the previous findings cannot be generalized outside of the specific states that were examined.
  2. Unlike the data in the previous studies, the raw data provides information on arrest rates for non-violent crimes as well, such as property crimes and crimes against society.

Thus, I think it’s worth quickly checking whether arrest rates by race and offense rates by race are in agreement. For each crime type, I simply calculated and compared the black percentage of reported offenders and the black percentage of arrestees. Note: when calculating the percentages, I excluded offenders/arrestees of unknown race from the denominator.

For example, there were a total of 1,644,158 reported assault offenses in 2019, with 615,467 involving a black offender and 83,624 involving an offender of unknown race. Thus, black offenders accounted for 615,467 / (1,644,158 – 83,624) = 39.4% of offenders of known race. There were a total of 565,840 assault arrests in 2019, with 182,071 arrests involving a black arrestee and 17,412 involving an arrestee of unknown race. Thus, black arrestees accounted for 33.2% of arrestees of known race. The results for each of the crime types under “crime against persons” are reported as follows:

As you can see, for all but one crime type reported, the black share of arrestees is lower than the black share of reported offenders. The one exception is homicide, where the percentages are virtually identical (53.53% vs 53.48%). Across all crimes against persons, the percentage of offenders reported as black (39%) is considerably greater than the percentage of arrestees reported as black (33%).

Similar findings hold for crimes against property:

Again, the black arrestee share is lower than the black offender share for nearly every crime type. The exceptions to this rule are arrest rates for bribery, embezzlement, extortion/blackmail, and stolen property offenses (interestingly, these constitute 4 of the 5 least common types of property offenses). Across all crimes against property, the percentage of offenders reported as black (37%) is considerably greater than the percentage of arrestees reported as black (31%).

Finally, consider crimes against society, the least common type of offense reported.

Unlike crimes against persons and property, the black arrestee share is fairly similar to the black report offender share for crimes against society (28.7% vs 28.1%). Blacks seem most disproportionately likely to be arrested for gambling, whereas whites seem to be most disproportionately arrested for animal cruelty and pornography/obscene material.

Conclusion

For the studies that compare arrest rates by race using NIBRS data, the main findings were as follows:

  • Unadjusted differences: before controlling for other characteristics, the studies found that black offenders had lower arrest rates than white offenders for robbery and assault offenses. For homicide and sex offenses, there were not large racial differences in arrest rates and the gaps were often not statistically significant, although most recent data suggests slightly lower arrest rates for black offenders.
  • Adjusted differences: after controlling for other characteristics, the studies typically find that racial differences in arrest rates diminished. The gap in arrest rates by race for sex offenses and homicide remained trivial. The gap for assault offenses decreased as well, sometimes becoming statistically insignificant depending on the study. However, there remained a sizable gap in arrest rates for robbery even after adjusting for other characteristics, with white offenders having greater arrest rates than black offenders.

Moreover, when comparing total arrests and total reported offenses in the 2019 NIBRS data, the black share of arrestees nearly perfectly matched the black share of reported offenders for each crime type. In fact, the black arrestee share was typically somewhat lower than the black reported offender share.

These findings support the Parity Hypothesis, because the hypothesis predicts that blacks will be reported for crimes at a similar rate that they are arrested for crimes. The hypothesis predicts this because (1) we would expect that blacks will be reported for crimes at a similar rate that they commit crimes, and (2) the hypothesis posits that blacks commit crimes at a similar rate that they are arrested for crimes.

A defender of the Bias Hypothesis might try to explain these findings by stipulating that police reports actually overrepresent the black proportion of offenders. That is, one might question (1) in the previous paragraphs. For example, perhaps individuals are more likely to report a crime when the offender is black or they are more likely to incorrectly attribute a crime to a black person. However, even when one accounts for unreported crimes using the NCVS (see Fogliato et al. 2023), the rate of arrests by race still reflects the rate of reported offenders by race.

Addressing possible arrest bias


Before finishing, I want to address the topic of arrest bias in policing. This is a slightly different topic from the main question of this post, which concerns whether arrest rates by race reflect crime rates by race. I define “arrest bias” to occur if offender race itself is associated with arrest, all else equal.

This data reviewed in this post suggests that arrest rates by race reflect crime rates by race overall. However, this does not imply the absence of arrest bias. That’s because the relationship between overall arrest rates and overall crime rates by race depends on a number of factors, not just the presence of arrest bias. For example, it could be the case that police are less likely to arrest black offenders than white offenders all else equal, but black offenders are more likely to commit serious crimes that prompt more police attention which leads to more arrests. In this hypothetical, there would be an anti-white arrest bias which is offset by higher involvement with serious crimes by blacks, which on net ultimately results in the arrest rates by race reflecting the crime rates by race.

Another thing to note is that arrest bias doesn’t necessarily imply that there exists bias in law enforcement agencies. There could be other ways that race itself has a causal influence on an offender’s likelihood of arrest. For example, there could be arrest bias if people are more likely to falsely report an innocent black person for a crime, even if law enforcement agencies are no more likely to arrest a black reported offender conditional on a reported crime.

Police arrest bias was not the main focus of this post. But I came across a number of studies examining this topic while conducting research for this post, so I thought it was worth documenting some of the findings I encountered. The most interesting study I found on this topic was Lantz and Wenger (2019), which compared the odds of arrest between black and white co-offenders of the same crime. This provides an interesting way to attempt to measure whether race itself was associated with arrest, holding all else equal. Before exploring the results of this study, I want to review some of the other studies on this topic and explain why I think they are not very useful.

Other studies on arrest bias

There have been many studies analyzing the relationship between race and arrest likelihood. In a literature review of these studies, Lantz and Wenger (2019) concluded that the studies have come to mixed results: “reviews of the literature have largely declared results inconclusive or mixed, and called for more research”. For example, many studies have found that race is not associated with likelihood of arrest after introducing various controls (Pollock et al 2012, Gase 2016, Bolger 2018, Stolzenberg et al. 2021). Other studies have found that race is associated with likelihood of arrest after controlling for various proxies of criminality, with black individuals being more likely to be arrested (Kirk et al. 2008, Ousey and Lee 2008, White et al. 2014, Finkeldey and Demuth 2021). Meta-analyses do suggest that black or minority individuals are more likely to be arrested (Kochel et al. 2011, Lytle 2014).

However, there are a number of problems with using these studies to address arrest bias:

  1. These studies usually have poor measures of crime or arrest. For example, many of the studies measure criminality and/or arrest through self-reports. But many previous studies have shown that black individuals may underreport their history of deviant behavior, crime, and/or arrest (Kleck 1982, Fendrich and Johnson 2005, Kirk 2006, Ledgerwood et al. 2008). Also, some of the studies don’t actually measure arrest, but instead focus on police stops or encounters.
  2. Many of the studies focus on specific cities or sometimes even specific types of police interactions in particular cities (e.g., arrests during traffic stops in Las Vegas), which may not be representative of crimes/arrests in the country as a whole. For example, many of the studies investigate racial differences in arrest using field observations of traffic stops or police responses to domestic violence.
  3. Most of the studies focus on arrests for any crime or focus on relatively low-level crimes (e.g., traffic offenses). This ignores racial differences in the more serious crimes such as violent crimes (i.e. murder, rape, assault, robbery) or property crimes (e.g., burglary, arson, motor vehicle theft, etc.).

For these reasons, I give very little weight to the above studies to provide evidence on the presence or absence of arrest bias. Lantz and Wenger (2019) avoids many of these problems because it analyzes the NIBRS data on more severe violent crimes from many different states in the country.

The co-offending counterfactual

Lantz and Wenger (2019) [archived] used NIBRS data on assault, robbery, sexual assault, and homicide during years 2003 to 2012. Their investigation of the NIBRS data differs from previous studies in that they constructed a limited “co-offending” sample which only included incidents involving pairs of offenders that were either mixed gender (i.e. a male and female offender) or mixed race (e.g. a black and white offender). The authors performed three separate analyses in total:

  1. They analyzed the relationship between race and arrest among solo offenders in regression models that controlled for other observable characteristics (this analysis is similar to the analyses mentioned in the previous NIBRS studies).
  2. They performed the same analysis after limiting the sample to the co-offending sample mentioned earlier.
  3. Finally, they used the co-offending sample to analyze within-partnership relative rates of arrest, i.e. to analyze whether black offenders are more or less likely to be arrested than a co-offending white offender.

Each analysis controlled for additional observable factors such as weapon use, injury, sex, etc. As expected, the first two analyses produced results similar to the findings of the previous NIBRS studies, showing that black offenders have lower arrest rates than white offenders.

However, the final analysis indicated that black offenders are slightly more likely to be arrested than their white co-offending partners:

Finally, Table 4 presents effect sizes (in terms of odds ratios) for the effect of race on arrest likelihood according to the three different analyses. These samples are not necessarily directly comparable, but they are useful to consider jointly nonetheless. In the logistic regression analysis of solo offenders, the odds of arrest for black offenders were roughly 25.1% less than the odds of arrest for white offenders. This negative effect was even stronger for co-offenders at the incident level, such that the odds of arrest for black co-offenders were roughly 42.2% less than the odds of arrest for white co-offenders. But, in the within co-offender partnership analysis, the odds of arrest for black offenders were 3.1% greater than the odds for white offenders.

The effect sizes for each analysis were presented in Table 4:

Note: the effect sizes here indicate the relationship between race and arrest after controlling for a number of observable characteristics such as whether a weapon was used, whether alcohol was involved, the number of victims, etc.

As you can see, among solo offenders, the odds of arrest for black offenders was about 25% less than the odds of arrest for white offenders. In the co-offender sample, the odds for arrest for black offenders was over 40% lower than the odds of arrest for white offenders. By contrast, when the odds of arrest for black co-offenders are compared to the odds of arrest for their white co-offending partners (i.e. the “within-partnership analysis”), the odds of arrest for black co-offenders was about 3% greater than the odds of arrest for their white co-offending partners. Again, all of these figures are the effect sizes after controlling for observables.

Their findings were described as follows by the authors:

If we consider the first two analyses to be estimates of racial differences between incidents (including unmeasured omitted variables) and the co-offender analysis to be an estimate of racial differences controlling for unmeasured variables that do not vary by offender, these results suggest that (1) accounting for potential unmeasured differences between offenses and victims is an important aspect of generating unbiased estimates of the effects of race on arrest likelihood; and (2) overall, black co-offenders are slightly more likely than their white co-offending partners to be arrested for an offense, net of control measures.

In other words, black offenders are less likely to be arrested than white offenders when comparing all black and white offenders and controlling for observables. However, it may nevertheless be the case that black offenders are more likely to be arrested than white offenders, all else equal. As the study mentions, there may be unobserved selection or omitted variable bias that explains why black offenders are less likely to be arrested in aggregate even though black offenders are more likely to be arrested all else equal. For example, perhaps black offenders in general are more likely to commit crimes in areas with less efficient policing, which lowers the probability that they will be arrested.

While this study is interesting, I wish to emphasize a few limitations that they study also mentions. First, while this co-offending analysis implicitly controls for many unobserved characteristics at the offense and the victim level (e.g., the analysis implicitly controls for characteristics of the victim, the time/place of the crime, etc.), the analysis doesn’t control for unobserved characteristics at the offender level, which may also be relevant. The authors also mention this limitation:

These results suggest that, within co-offending partnerships, black offenders are more likely than white offenders to be arrested. Because these analyses are conducted within partnerships, the potential alternative explanations are limited. Because of the research design, potential explanations related exclusively to the characteristics of the offense and the characteristics of the victim are eliminated; alternative explanations must be related, in some way, to the characteristics of the co-offenders. Following this, there are at least three possible explanations for the current findings that cannot be eliminated due to the nature of the NIBRS data. First, and most importantly, it is possible that the observed arrest patterns are related to offender demeanor or the role that offenders play in the course of an offense. This analysis can control for a multitude of potential confounding measures; unfortunately, however, because the data are still based on official records, there is no way to measure the role played by the offender in the course of an offense

The authors go on to mention one possible reason why black offenders are more likely to be arrested than white co-offenders. For example, one possibility is that black offenders are more likely to act as primary offenders or instigators when co-offending with a white partner, and this might raise their likelihood of arrest. But I want to mention another possible characteristic of black offenders that might raise their likelihood of arrest. It may be that black offenders tend to be more impulsive or less intelligent than their white co-offending partners, and these psychological differences (rather than race per se) might explain why black offenders are more likely to be arrested than their white co-offending partners. This is actually a plausible hypothesis, as there are large black-white differences in cognitive ability and impulsivity, and there is some evidence suggesting that criminals with higher IQs are more likely to evade arrest (Boccio et al. 2018).

A second possible limitation is that the study doesn’t control for events after the offense. For example, black offenders may be more likely than their white co-offending partners to interact with the police after the offense, which puts them in more situations where they could be arrested. Again, this is a plausible hypothesis given that blacks tend to have more police encounters. A third possible limitation is that the sample of co-offending crimes may differ from crimes in general in unknown ways, which would limit the generalizability of the findings.

Any of these limitations could plausibly account for the small association between race and arrest, where black offenders had just 3% higher odds of arrest than their white co-offending partners for violent offenses. But even if we assume that this gap was purely due to race per se, this is nevertheless a very tiny gap, especially given that blacks are over 3 times as likely to be arrested for a violent crime compared to whites. Even if we assume that this figure properly estimates arrest bias against blacks, one can only conclude that arrest bias explains only a trivial fraction of the overall gap in arrest rates between blacks and whites.

Conclusion: The Parity Hypothesis is true


Let us review the findings in this post:

  • The FBI-reported homicide arrest rates by race is corroborated by the homicide victimization rates reported by both the FBI and the CDC. The FBI homicide arrest data show that blacks account for about 50% of arrestees for homicide. The CDC and FBI homicide victimization data also show that blacks constitute about 50% of homicide victims.
  • The FBI-reported assault arrest rates by race is corroborated by the rates of hospitalizations and ED visits due to assault by race. The FBI assault arrest data show that blacks account for 31 to 33% of arrestees for assault, with an arrest rate that is 2.7 to 3.0 times the rate for whites (including Hispanics). The studies on hospitalizations and ED visits consistently find that black patients represent an even greater share of patients injured through assault, especially for firearm injuries where black patients often had more assault injuries than patients of any other race. The findings regarding Hispanics were a bit mixed. Hispanics were consistently overrepresented as victims of firearm assault injuries, but they were sometimes not overrepresented when including non-firearm assault injuries.
  • The respondents to the National Crime Victimization Survey (NCVS) reported rates of offending by race that perfectly lined up with FBI-reported rates of arrest by race. That is, black overrepresentation in FBI-reported arrestees nearly perfectly matches black overrepresentation in victim-reported offenders.
  • According to the arrest and police report data collected by the NIBRS, black offenders have lower arrest rates than white offenders for robbery and assault offenses, whereas racial differences in arrest rates are small and often statistically insignificant for homicide and sex offenses. When comparing total arrests and total reported offenses in the 2019 NIBRS data, the black share of arrestees nearly perfectly matched the black share of reported offenders for each crime type.

All of these findings are naturally predicted by the Parity hypothesis. If the Parity hypothesis were true, i.e. if true crime rates by race reflected the FBI-reported arrest rates by race, then we would expect the black share of homicide victims to reflect the black share of homicide arrestees. We also would expect the black share of assault injury victims to reflect the black share of assault arrestees. We would also expect the black share of victim-reported offenders to reflect the black share of arrestees for each crime type. Finally, we also would expect the black share of offenders reported to police to reflect the black share of arrestees for each crime type.

Because each of these predictions has been confirmed above, this constitutes sufficient evidence to believe that the Parity Hypothesis is true. Of course, this is provisional; if the evidence above is invalidated or overridden by counterevidence, then we should give more credence to the Bias Hypothesis.

A supporter of the Bias Hypothesis, i.e. one who thinks that the black-white differences in FBI-reported arrest rates are greater than the true black-white differences in crime rate, might object that each of these findings are actually compatible with the Bias Hypothesis if we just posit a few auxiliary hypotheses. For example, a defender of the Bias Hypothesis might try to explain all of the above findings as follows:

  • To explain the congruence between the arrest data and the victim data by the NCVS, one might posit that survey respondents give biased responses when reporting the race of their offender.
  • To explain the congruence between the arrest data and the police report data, one might posit that police reports are biased against black offenders. For example, perhaps crimes by black offenders are more likely to be reported to police. Sure, the NCVS suggests only small differences in rates of police notification by offender race, but perhaps the NCVS respondents give biased responses when reporting whether they notified police of their victimization (this would require two separate biases working together: people over-report black offenders to the police, but this is not shown in the NCVS data because the NCVS respondents under-report the rates that they report black offenders to the police).
  • To explain the congruence between the arrest data and the assault injury victimization data, one might posit that injuries that require hospitalization and/or ED visits are very severe visits and may not be representative of assaults in general. Perhaps blacks are really overrepresented in these very severe assaults, but they are much less overrepresented in less severe assaults, which implies that the total black assault rates are much lower than the FBI-reported arrest rates would suggest. Or one might posit that victimization rates by race is not a reliable indicator of offending rates by race (see next point).
  • To explain the congruence between the arrest data and the homicide victimization data, one might posit that victimization rates by race are not a reliable indicator of offending rates by race. For example, perhaps black homicide victims are much more likely to be killed interracially than are non-black homicide victims. Sure, the arrest data and police report data suggest that homicides are almost entirely intra-racial (and, if anything, black offenders are more likely to offend interracially than are non-black offenders). But all of this data could be biased as well.

In response, I must first concede that each of these auxiliary hypotheses is possible. However, the conjunction of each of these auxiliary hypotheses is less probable than the Parity Hypothesis. Why? Because the Parity Hypothesis gives a much more parsimonious explanation of each of the findings. In order to explain the findings above, the Parity Hypothesis only needs to assume that blacks have higher crime rates than whites. That assumption is sufficient to predict each of the above findings without making any additional unnatural assumptions. By “unnatural” assumption, I mean an assumption that we wouldn’t have already posited prior to reviewing the evidence above.

On the other hand, the Bias Hypothesis must stipulate many unnatural assumptions, i.e. assumptions we would have no reason to posit unless we were already committed to trying to make the Bias Hypothesis explain the aforementioned data. For example, unless someone was already committed to explain the above data with the Bias Hypothesis, no one would have assumed that black homicide victims are more likely to be killed interracially than non-black homicide victims. Likewise, no one would have assumed that NCVS respondents gave racially biased responses when reporting whether they issued a police notification for a crime.

More importantly, the additional unnatural assumptions required by the Bias Hypothesis are all independent. For example, the assumption that the police report data is biased against blacks seems unrelated to the assumption that black homicide victims are more likely to be killed interracially than are non-black homicide victims. For another example, the assumption that people over-report black offenders to the police seems unrelated to the assumption that the NCVS respondents under-report the rates that they report black offenders to the police. Therefore, the Bias Hypothesis cannot just posit a few unnatural assumptions to explain the above findings; instead, it must posit many independent, unnatural, assumptions to explain the findings. All else equal, if a theory requires more independent, unnatural, assumptions to explain the data, then the theory is less likely to be true because there are more ways for the theory to be false.

Leave a Reply