Comparing the crime-race and crime-SES relationships

If racial differences in criminality were the result of racial differences in SES, then one might expect the association between crime and SES to be stronger than the association between crime and race. This pattern is not detected when we investigate crime aggregated at the county, city, and state level. For example, across states, the crime-black association is consistently found to be stronger than crime-SES association:

• Rubenstein (2005) [archived] measures how the violent crime rate in a state (or Washington D.C.) correlates with various variables such as poverty, unemployment, high school completion rate, and percent of the population that is black/Hispanic. The finding was that the correlation between violent crime rate and percent black/Hispanic (r=0.81) is far larger than the correlation between violent crime rate and poverty (r=0.36), unemployment (r=0.35), and percent that has not completed high school (r=0.37) (Figures 14-17).
• Templer and Rushton (2011) [archived] report similar findings when investigating crime rates across the 50 states of the country. They find that the correlation between the percent black and murder rate is extremely large (r=0.81), larger than the correlation between murder and IQ (r=0.54), life expectancy (r=-0.61), or income (r=-0.40).

The classic work on structural correlates on crime is Land, McCall, and Cohen (1990) [archived]. Researchers reviewed 44 models across 21 studies that investigated the relationship between homicide and various covariates (an update to the paper – McCall et al. (2010) [archived] – was published 20 years later). Here are some of the main findings:

• Percentage black was associated with homicide more consistently than any other covariate. Table 1 shows that: 86% (38/44) of the models found a significant positive association between homicide and percent black. For comparison, 69% (22/32) of the models found a significant positive association with poverty, 29% (8/28) found a significant positive association with income inequality, 20% (1/5) found a significant negative association with median family income, and 0% (0/9) found a significant positive association with unemployment.
• In 1980, percentage black had a stronger effect on a city’s or metropolitan area’s homicide rate than any other variable (Table 2). The other variables were percentage divorced, percentage of kids not with both parents, median family income, percentage families below poverty, Gini index, and unemployment rate.

The authors of the study then created their own regression modal that combined various elements into a single variable: median family income, the percentage of families in poverty, income inequality, percentage of the population that is black, and the percentage of children not living with both parents. This new variable was called the resource deprivation/affluence component. Unsurprisingly, the resource deprivation/affluence variable had the strongest association with homicides compared to all other variables between 1960 and 1980 (Table 3). Interestingly, the researchers note that “among these elements, the single element that is most consistently (although not invariably, as documented earlier in tables 1 and 2) associated with homicide rates is the percentage of the population of a city, metropolitan area, or state that is black” (page 954).

Kposowa, Breault, and Harrison (1995) found similar results when investigating covariates with crime at the county level in 1980. The study used regression analysis to estimate the association between homicide and a variety of covariates for large counties (population > 100,000). The results showed that the relationship between homicides and percentage black (β=0.739) was far stronger than the association between homicide and poverty (β=-0.119), Gini index (β=0.177), divorce (β=0.187), density (β=0.126), unemployment (β=-0.026), and education (β=-0.027) (Table I). The variable with the second largest association with homicide was percent Hispanic (β=0.265). Table II shows that percent black was also the strongest predictor of homicides in small counties (population < 25,000). Finally, when considering all counties together, percent black was found to be the strongest predictor of homicides ( Table V) and violent crime (Table IV), but not property crimes (Table III).

Finally, Pratt and Cullen (2005) [archived] meta-analyzed 162 studies that investigated the relationship between 34 variables and a region’s crime rate. Some key results are as follows:

• 72% of studies found a significant positive association between percent black and crime, with an average effect size of .294. Among the variables with at least 20 studies, only 2 variables had a stronger effect size (incarceration effect and percent nonwhite which had effect sizes of –0.332 and 0.328, respectively), and only 3 variables were found to have a significant effect in a larger proportion of studies (racial heterogeneity, percent non-white, and social support/altruism were found to have a significant effect in 72.83%, 76.38%, and 74.47% of studies). Percent black had a larger average effect size and was more consistently significantly associated with crime than poverty, inequality, socioeconomic status, and unemployment.
• The authors also adjusted the mean effect size estimates of each variable for independence correction. The variables with the largest independence-adjusted mean effect size estimates were ranked (Table 1). Percent black was ranked 7th out of 31 variables (independence-adjusted mean effect size of 0.294), ranked above poverty, inequality, urbanism, and socioeconomic status. Also, of the 6 variables ranked above percent black, only two variables had more than 20 studies (incarceration effect and percent nonwhite had independence-adjusted mean effect sizes of 0.330 and –0.317, respectively).

The above studies do not prove that black criminality is not ultimately the result of low-SES, but they do provide reasons to be skeptical of any such claims. To make more decisive claims about whether black criminality is the result of SES, we should consider studies that estimate racial differences in criminality after controlling for SES. If controlling for SES does not eliminate a significant portion of the gap in criminality, then we can conclude that SES differences do not cause a significant portion of the racial differences in criminality.

Controlling for family SES

I only know of two studies that measure racial differences in criminality after controlling for family SES. Unfortunately, the studies only control for one component of SES each – the first study controls for family wealth whereas the second study controls for family income. Unfortunately, the studies don’t directly measure differences in criminality. Rather, they measured racial differences in incarceration (although incarceration rates are a good indicator of criminality, as I argued in a separate post).

The first study is Zaw et al. (2016) [archived] where researchers examined the impact of race and wealth on the probability of being incarcerated. The study reveals that black males at every wealth bracket (except for the top wealth bracket) are more likely to be incarcerated than white males at every other wealth bracket (Figure 2b). The study notes that the exceptional finding for the wealthiest blacks is “inconclusive” given the small sample sizes (page 108). The study reports found that “at low levels of wealth both blacks and Hispanics still had a higher incarceration rate than whites. At higher levels of wealth at the baseline, although the black-white incarceration disparity was reduced for males, it was not eliminated” (page 112).

These results were corroborated by the second study Chetty et al. (2018) [archived] which examined the existence and magnitude of racial and gender differences conditional on parental income. Researchers find that black men are substantially more likely to be incarcerated than white men, even after controlling for parental income. For example, among children with families in the lowest income bracket (bottom 1%) on April 1, 2020, 21% of black males were incarcerated whereas only 6.4% of white males were incarcerated. Furthermore, among children with families in the top 1%, 2.2% of black males were incarcerated whereas only 0.2% of white males were incarcerated (page 23). Black males are several times as likely as white males to be incarcerated at every level of family income (see Figure VII).

Controlling for neighborhood SES

One criticism of some of the above studies is that they focus on the wrong unit of analysis. Either the unit of analysis is too broad (e.g. data is aggregated at the city or state level) or too narrow (e.g. data is aggregated at the individual or household level). One issue with these units of analysis is that blacks might find themselves in worse neighborhoods even after controlling for city SES or household SES. For example, a 2010 report [archived] by Brown University shows that affluent blacks live in neighborhoods with higher poverty rates than the neighborhoods in which poor whites live (Table 2). If blacks find themselves concentrated in disadvantaged neighborhoods regardless of city/state SES or individual/household SES, then controlling for SES at levels of analysis that are too broad/narrow might not control for the relevant environments.

Some studies have overcome these issues by analyzing data aggregated at the neighborhood or block level. I will review such studies below (based on the data that I’ve found). As I’ll show below, the results of these studies are mixed. Some studies show that the relationship between race and criminality is completely or mostly accounted for (statistically) after controlling for neighborhood SES. Others find that the majority of the relationship persists even after accounting for these controls.

Studies

Levitt (1999) [archived] used homicide data between 1965 and 1995 to explore how income inequality affects criminal victimization patterns in over 70 Chicago neighborhoods. While the primary purpose of the study was not to investigate the association race and crime, the data in the study provides useful information on this topic.

• Results: Significant racial differences in homicide victimization were found even after controlling for neighborhood median family income. During 1986-1995 (the latest time period available), the homicide rate for black residents in neighborhoods with the highest incomes was about 3 times higher than the homicide rate for white residents in neighborhoods at all income levels (Table 6). Levitt notes that “even the most dangerous neighborhoods for whites experienced homicide rates of only about 10 per 100,000, about one-fourth the median homicide rate among blacks” (page 92). A regression analysis was performed to estimate the impact of each variable of interest (median family income, percentage black, female-headed household rate, etc.) on homicide rates after controlling for every other variable (Table 7). As expected, these controls reduce the relationship between percentage black and homicide. However, interpreting the data any further is difficult because the table does not indicate the units of the values nor which values are statistically significant.

Shihadeh and Shrum (2004) examined the relationship between SES, race, and crime in different block groups in Baton Rouge, Louisiana. The sample included the 276 block groups located within the incorporated city limits. Analysis was based on 1990 census data and arrest records during 1989–1991.

• Results: After controlling only for basic demographic controls, percent black was a strong predictor of serious crimes. Specifically, the authors note that “racial composition of neighborhoods is the strongest and most consistent predictor of serious crime rates, with the exception of rape and larceny theft. For violent crimes in particular, a one standard deviation rise in the proportion Black is associated with about a one-third standard deviation rise in the rates of crimes (βs for completed homicide = 0.289, attempted homicide = .284, aggravated assault = 0.311).” (page 522). However, controlling for economic disadvantage, social disorganization, and other background characteristics “eliminates the effect of race on crime rates” (page 524).
• Controls: Demographic controls were population size, population density, and age structure (proportion of all residents aged 15 to 24). Social disorganization was measured based on vacant household rate and rate of recent movers. Family structure is calculated as the proportion of all families that are headed by a female with no husband present. Economic deprivation is a composite index composed of poverty rate, income inequality, and unemployment rate.

Gjelsvik, Zierler, and Blume (2004) [archived] investigated the risk of homicide victimization of men by block group-level SES characteristics in Massachusetts and Rhode Island. The study comprised all white, black, and Hispanic 15- to 44-year-old men (N=1,593,256) included in the 1990 US Census as Rhode Island or Massachusetts residents. Researchers estimated the differences in risk of homicide victimization by race by block group-level SES characteristics and the effects of race on homicide risk after block-group-level conditions were taken into account.

• Basic results: The risk for homicide of black men is over 10 times higher than the risk for white men among nearly every SES bracket (Table 1). As the authors note, “except for proportion of female-headed households, white men had not only a substantially lower risk of homicide compared to black and Hispanic men within the same block-group socioeconomic level, but also the highest risk for white men (over all socioeconomic levels) was still substantially lower than the lowest risk of homicide for black and Hispanic men (over all socioeconomic levels) (Table 1)” (page 706).
• Regression analysis: Researchers constructed a regression model involving race, age, and each of their SES measures to estimate the effect of each variable after controlling for every other variable. They then estimated the relative change in homicide risk associated with a 10% increase in each independent variable (e.g. percent black, percent Hispanic, poverty rate, percent female-headed households, etc.) (Table 3, Model B). The results indicated that a 10% increase in percent black or percent Hispanic was associated with a far greater increase in homicide risk than a 10% increase in any other variable. For example, a 10% increase in percent black was associated with a 440% increase in homicide risk, whereas a 10% increase in female-headed households was associated with only a 35% increase in homicide risk (Table 3, Model B).
• Controls: In their full regression model, researchers measured the effects of percent black, percent Hispanic, age, and the following 7 measures of socioeconomic status: household poverty rate, home ownership rate, higher educational attainment rate, high school completion rate, unemployment rate, female-headed household rate, and racial composition (ratio of black residents compared to white residents in each block-group) (page 704).

Sampson, Morenoff, and Raudenbush (2005) [archived] analyzed various individual, family, and neighborhood factors related to racial gaps in violent crime from 1995 to 2002. The study investigated 3 waves of data on 2974 participants aged 8 to 25 years living in 180 Chicago neighborhoods. The study aimed to study the cause of racial differences in individual violent crime rates and neighborhood violent crime rates.

• Individual violence: Prior to controlling for any covariates, the odds ratio for the black:white violent crime rate was 1.85 (page 228), meaning that blacks were 85% more likely than whites to engage in violence. After controlling for gender, age, and immigration status, the odds ratio dropped to 1.7 (page 229). Controlling for family structure, SES, and number of years at the same address reduced the ratio to 1.49 (page 229), accounting for only 30% of the gap (0.21 out of 0.7) that remained after controlling for gender, age, and immigration status. Surprisingly, the study found that SES was not associated with violence: “family socioeconomic status is not directly associated with violence. What matters instead are years of residence in the neighborhood and having married parents, both of which are protective” (page 229). Further controls for verbal/reading ability, impulsivity/hyperactivity, and neighborhood racial demographics further reduced the ratio to 1.28 (page 230).
• Neighborhood violence: Prior to controlling for any covariates, there was a significant correlation between the violent crime rate in a neighborhood and the percentage of the neighborhood that was black. This correlation was rendered statistically insignificant after controlling for immigrant concentration, percentage professional/managerial, concentrated disadvantage, and residential stability (page 230). Contrary to many other studies, concentrated disadvantage was not a significant predictor of violence (page 230).
• Controls: Socioeconomic status (SES) was a composite of parent’s income, education, and occupational status. Verbal/reading ability was measured using Wechsler Intelligence Scale for Children vocabulary test  for the 9- to 15-year-olds and using the Wechsler Adult Intelligence Scale vocabulary test for the 18-year-old cohort. Impulsivity/hyperactivity was measured based on reports by the primary caregiver for the 9- to 15-years-olds and based on self-reports for the 18-year-old cohort. Concentrated disadvantage was an average of percentage of poor families, percentage of single-parent families, percentage of families on welfare, and the unemployment rate in the neighborhood.

Jones-Webb and Wall (2008) [archived] investigated how well various neighborhood characteristics can explain racial gaps in homicide in 10 U.S. cities between between 2003 and 2005. The sample includes 3,915 census block groups in in Oakland, San Francisco, Santa Ana, St. Paul, Minneapolis, Atlanta, Baltimore, Boston, Kansas City, KS, and Kansas City, MO. Each census block consists of on average of about 400 households or about 1,200 individuals and is roughly 1 to 2 square miles.

• Results: After adjusting for basic background variables, percent black was significantly associated with homicide. Specifically, a 10% increase in percent black was associated with a 23% increase in homicide risk (Table 3, separate predictors). After controlling other covariates with homicide, a 10% increase in precent black was only associated with a 13% increase in homicide (Table 3, model 2), “indicating an attenuation effect of 43%.” Similar patterns were found for percent Hispanic. Interestingly, percent Asian was consistently associated with fewer homicides, before and after controlling for covariates.
• Controls: Background variables were year, spatial neighbor mean homicide, proportion of males aged 14 to 24 years of age, and area in square miles, and a fixed effect for city. Social disadvantage variables were unemployment rate, percent persons with less than a high school education, median household income, and percent female head of household.

Peterson and Krivo (2009) [archived] analyzed police-reported crime counts for 8,286 neighborhoods across U.S. 87 cities to investigate how the association between race and crime is affected by various structural controls. The focus was on the number of violent crimes (homicides and robberies) reported to the police averaged over three years (1999–2001).

• Results: After controlling for only age and sex structure and city characteristics alone, there are large racial differences in violent crime rates. After these controls, the violent crime rate for black neighborhoods (70% or more black) was 4.29 times the rate for white neighborhoods (Table 2). After further controlling for a variety of neighborhood characteristics – residential instability, residential loans, immigration, and socioeconomic disadvantage (see “Tract Level” independent variables in Table 1) – the violent crime rate for black neighborhoods was only 1.78 times the rate for white neighborhoods (Table 2), a 76% reduction.
• Controls: Residential instability is measured by a combination of “the percent of renter-occupied units and the percent of residents aged five or older who lived in a different dwelling in 1995” (page 917). Residential loans are measured as the amount of home mortgages originated in the census tract in 2000. Socioeconomic disadvantage is measured as “the extent of joblessness, professional or managerial occupations, high school graduates, female-headed families, secondary sector workers (“those in the six occupations with the lowest average incomes”), and poverty (page 917). City characteristics (see “City level” independent variables in Table 2) include “racial residential segregation, socioeconomic disadvantage, manufacturing jobs, population size, percent non-Latino black, percent of recent movers, percent foreign born, percent young males, and region” (page 927).

Feldmeyer, Steffensmeier, and Ulmer (2013) [archived] analyzed the relationship between violent crime, race, and other covariates in 479 census places in California and New York. Violence measures are calculated using 5-year averaged arrest figures for 1998–2002. The researchers created two regression models to investigate the relationship between violent crime and a variety of independent variables such as percent black, percent latino and a robust set of structural controls. Model 1 included all structural controls except for their structural disadvantage index (defined below). Model 2 was their full model which included all controls.

• Basic results: Model 1 “reveals that these race/ethnic composition effects are consistently among the strongest predictors of violence, especially for the more serious offenses of homicide (B = 0.34 for %Black; B = 0.49 for %Latino) and robbery (B = 0.37 for %Black; B = 0.44 for %Latino)”. In fact, percent black and percent latino were the strongest predictors of homicide and robbery in this model (Table III). After controlling for structural disadvantage (model 2), the association between percent black and violence reduced, but only partially; in fact, for certain crimes (homicide and robbery), the majority of the association remained (Table III). As the authors note, “the percent black effects are reduced by about one-third for homicide (Model 1, B = 0.34; Model 2, B = 0.23) and robbery (Model 1, B = 0.37; Model 2, B = 0.26).” Additionally, in their full model, percent black was the second strongest predictor (after their structural disadvantage index) of homicide (B = 0.23) and robbery (B = 0.26), even after accounting for the other structural controls.
• Results by census place size: The authors also performed a similar analysis disaggregated by census place size. The census places were segregated into small (10,000–24,999 residents), medium (25,000–49,999 residents), and large (50,000+ residents) census places. The authors also find that, for small census places, “racial/ethnic composition effects on violence are largely nonsignificant after accounting for concentrated disadvantage, residential mobility, and other structural and demographic characteristics of census places.” However, while structural controls can account for the association between race and crime in smaller areas, this is not true for larger areas. Specifically, the authors report that their findings “provide consistent evidence that black and to a lesser extent Latino population concentration in larger places is linked to higher levels of violence, and that these effects cannot be fully accounted for with structural controls.”
• Controls: Structural disadvantage is a composite index measured in terms of poverty rate, unemployment rate, education (percent residents over 25 without a high school degree or equivalent), and family structure (percent female-headed families with children under 18 years old). Residential mobility is measured as the percentage of households that experienced housing turnover during 1995–2000. Immigration measures the percentage of census place residents that are foreign born and arrived in the United States between 1990 and 2000. There were two measures of racial segregation – black–white segregation and Latino–white segregation using Index of Dissimilarity (D) measures of racial/ethnic residential evenness. Finally, there were controls for the following demographic factors – population density and size of census place, young male population, state, and police per capita. In the full regression model (model 2), the only statistically significant independent variables (p<.05) were (ranked in descending order of standardized regression coefficient) structural disadvantage, percent black, percent Latino, total population size, black-white segregation, young male population, and police per capita (Table III).

Takeaways

I have two main takeaways from the data considered here:

1. Controlling for a variety of neighborhood structural factors appears to statistically account for a significant portion of the black-white gap in criminality. The factors most significant in explaining the gap across several studies appear to be: female-headed households, income (or poverty), educational attainment, income inequality, and various measures of occupation (e.g. employment rate and occupational prestige).
2. The data is mixed regarding the degree to which these factors can account for the black-white gap in criminality. Some studies and can account for the majority (Peterson 2009) or even the entirety (Shihadeh 2004) of gaps in criminality after adjusting for a variety of neighborhood structural controls. However, other studies (e.g., Feldmeyer 2013, Jones-Webb 2008, and Gjelsvik, Zierler, and Blume 2004) find that the majority of the gap for certain serious crimes (e.g., homicide and robbery) persists even after controlling for these factors.

More data is needed for clarity on the precise magnitude of the gap that is eliminated after adjusting for these controls. My hunch is that these structural controls probably can (statistically) account for some portion of the racial gap in crime, but they cannot account for the entire gap. Further, as the results of Feldmeyer et al. (2013) demonstrate, the precise portion of the gap that can be explained probably depends on a variety of factors, such as the type of crime, the region, etc.

Possible confounders?

The above studies demonstrate that various neighborhood-level variables can (statistically) account for some portion of the black-white gap in criminality (the precise portion of the gap cannot be determined). However, this does not imply that these variables cause that portion of the black-white gap in criminality. This merely establishes a correlation between racial differences in these SES variables and racial differences in crime, but such correlation is not sufficient to infer causation. This is because the correlation between racial differences in SES and racial differences in crime may be the result of a spurious, rather than causal, relationship. There are plenty of possible explanations for the relationship between racial differences in SES and racial differences in crime. The hypothesis that SES differences cause crime differences is just one possible explanation. In order to infer causation, we need to rule out possible alternative explanations.

The simplest alternative explanation is that there is an omitted confounder that is explaining the relationship between racial differences in SES and racial differences in crime. This confounder might be responsible for both low SES and high crime among the black population. If so, then controlling for SES would implicitly controls for that confounder, which means the consequent reduction in crime gaps would be due to controlling for the confounder (rather than controlling for SES).

Genetic confounders

One possible confounder is racial genetic differences. Genetic differences might be responsible for both racial differences in crime and racial differences in SES (e.g., if the genetic differences were responsible for differences in important psychological traits related to both outcomes such as intelligence, self-control, etc.). Because the above studies made no effort to control for genetics, their outcomes are not inconsistent with a genetic explanation of racial gaps in criminality.

This same criticism is appropriate when investigating the causes of black-white differences in IQ. Some have argued that black-white differences in IQ are partially due to black-white differences in SES on the basis that the IQ differences are reduced after controlling for SES differences. Hereditarians have named this mistaken inference the “sociologist’s fallacy.” Prominent hereditarians Rushton and Jensen (2005) have explicitly noted that such reductions are compatible with genetic explanations (page 267):

The most frequently stated culture-only hypothesis is that the mean IQ differences are due to SES. In fact, controlling for SES only reduces the mean Black–White group difference in IQ by about a third, around 5 IQ points. The genetic perspective does not regard this control for SES as being entirely environmental. It holds that the parents’ socioeconomic level in part reflects their genetic differences in intelligence.

This point has also be recognized by environmentalists as well. For example, Nisbett et al. (2012) [archived] acknowledged that “there is no way of knowing how much of the IQ advantage for children with excellent environments is due to the environments per se and how much is due to the genes that parents creating those environments pass along to their children” (page 7). Also, Neisser et al. (1996) [archived] have acknowledged that “The living conditions of children result in part from the accomplishments of their parents: If the skills measured by psychometric tests actually matter for those accomplishments, intelligence is affecting SES rather than the other way around” (page 95) to critique the hypothesis that SES differences explain black-white IQ differences.

What was said about IQ differences can also be said about crime differences. Just as we cannot assume causation from a correlation between racial IQ gaps and racial SES gaps, we also cannot assume causation from a correlation between racial crime gaps and racial SES gaps. Genetic confounding may be playing a role in both cases. In fact, Barnes et al. (2014) have criticized criminological studies for failing to control for genetic confounding. They argue that much criminological research rely on “the standard social science method (SSSM)” which “does not allow the researcher to account for genetic influences” (page 472). This criticism is rather damning since “the available evidence indicates that genetic influences account for about half of the variance in antisocial behavior” (page 473).

Other confounders

Genetic differences are just one possible explanation of the relationship between racial differences in SES and racial differences in crime. There may be other traits that can explain the relationship without necessarily depending on genetics. Consider the following examples:

• IQ. We know that (1) IQ predicts future SES, (2) IQ predicts future criminality, and (3) there are significant IQ differences between blacks and whites (see my previous post). Therefore, it is plausible that the IQ gap explains the relationship between the SES gap and the crime gap. That is, by controlling for SES, one also implicitly controls for IQ, which is what explains the reduction in the crime gap.
• Self-control. We know that (1) self-control predicts future SES, (2) self-control predicts future criminality, and (3) there are significant differences in self-control between blacks and whites (see previous posts here and here). Therefore, it is plausible that the self-control gap explains the relationship between the SES gap and the crime gap. That is, by controlling for SES, one also implicitly controls for self-control, which is what explains the reduction in the crime gap.

So there are plausible alternative explanations for the relationship between racial differences in crime and racial differences in SES. One plausible explanation is that genetic differences account for the relationship. Another plausible explanation is that other psychological differences (e.g. differences in IQ or self-control) account for the relationship. In other words, it may be that individuals with the traits (whether genetic or environment) necessary to avoid criminality also tend to find themselves in high-SES neighborhoods (and vice-versa), rather than the high-SES neighborhoods causing low criminality in its inhabitants. Until these alternative explanations are ruled out, we cannot infer that racial differences in SES are the cause of racial differences in crime. None of the above studies have sufficiently ruled out alternative explanations. Therefore, we are in no position to conclude that racial differences in SES are in any way causal of racial differences in crime.

A defense of alternative explanations

As may be obvious from my post, I accept an alternative explanation of the data from the above studies. Apart from family structure, I do not believe that racial differences in SES play a large role in explaining racial differences in criminality. Instead, I believe racial differences in certain psychological traits – namely, differences in cognitive ability and self-control – are responsible for the statistical correlation between racial differences in SES and racial differences in criminality. When one controls for SES, they implicitly control for cognitive ability and self-control, which explains the reduction in the crime gap. In other words, I believe the above data is best explained by the hypothesis that individuals prone to criminality find themselves in poor neighborhoods (or better, they create poor neighborhoods) rather than by the standard sociological hypothesis that poor neighborhoods produce individuals prone to criminality.

I will provide evidence for these claims in a future post. For now, I’ll just explain why I believe my explanation has some advantages over the standard sociological explanation.

1. The greater levels of black criminality relative to other low-SES groups (e.g. Hispanics and Native Americans in the U.S., Bangladeshi and Pakistani in the U.K., certain Asian groups in Canada, etc.) cannot be explained by racial differences in SES (see my previous post), but they may be explainable by racial differences in cognitive ability and self-control.
2. One needs an explanation as to why blacks consistently find themselves in low-SES neighborhoods and households. This finding can be perfectly explained by differences in cognitive ability (see my post here).

Again, when I say SES, I’m excluding family structure (i.e. differences in female-headed households). I exclude family structure because the above two criticisms do not apply: (1) blacks have much higher rates of female-headed households, even compared to other low-SES groups, and (2) high rates of female-headed households cannot be explained by differences in cognitive ability. The remaining components of SES are rough measures of economic resources (e.g. poverty, income, inequality, unemployment, education, etc.). These components of SES, I believe, do not account for a large role of the racial gap in criminality. I plan to substantiate my position here in a future post.