Percent Black and Crime

An interesting predictor of local crime rates is the percent of the region examined that is black. Some may point to this and say that this is a result of discrimination. This is unlikely, as Rubenstein (2016) finds “Victim and witness surveys show that police arrest violent criminals in close proportion to the rates at which criminals of different races commit violent crimes.” Other reasons could include socioeconomic standards in the neighborhood, race differences in IQ, and underlying genetic differences in crime-committing-variables (aggression, ASB, etc.). Whatever the cause, the evidence for the association is documented in this blog post.

The most well known study on percent black and crime, perhaps the foundation of all modern research of the topic, is Land et al.’s (1990) paper Structural Covariates of Homicide Rates: Are There Any Invariances Across Time and Social Space? In this paper, they review the existing literature from 1960 through 1980; they find that 86% (38/44) found positive, significant correlations. At least at the time of Land et al.s paper, this was a very replicable finding. According to modern evidence, this has held up.

Worrall (2008) also summarize previous longitudinal findings, which appear to be much more mixed in outcomes:

They argue that this is because of statistical issues rather than a non-relationship. For example, these studies suffer from fixed effects preventing a visible relationship. Beck (2001) states that, “Although we can estimate [a model] with slowly changing independent variables, the fixed effect will soak up most of the explanatory power of these slowly changing variables. Thus, if a variable … changes over time, but slowly, the fixed effects will make it hard for such variables to appear either substantively or statistically significant.”

Another issue described by Worrall is that of collinearity in the primary variables. This is admitted to be the reason some other studies will find null effects, such as that of Messner and Tardiff (1986). As they state,

“A possible explanation for this pattern of “null effects” is that it reflects large standard errors generated by high levels of multicollinearity (Hanushek and Jackson, 1977: 86-93). The simple correlation between the poverty measure and the measure of racial composition is .72.7 Neighborhoods with large black populations tend, not surprisingly, to be neighborhoods with relatively large numbers of very poor residents”

– (Messner and Tardiff, 1986)

Worrall tests the percent black-crime association with other statistical models in order to counter these effects and consistently finds significant, positive relationships. The data is shown in the tables below.

Percent black was not significantly related to aggravated assault and burglary, but had correlations with robbery and murder at 0.61 and 0.57 respectively in an analysis from Bartels et al. (2010). Block (1979) finds a correlation of 0.69 between percent black of an area and the homicide rate.

Beasley and Antunes (1974) and Mladenka and Hill (1976) both found very strong correlations between percent black of a neighborhood and the reported violent crime rate across 20 police districts. The former found a correlation of 0.74, whereas the latter found a correlation of 0.81. In a study of Iowa, across all analyses, percent black had a strong, positive, significant correlation with total crime rate (Carrothers, 2016).

Rubenstein (2005) finds a very strong correlation (r=0.81) between percent black and Hispanic of a state and the violent crime rate of the state. Conversely, the association between poverty and crime and unemployment and crime are 0.36 and 0.35 respectively.

Unz (2013) is much more optimistic about the relationship between poverty and crime. However, he finds the relationship between percent black and crime is still stronger in the 0.8-0.9 range. He also finds this relationship has been strengthening over the past few years.

Stuart and Taylor (2019) review the data from 1920s through the 1960s and finds a consistent relationship between percent black and crime across that period of time.

Whitney (1995) also cites existing evidence of the relationship between percent black and crime rate. He states,

We can do a pretty good job of predicting differential murder rates, simply by considering racial composition of the population. For example, in the fourth figure we have aggregate data across the 50 states of the United States. The simple correlation between murder rate and percent of the population that is black, is r= +0.77. For Figures 4 and 5, the homicide data are from the U .S. Department of Justice (1981), while the population percentages are from the 1980 census (Race, 1981). I know of no environmental variable that accounts for more of the variation. Rather than the 50 states, we can look at all of the 170 cities in the United States that had a 1980 population of at least 100,000. With 170 data points, it would make a messy scatter-plot; the overall correlation between murder rate and percent of the population which is black is r= +0.69 (Kleck & Patterson, 1993; Kleck, 1995).

Simply for illustrative purposes, the fifth figure is the rate-by-state as in figure 4, but with the values for Washington, DC included. As you can see, the very high murder rate for Washington, DC is simply what one would predict, given knowledge of its population composition.

– (Whitney, 1995)

The figures presented by Whitney are below:

Some opposing evidence exists. For example, Arvanites and Defina (2006) find a very small, negative relationship between percent white and crime rates, all of which were statistically insignificant (or approaching significance). This may be due to the metric used – percent white. A lower percent white may also mean a greater proportion Asian (whose crime rates are lower) or Hispanic (whose higher crime rates are debatable), rather than a greater proportion of black people present. As discussed earlier, this can be to fixed effects dampening the relationship as well.

Feldmeyer et al., (2013) attempts to find various confounding variables in the percent black-crime association, however the relationship persisted beyond these variables. Before controlling for confounding, they found a correlation of 0.438 between homicide rate and percent black, 0.513 between robbery rate and percent black, 0.304 between rape rate and percent black, and 0.199 between aggravated assault rate and percent black.

In their reduced models, they controlled for demographic variables and then structural disadvantage and they find these had “small to moderate effect on violence rates.” The relationship between percent black and the various crimes were slightly mediated but nevertheless persisted. To the dismay of any pushing a theory of socioeconomic disadvantage, Feldmeyer et al. state,

“As we examine next, however, this is only partly the case. Table III clearly shows that measures of structural disadvantage and social-demographic characteristics of census place explain away some but not nearly all of the effects of race/ethnic composition on place-level rates of violence. Model 2 (Table III) shows that the violence-generating effects of percent black and percent Latino, though reduced, remain significant and are remarkably persistent across all four violence measures, even after accounting for structural conditions of locales. In fact, only one effect (percent Latino on rape) is reduced to nonsignificance after introducing the structural disadvantage index.”

– (Feldmeyer et al., 2013)

However, they go on to find that place of living makes a much larger difference in the relationship. The relationship was substantially reduced or entirely eliminated after controlling for size of place (e.i. urban vs. rural). Essentially, the relationship between percent black and crime only really persisted in urban areas.

Shihadeh and Shrum (2004) find a large, significant effect of % black on crime rates, but this became statistically insignificant when controlling for socioeconomic, structural, and demographic variables.

However, one should also ask why these neighborhoods are poor? Shihadeh and Shrum unfortunately do not bother to leave this up to interpretation. They rather assume that socioeconomic disadvantage is something which just occurs among black neighborhoods, saying,

“Our results show that the association between block group racial composition and crime rates is due to an underlying association between serious crime and structural factors that are often implicated in the Black urban experience. Middleclass African American neighborhoods in Baton Rouge are not plagued by homicide, assault, and burglary. In other words, not race, but economic deprivation in the form of poverty, unemployment and income inequality account for the disproportionate concentration of serious crime in Black neighborhoods.”

– (Shihadeh and Shrum, 2004)

But, if percent black was causal of socioeconomic disadvantage, perhaps, then maybe controlling for SES disadvantage could be very fallacious. One could easily do a structural equation model to test this theory. The idea behind this model is very simple. If we argued that genetics are a large factor in the between group differences in variables that cause crime and poverty, then we could come to Herrnsteinian conclusion about the between-group heritability of crime and poverty. There are a million ways we can give evidence of such a relationship, such as through potential admixture analysis or testing the role of shared environments in predicting individual and group level crime rates.

Blacks and whites appear to differ in allele frequencies related to educational achievement and cognitive ability (Piffer, 2019), aggression (Beaver et al., 2013), and general personality features (Murray, 2020). Overall, there is a lot of evidence of a strong relationship between percent black of a region and the crime rates. There are a multitude of possible explanations for this including environmental differences, genetic variation, possible systemic racism, etc. This post is not to endorse any of these views but rather to fairly analyze the different arguments

3 Comments

    1. I did have an issue in the math, but it wasn’t because of Cohen’s d.
      The equation I used comes from Jensen’s (1973) Educability and Group Differences, but I had learned about it from Levin’s (1997) Why Race Matters.Essentially, once you find the necessary environmental difference, you have your z. Let E be necessary environmental difference. Then you divide that by sqrt(2) for the difference distribution. Then you can find probability, p, from the z=E/sqrt(2) through the cumulative standard normal deviation. This ended up being 0.25, hence why I have just made an adjustment to the article.

      Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s