The Truth About Murder

Occam’s Razor Strikes Again

You know the old line “13–50,” yes, African Americans commit a lot of murder compared to other races. Many people assert that it is for socioeconomic reasons, and if you ask any more questions, you will be told how socioeconomics is really hard to quantify (it is easy), and it is just too complicated to even explain — but trust me it is the socioeconomics!

Should somebody assert that the environment is causing these differences in homicide rates in a free market system, we must ask why these environments are the way that they are. Is it due to the bell curves? Probably so. This is why “controlling for socioeconomics” is such a stupid idea, but I will do it anyways.

I think that it is important to acknowledge that not everyone is their group average, and there is no group where the average is a violent felon, but we cannot ignore the facts in terms of policy. Public policy itself is derived from generalizations, not this whole teasing apart the individual data. When you stop at a stoplight at 4 in the morning, it is not because you are in danger if you run it, the road is presumably empty. You stop at the light because it is the law — which does not have exceptions for your circumstances because subjective judgement would lead to disaster. If you disagree with that view of policy, that means that you must be an anarchist and oppose all wealth redistribution. Otherwise, you agree that policy formed from generalities under the guise of the common good/maximizing collective prosperity. Applying this sort of “maximize prosperity” logic to race in an ad hoc fashion under the guise of “equity” actually creates burden for all races, including the one that you are trying to help. The Ferguson Effect has been observed after anti-police riots several times. Affirmative action tends to cause lower graduation rates/program completion (Rose, 2005, Baird, 2016) for the beneficiaries.

The Data

First let’s talk about the data on the environment. The environmental gap between White Americans and African Americans is not massive, but it is real. Russell Warne factor analyzed various indicators of socioeconomic status and found that the gap is probably between 3/5 and 2/3 a standard deviation. This would be classified as a moderate or medium effect size in Cohen’s d units.

“All of these socioeconomic status variables are composite variables based on parental income, parental occupation, and educational level, though the HSLS dataset also includes a measure of urbanicity of the child’s high school.”

There is a major genetic confounding component to most of these things. Quantifying the genetic component of occupation is really hard, but if your dad is in congress, there are complex traits with partially genetic components that probably helped get him there (extraversion, height?, intelligence, maybe attractiveness, work ethic, and so on). There is a major genetic confounding effect for educational attainment. Youth achievement in England has a heritability of 61% at age 16 (Morris et al., 2018). The SNP heritability is persistently around 20%–30%, however, SNP heritability gives conservative estimates due to the limitations the significance threshold imposes on rare SNPs (Rimfeld et al., 2018). The heritability of income in Finland is over 50% for men and near 40% for women (Hyytinen et al., 2019). In Sweden, the 20 year average income heritability is about 0.58 for men (Bejamin et al., 2013). The genetic correlation between income PGS and Intelligence PGS is 0.69 in England (Hill et al., 2019).

Without controlling for the genetic confounding you sill have a gap that is relatively small compared to the behavioral differences between African Americans and white Americans when you hold the environment constant.

We can examine data on homicide rates at the neighborhood level in Chicago over time and the correlations to income are minimal for African Americans between the 25th and 90th percentile of neighborhoods from 1986–1995 and the data on whites is no correlation. Regardless of how you feel about the poverty-crime correlation, race matters here. These numbers are coming from one of the co-authors of the book “Freakonomics” so you know that these are not racially motivated numbers. In fact, he comes up with all sorts of peculiar explanations to explain away the data I believe. Not too familiar with the paper I am citing because I am not particularly enamored with that sort of foolishness.

At the state level, things are also very obvious. Sorry if these graphs are unclear about the specific states, but you can see the trends I bet.

The SPLC says don’t let your eyes fool you about race and crime, don’t let your eyes fool you with these graphs! The R² is stronger for race!

You are probably smart enough to know that the race-IQ stuff explains some of the race-poverty stuff, but we can pretend that race and IQ are non-issues.

In net of the correlation between poverty and crime, which explains about 45% of the variance, 30% of the remaining variance can be explained by % black. You would think that there would be no remaining correlation between race and criminality after you control for the socioeconomics of the state!

2017 Homicides by state

2017 demographics by state

2016–2017 poverty data (it is pooled, I think that a year of time lagged data is more accurate — whatever sociological processes would cause crime would not flip like a switch

At the county level, data from the 2020 county health rankings yielded unsurprising results.

Most of these variables fail to meet the significance threshold but are worth including for the purpose of showing that they have no added value

It turns out that despite what the SPLC says, race is a really good predictor of homicide. In stepwise regression style (not pictured), the first variable detected is race. Some might argue that the residual for % African American will negatively correlate with the residuals for income which will in turn cause higher levels of criminality. We can evaluate that hypothesis with the residuals of incomes and correlate them to homicides.

If you are not fully convinced, we can look further.

Here is the effect size of the median household income differences relative to homicide rate differences.

*I am aware of assumption violations in this method

The effect size is much larger — about half a standard deviation larger — for homicides!

The county level median income distributions are not normal.

For serious calculations, I will use log transformations.

Here is the untransformed raw data for homicides, I will transform that to logs as well.

After a logarithmic transformation, the effect sizes are kind of unimportant in terms of numbers (for example, for Black and White incomes when aggregated into counties, the effect size is about 2.5 standard deviations). The reason that I do this transformation is because things occur on exponential levels. The rank order does not change, however, when dealing with things like this, it is a good way to visualize correlations.

It is pretty hard to tease apart what the causal variable is here. Within nearly all counties that reported data, there is a higher African American homicide rate than white one, wealthier people commit less crime, and there are race differences in income within counties. Syllogistically, this might make a lot of sense for you to assume that the money is causing the differences between the races based on purely correlational data if that fits your internal biases, so we can use Hispanics as a dummy variable whites (since only 3 counties have whites poorer than blacks and none had aggregated homicides by race). If Hispanics being poorer than blacks is a predictor of Hispanics having higher homicide rates than blacks, we can assume it is about the money.

There are 470 or so counties that have data on the African American and Hispanic median incomes in this data set. Most of them do not show the homicide rates by race as well which is unfortunate, but of the ones that we do have, there is considerable difference in the homicide rates.

Doing matched pairs T-Tests for the 47 counties that reported all of these measurements and fit the criteria of the median African American income being higher than the median Hispanic income, we see that it is not all too close in terms of the difference. A group mean differences of nearly 0.4 on a log scale is not a small difference in homicide rates no matter what the rate is due to the nature of powers of 10!

After undoing the log transformation, we can look at the data in practical terms and the effect sizes are still very large. A difference in means of under 10k in income by the races in these counties and a homicide rate differences of nearly 8 per 100k! The homicide rate in America has been generally around 5 in the last several years!

When these sorts of tests are done for comparing African Americans to white Americans, similar results are found.

There are too few counties to show that African Americans making more money than whites have a higher homicide rate than whites, but we can do is show what happens in communities of similar SES.

A log of -1 means 0.1, this is not a computational error that you see negative values on the white graph

An intercept of 8.185 and an unstandardized slope of -1.605 means that at the county level, a white median income of 50,000 (4.699 in log form), would predict a homicide rate of to the power of 8.185+(4.699)(-1.605). This is 10^0.643, which is equal to about 4.4 (per 100,000 residents is the implied unit).

Plugging that same income into the equation for African Americans gets you

5.708+(4.699)(-0.975)=1.126475

10^1.126475=13.38 (per 100,000 is the implied unit)

Extrapolation is tempting here, but we cannot do it! The formulas differ by race. One thing is for sure, blaming crime on environmental factors is absurd.

To clarify, it is not that income is the ONLY environmental factor, it is that income correlates very strongly with every other environmental factor like nutrition, water quality, schools, etc. To blame “other” environmental factors without providing hard evidence that that income does not capture enough of the variability in that environmental factor is a non-argument. You think about what income correlates to — IQ, education, time preference, etc. and realize, yeah, that is good enough. If you want to do the general socioeconomic factor, we can use the one discussed by Russell T. Warne where you get their job, education, income, and ubranicity of the children’s school. If African American lawyers feed their children worse food than white lawyers do, so be it.

On the global level, these trends exist too.

Per 100,000 rates of crimes From Race, Evolution, and Behavior

Rushton included Indonesia, Malaysia, and the Philippines for Mongloloids which increases their rate of violent crime tremendously. When you exclude those nations, you start seeing far lower numbers

Japan, for example, has retained strikingly low crime rates for decades. At no point was their homicide rate above 4/100k in the last 70+ years.

I had to look hard for the rates in China but we got em!

China had about 1.1 billion people in 1988, and 1.2 billion in 1995.

15,959/1,200,000,000=0.000013299

A per 100,000 homicide rate of 1.3299 per 100,000 in 1988 is astonishingly low. Even with the increase over time (likely due to a growing Muslim population), it remained well below 4.

The rate did increase in the subsequent years, however, this was true globally. China’s Baby boom was in the 1960’s so the 1990’s was when their equivalent of baby boomers were young adults.

In conclusion, when you look at the data, you have to be a total moron to think that these differences are not due to systemic biological differences. It is not my job to explain the exact physiological and biochemical process at play for the same reason that I do not have to “find the genes” for the IQ stuff.

The policy implications of these truths are simple. Do not burn down buildings and adopt a complete victim identity for a single bad police officer, defund the police, or burn a city block down. Trump calling it “sleepy Joe’s America” is another bad way to address it.

Is Africa Different? Not Really.

The rates of homicide, by global region (and country), are on wikipedia — UNODC is UN office of Drugs and Crime. This literally is not a secret.

We need an honest discussion. The truth is that the homicide rates in America for African Americans are very similar to those in Africa. While we do not have the exact rate of African American homicide (since many victims have unknown killers — about 40% of all murders!) we do have our lower end estimates. For computational convenience, I think we could assume unknown murderers were the same race as the person that they killed, but we don’t and I do not want to go that extra step just to show how many people African Americans murder. Seems pointless.

There is tremendous heterogeneity in Africa for the homicide rates. The homicide rate in Nigeria was 34/100k in 2016 according to the UN. Most African Americans (ADOS) are of Nigerian ancestry, and many immigrants have come here from Nigeria as well (I believe Trump blocked them from moving here in recent history).

African Americans commit homicide at about 58% the rate of Nigerians in Nigeria when age is not adjusted for.

I do not know the exact formula of age adjusting, nor do I claim to be an expert in this subject, but most homicides are men between 15 and 29. In Nigeria, that is 13.7% of the population. In America, that was about 10.5% of the African American population in 2000 (I assume that it has gone down even more, but we can roll with 10.5%).

13.7%/10.5%=1.3047, so you would expect a homicide rate of about 20%-30% higher given that they have 30% more of their population in the most homicide prone age+sex section of the population. I would argue that there is a nonlinear increase in age structure because the frequency of interactions with other young men increases the probability of a violent interaction.

30% of 20 is 6, so that moves the gap from 14/100k to 8/100k if we want to consider the fact that they have the issue of domestic terrorism (I am not talking about people with tiki torches and goofy robes, I mean Boko Haram). Boko Haram killed over 30,000 people between 2009 to 2020. If we estimate that to be somewhere around 3000 a year, and we use the midpoint (2015) population of 181 million — that is a homicide rate of 1.657/100k that is due to domestic terrorism. In America, that number is seriously like zero. Black people do not do much domestic terrorism aside from sporadic BLM moments when they burn down things like police stations.

After considering a little bit of age and eliminating the activity of a single terrorist group, the rate in Nigeria is only about 6.343/100k higher than in amongst African Americans. Should we consider other things like admixture and policing, it really becomes the same without mentioning any socioeconomic factors. I mean think about it — if the average African American is 15–20% European, that would mean that we would expect 15–20% of the gap between Europeans and Africans to be eliminated. It is applicable to the IQ numbers and many other traits, why not homicide rates?

Lastly, depending on what YEAR you get your numbers and how they are calculated (ie age adjusted and varying estimates of homicide totals, year to year variance, and varying population estimate), you may find that the rates of homicide amongst African Americans and Africans is strikingly similar:

Somebody did this in 2016 and and found a more shocking result than I did!

For those that mention the Russian homicide rates as a gotcha point, you do realize that political assassinations do not count, right? By this indignant standard of proof, we may as well add gang related homicides to the list of hate crimes. He was born a crip, you cannot hate on his identity!

After you remove the political ones, you have to look at who is committing the crimes. It seems like there is a problem with Siberians.

The homicide rates and alcohol problems remind me of what happens in a country that gets blamed for the oppression of Siberian Americans.

unpublished stuff *anon*