+Here and through the remaining chapters up until the conclusion, Murray elects to switch to the nomenclature "European"/"African"/"Latin" rather than white/black/Latino (respectively, with "Asian" remaining unaltered), on the grounds that using less familiar terms for these groups will drag along less cultural and political baggage without resorting to outright obfuscation ("populations A, B, C, and D"). It doesn't feel that effective to my ear, and I kind expect it to backfire for a lot of readers, to whom the continental African/European/Asian terms probably sound _more_ racially essentialist than I think Murray wants to come off as!
+
+The next four chapters follow a formula: "Race Differences in Cognitive Ability", "Race Differences in Violent Crime", "First-Order Effects of Race Differences in Cognitive Ability", and "First-Order Effects of Race Differences in Violent Crime." (Those chapter titles felt awful just to type!! Am I really doing this?) Much of the value of these chapters is in the graphs and tables documenting statistics that many readers will be unfamiliar with. The scatterplots of nationally-representative test scores are interesting. The black–white gap _did_ shrink between '70s when it was about 1.3 standard deviations, until about the 1990s, but has been stubbornly stable since then at about 0.85 standard deviations (a.k.a. [Cohen's _d_](/2019/Sep/does-general-intelligence-deflate-standardized-effect-sizes-of-cognitive-sex-differences/)). Murray estimates the current white–Latino difference at 0.62 standard deviations, and the current white–Asian difference at 0.3 standard deviations (favoring Asians).
+
+Murray briefly addresses two popular (but mutually in tension) classes of objection: that the gaps will vanish with better (more equitable) education policy, and that the tests are biased. The response to the we-can-fix-it objection is basically, "We tried that and it didn't work": a lot of money and effort has been poured into attempts to narrow the racial achievement gap over the past thirty years of its stability, but we just don't know of any interventions that lastingly increase cognitive ability for _anyone_. The response to the tests-are-biased objection is basically, "We checked for that and it doesn't work": psychometricians have all sorts of technical measures to verify that their tests are doing what they think, but beyond that, if the tests were biased, you would expect them to underpredict school and workplace performance, and that's not what we see: for example, black people get worse college grades than their SAT and ACT scores would predict, not better.
+
+Perhaps you could argue both that the tests are racist and that teachers are _even more_ racist?—but you'd want to be specific about what alternative metric of "merit" you think they're being racist _with respect to_. The thing about tests is, a vast space of "objective" procedures can come up with a number, without giving anyone a reason to care about that particular number; if you care about group rank orderings, you _could_ come up with a measurement that gets the the group rank ordering you want. (Black people have more melanin than white people, on average! People with more letters in their name take longer to say their name out loud, on average! Cats do better than humans on a test of scratching, on average! "Hispanic" comes before "White" in alphabetical order!) The problem is that it looks like the "cognitive ability" thing that psychometricians are trying to measure is actually a pretty robust abstraction that summarizes variation _in individuals_ that people care about (like the ability to master a profession); if it were just a tool of racial oppression, it's hard to see why it would work so well _within_ groups.
+
+[TODO: mention https://en.wikipedia.org/wiki/Black_Intelligence_Test_of_Cultural_Homogeneity ]
+
+In the chapter on violent crime, Murray presents a table of black/white and Latino/white ratios of arrests for violent crimes in thirteen cities for which data was available. The median black/white ratio was 9.0 (that is, 9 black people arrested for violent crimes per 1 white person so arrested) and the median Latino/white ratio was 2.4.
+
+To argue that these ratios are driven by real differences in behavior rather than biased police, Murray attempts to "triangulate" the true crime rate with other data. For example, arrests for murder specifically are going to be less biased by selective enforcement or fraud: even evil and corrupt cops who don't consider themselves above, say, planting evidence of drugs, seem less likely or able to fake a human corpse. So if racial differences in murder charges match differences in violent-crime arrests more generally, that's probabilistic evidence that arrests are tracking a real difference in criminal behavior.
+
+Interestingly, Murray argues that this is true even if you don't think police are generally getting the right suspect (!!), as long as the suspect who is arrested is of the same race as the actual perpetrator, which will usually be the case given how many murders are crimes of passion where the victim and perpetrator knew each other (in highly segregated communities), or tied to gang activity (where gangs are almost always monoracial). The scenario most prone to racist police falsely accusing a black person—non-gang-related murders where the alleged perp is black and didn't know the victim—only accounted for 4% of all homocides. Meanwhile, the group ratios for murder arrests are more stark than for violent crimes more generally: a median black/white ratio of 18.1, and a median Latino/white ratio of 4.7, which is not the pattern we would expect to see if cops were using their discretionary powers to falsely imprison blacks and Latinos on lesser charges. Another source of data for triangulation is in reports of crimes _to_ the police: if crime _victims_ aren't racist in the same way the police themselves might be, then that would show up in the numbers—and it turns out that even black and Latino victims also report more black and Latino perpretrators, even in neighborhoods where they are a minority.
+
+As an argument about patterns of behavior in our own world, I find this quite convincing, but again, Murray's agnosticism about the causes of behavior limits its applicability; I expect the sophisticated advocate of structural-racism theories to be entirely unmoved. Even if the police sometimes getting the wrong man can't change the conclusion about which races do how many murders, a world in which the police sometimes get the wrong man of the same race is _exactly_ the kind of factor that would contribute to structural racism—if the System is going to treat members of your racial caste interchangeably anyway, that changes your _incentives_ to commit crime, relative to the world where people were [literally raceblind](/2021/Mar/link-see-color/). We don't know what other equilibria might be possible for a civilization with a discipline of Actual Social Science, even if there's no obvious way to jump out of our own equilibrium with the crude "policy" levers available to actualy-existing governments.