Links for January ’22

I will try to get in the habit of collecting the most interesting studies, articles, posts, etc. related to human biodiversity in a monthly post, together with some commentary. The links are not necessarily to brand-new stuff; they are just what I happened to come across recently.

Genetics

• A Saturated Map of Common Genetic Variants Associated with Human Height from 5.4 Million Individuals of Diverse Ancestries by Yengo et al. This GWAS found around 12,000 independent SNPs that are associated with height at p < 0.00000005. You can combine them into a polygenic score that explains about 40 percent of height variation in whites, less in non-whites. The SNP effect sizes are slightly smaller in sibling comparisons than in unrelated individuals, but not more than is expected given assortative mating for height. This means that common variants contributing to height is now, in a sense, a solved problem, although the reduced accuracy in non-whites suggests that many of the SNPs are not true causal variants but only correlated with them. The heritability of height is around 80 percent, so the remaining 40 percent is accounted for by rare variants, the observation of which requires whole-genome sequencing (WGS). The theoretical expectation is that rare variants have a much larger average effect size than common variants, so detecting them is not necessarily as daunting a task as it may seem once large WGS samples become available.
• Classical Models for Twin Data by Karl Jöreskog. The workhorse of twin research is the ACDE model which in principle allows the estimation of the effects of additive genetics (A), the shared environment (C), genetic dominance (D), and the unshared environment (E). However, C and D are confounded when there are only data from MZ and DZ twins raised together. Therefore, the full ACDE model cannot typically be used and submodels like ACE and ADE are used instead. Jöreskog, the grand old man of structural equation modeling, proposes in his article that the ACDE model can be estimated in a typical twin study setting through a reparameterization of the classical model. This is a strong claim given that the non-identifiability of the ACDE model has been held as a truism in behavioral genetics for decades. The received wisdom is that the model is not feasible without additional data sources such as adoptees (or higher-order moments as in Ozaki et al., 2011, but that runs into problems, e.g., you must use the generally poorly performing ADF fit function rather than maximum likelihood). If Jöreskog’s approach pans out in further testing, it will be an important advance, especially if assortative mating can be added to it.

IQ, psychometrics, education

• Effects of a statewide pre-kindergarten program on children’s achievement and behavior through sixth grade by Durkin et al. This is a new follow-up study of the Tennessee Pre-K program RCT. It found negative treatment effects that were small in magnitude and mostly statistically significant, generally, it seems, even after multiple testing adjustment (which was not done). This was the case for most cognitive and non-cognitive outcomes. The results will not please universal pre-K advocates. However, they are also not easy to explain if, like yours truly, you do not believe that preschool programs have any long-term effects one way or another; I would have expected null effects by sixth grade. The control condition in the RCT was mostly home-based care, not some other center-based care, at least in the subsample for which those data were available.
• Many black-white disparities in important life outcomes are mostly or entirely eliminated after controlling for youth standardized test scores by jay520. This is an excellent survey of studies showing that IQ is the main, if not the sole, factor behind many socioeconomic and behavioral disparities between blacks and whites. Many people are skeptical of the idea that IQ would have such a strong effect (whether causal or just statistical) on race differences because the within-race effect of IQ on most outcomes is much smaller, say R² = 10–30%. However, if race differences in the other determinants of outcomes are small or non-existent, then IQ will necessarily explain the lion’s share of race differences. Say the effect of IQ on some outcome y is given by y = α + β×IQ + e where α is an intercept, β is the effect of IQ, and e is a residual that is independent of IQ and captures all other influences on y. When black-white differences in the residual e approach zero, the proportion of race differences in y that is explained by IQ approaches 100%, no matter how small or large β is.

Miscellaneous

• The impact of a poverty reduction intervention on infant brain activity by Troller-Renfree et al. This RCT found that giving $333 to poor mothers each month did not lead to significant changes in the brain activity of their infants, as measured by EEG wave patterns that have (dubiously) been linked to better cognition. This would be an entirely forgettable study if it were not for the fact that it was published in PNAS, spinned as a positive finding by the authors, and given effusive, largely uncritical coverage in the prestige media. Andrew Gelman and Scott Alexander, among others, have shown in detail that this is indeed a null study. It is notable that the research project has received about nine million dollars in funding from the NIH alone, plus untold amounts from a veritable who’s who of the big money foundations–Bezos, Gates, Chan Zuckerberg, Ford, etc. The authors certainly have a motive to oversell their findings.
• Black Lives Matter’s Bloody Toll on Black People by Steve Sailer. This post contains another one of Sailer’s striking graphics which is based on newly released CDC data:
  
  The “Floyd effect” is like an ongoing black 9/11. With this carnage in mind, the casual way people still show allegiance to BLM in, say, their social media profiles seems grotesque. Now that the data are coming in, it will be interesting to see how this is framed in the research literature. The Floyd effect appears well-suited for study with causal inference methods from econometrics, but few laurels await people who reach straightforward conclusions about it.