Hsu’s blog post makes two claims about race. The first argument is that ‘Hypothesis 2’ could be correct—i.e., that there could be genetically driven differences in exciting traits like IQ between races (or ‘groups,’ but I think we all know which ‘groups’ we’re really interested in). I agree with this argument.
I completely disagree with the second claim, which is that genetic clustering studies constitute ‘the scientific basis for race.’ It’s true that scientists can extract clusters from genetic data that match what we call races. If you gave me a bunch of human genotypes sampled from around the world and let me fuck around with that data and run it through PCA for a few hours, I’m sure I could do the same. But it doesn’t automatically follow that my classification is correct.
For example, if you sample some whites, sample some blacks, and expect those two categories to automatically pop out of your analysis, you might be surprised. Here’s a recent paper that estimated the European ancestry in African-Americans by analyzing genotypes from samples of US whites, US blacks, and several subgroups of Africans. Running PCA on all of the genotype data, and plotting the first two principal components of the subjects’ genotypes in each sample gave these clusters:
If we treat the widely separated clusters as races, we don’t automatically recover a black race and a white race. We end up with a Mandenka race, a white race, and a Bantu + Yoruba race, with African-Americans smeared out between them.
The researchers could no doubt have come up with an alternative rotation of the axes that would’ve projected all of the African samples on top of each other, and the European sample far away from them. But what would justify the alternative projection over the original one?
Maybe my own personal concept of ‘race’ emphasizes differences among sub-Saharan Africans, instead of continental differences. Then I might do a PCA on a set of sub-Saharan African genotypes, find a couple of principal components that best separate out the sub-Saharan African subgroups, and only then plot the north Africans and non-Africans along with the sub-Saharans.
Here are a few plots from a study that did just that. Notice now that the most widely separated clusters are three, or perhaps four, sub-Saharan African clusters—and the rest of the world forms one little cluster in the middle of them!
If I were a scientist who had started with the idea that the main races consisted of several African subgroups, plus one other race containing all non-Africans, this analysis would seem to completely vindicate my initial beliefs! But the analysis turned out the way it did mainly because the way I did it was driven by my original taxonomy of ‘races.’
I’ve picked out two papers myself to make points, now I’ll write a bit about the ‘Risch et al.’ paper Hsu points to. Risch et al. calculated genetic clusters by running data collected for the Family Blood Pressure Program through the structure program. Hsu writes that the clusters that emerged ‘correspond very well to self-identified notions of race.’
Well, there’s no ready-made algorithm which takes genotypes as input and spits out objectively determined races, and structure is no exception. There are some subtleties to how the program works. For one thing, it doesn’t automatically confirm an optimal number of clusters and then sort the subjects into the appropriate number of clusters: the researcher tells structure to put subjects into some number k of clusters, and the program then does its best to fit the subjects into k clusters. So the fact that structure’s output contained an intuitively pleasing number of clusters doesn’t mean very much.
Another issue is that the kind of model structure uses to represent distributions of genotypes is suboptimal for cases where samples have been isolated due to distance and have suffered a lack of gene flow. But, if Hsu is correct, this is exactly the case for Risch et al.‘s data, since he writes that Risch et al.‘s ‘clustering is a natural consequence of geographical isolation, inheritance and natural selection operating over the last 50k years since humans left Africa!’
There is more I could write, but I might as well just link this book chapter, which discusses issues with trying to algorithmically infer someone’s racial ancestry. I’ve already written more than I meant to—sorry for the lecture—but it disappoints me when someone well-credentialed (a professor of physics!) uncritically waves around ambiguous results to shore up a folk model of race.
(Edited to fix last link.)
There are a few randomized trials of baclofen, if those count:
Addolorato et al. 2006. 18 drinkers got baclofen, 19 got diazepam (the ‘gold standard’ treatment, apparently). Baclofen performed about as well as diazepam.
Addolorato et al. 2007.61814-5) 42 drinkers got baclofen, 42 got a placebo. More baclofen patients remained abstinent than placebo patients, and the baclofen takers stayed abstinent longer (both results were statistically significant).
Assadi et al. 2003. 20 opiate addicts got baclofen, 20 got a placebo. (Statistically) significantly more of the baclofen patients stayed on the treatment, and lessened depressive & withdrawal symptoms. The baclofen patients also did insignificantly better on ‘opioid craving and self-reported opioid and alcohol use.’
Shoptaw et al. 2003. 35 cokeheads got baclofen, 35 a placebo. ‘Univariate analyses of aggregates of urine drug screening showed generally favorable outcomes for baclofen, but not at statistically significant levels. There was no statistical significance observed for retention, cocaine craving, or incidence of reported adverse events by treatment condition.’
Heinzerling et al. 2006. Just found this one: 25 meth addicts got baclofen, 26 got gabapentin, and 37 got a placebo. Going by the abstract, across the whole sample, neither baclofen nor gabapentin beat the placebo, but an after-the-fact statistical analysis suggested that baclofen had a significantly stronger effect than placebo among the patients who were stricter about taking the baclofen.
Franklin et al. 2009. Editing in this one too: 30 smokers who were thinking of quitting took baclofen, 30 took a placebo. Both groups smoked progressively fewer cigarettes a day during the trial, but the baclofen users had a significantly steeper decline than the placebo users. However, they did not report significantly less craving feelings.
Kahn et al. 2009. Last one, I promise: 80 cocaine addicts from around the USA got baclofen and 80 got a placebo. There were no statistically significant differences in treatment retention, cocaine use, measures of craving and withdrawal, or any of the other things the researchers tested for, except on a couple of post hoc tests. The researchers hint that the dose used (60mg) might have been too small.
Most of these studies are a few years old now, and there are also case reports, uncontrolled trials like this one and studies done on rodents. I’m kind of surprised no one’s tried doing a larger scale trial of baclofen for alcohol. I haven’t looked at these in detail—maybe the effect is only statistically significant and not clinically significant, or there’s some subtle methodological issue I’m missing.
(Edited this comment a few times because Chrome helpfully posted it prematurely for me.)