A while back, I wrote a blog post about a GWAS (Genome-Wide Association Study) on the genetics of educational attainment. I mentioned how it was interesting that it was a GWAS on “educational attainment” rather than “intelligence” because, while the two seem closely connected in many people’s eyes, the latter appears more like the sort of thing where it makes sense to do genetics of.
Sometime later, I came across a tweet about a study on the genetic variation in health coverage in the US context. The tweet (which I couldn’t find) had a dismissive tone. It was something along the lines of “where did their brains go?” along with a retweet of the classic “scientists were too occupied with whether they could; they did not stop to think whether they should”.
Those comments intrigued me, but let’s come back to them later and stick to the papers for now. The health coverage paper uses a twin design, which is the gold standard in nature/nurture debates much like how RCTs (Randomized Control Trials) are the gold standard in causal discovery. In a twin study, the presence or absence of some trait is compared between identical twins reared together or apart, as well as identical and fraternal twins reared together. If identical twins reared together are just as similar as identical twins reared apart in relation to this trait, then the trait is considered highly genetic. If the similarity between identical twins reared together is similar to the similarity between fraternal twins reared together, then the trait is considered at least partially environmental. You can make the process more complicated, of course, especially when you want a bigger sample and therefore have to use non-twin siblings, etc. The study finds that
46% of variance in uninsured vs. private/public coverage and 50% of variance in private coverage vs. uninsured among individuals younger than 65 explained by genetic effects.
As standard practice, the authors start their paper with a brief literature review aimed at justifying their research question and method. Here, they cite studies on the genetics of financial risk-taking, health conscientiousness, and human capital accumulation (which predicts prospects of getting jobs that would pay for health insurance), arguing that these may be possible pathways connecting genetics and insurance status.
I am not exactly qualified to judge the methodological quality of this study, but the paper reads pretty convincingly. It certainly is not unique in its kind. In addition to the papers it cites and the educational attainment paper I discussed before, there has been another newly published GWAS paper on the genetics of future income (see a blog post explanation of the paper).
While the science is by no means airtight, those of us who don’t like the implications of these papers can’t always rely on the science to fail. But let’s still stick with science for now.
What neither twin study nor GWAS provides is a mechanistic picture of how the genetic influence is supposed to play out. Both are correlational, after all. GWAS, the geneticist’s new “hammer” that turns everything into nails, is able to identify a number of “genetic locations” (SNPs; single nucleotide polymorphisms) that are salient for the trait in question. Sometimes there are past results and theories concerning the functions of some of these SNPs, which form the basis of a mechanistic story. But even when we have some story, it tends to not stand up to scrutiny. For one thing, if we only have theories about the workings of 20 SNPs among the 150 that are picked out by our GWAS, we don’t actually know whether the remaining 130 would inhibit the 20 or otherwise interact with them in causally significant ways. For another, our understanding of the function of SNPs tends to be partial — we usually know that an SNP does this thing, but not whether it also does other things. Lastly, the story tends to be so sparse that it amounts to little more than creative writing. For example, perhaps one SNP has been shown to correlate with a bigger frontal lobe, and the prefrontal cortex is in charge of decision making and future planning. A story can then be told about how genes determine better decision skills, which in turn determine income levels.
To be fair, many (but not all!) scientists resist this kind of creative storytelling and put some sort of warning against the temptation to infer from “these genes are associated with trait X” to “these are X-genes and X is entirely genetic” in their paper. But I don’t think these warnings are enough, and not just because “the public” would misinterpret it — I think scientists often do, also.
As Stephen Jay Gould says in the preface of The Mismeasure of Man, nobody denies gene-environmental interaction. Everybody (with a decent level of scientific base knowledge) knows that everything is the result of some kind of interaction between genetics and the environment, and so merely pointing out that interaction occurs is not going to help settle important disagreements.
What is more important is the nature of these interactions. As the genetics of educational attainment paper points out, the nature of an interaction may be nowhere near what we think of when we think about gene-trait pathways. For example, suppose a certain SNP give people a particular sitting posture that makes them uncomfortable when sitting in a particular type of chair. It just so happens that this type of chair is standard in K-12 schools. Since people with this SNP tend to be chronically more uncomfortable in their school chairs than their peers, their SAT score (an intelligence indicator by intent if not by name) is, on average, 2 percentile lower than their peers. This makes this SNP significantly correlated with intelligence scores. But I’m willing to bet that when most people think “gene-environmental interaction”, this is not what they imagine.
Here is another, arguably deeper, problem. Why is it that, when we read about studies on the genetics of income or health insurance, we immediately think they must be mediated by other, more “fundamental”, characteristics like self-control or intelligence? Or, rather, why is it that we don’t try to find pathway explanations through income and health insurance for the genetics of self-control and intelligence?
I’ve always thought of a particular kind of philosophy of science question in the following way. In the beginning, someone creative comes up with a method of discovery, along with some reasonably convincing story of how it’s supposed to work. People then use this method for a period of time (“normal science”, if you will). The story is supposed to preclude the method to be used on a certain range of cases, but somebody eventually tries it anyway. They then find out that, alas, the method “works” equally well in cases where it’s not supposed to work, according to the story. As time goes by, people also realize that the story actually doesn’t have any independent merit, other than that it sounds reasonable. Since the method works in cases where it’s not supposed to work, we now have to figure out whether 1) we’re wrong about the nature of the cases we think it’s not supposed to work for, or 2) it doesn’t actually work for the cases that we thought it worked for either. In either case, we have to reexamine the story.
This is what, I think, is happening with genetic studies of traits. It seems reasonable that things like self-control might be genetic. We conduct a twin study or a GWAS and find out that it is genetic, and so we leave it at that. It doesn’t seem reasonable that health insurance coverage is genetic in the same way. However, we apply the same method and get the same result. Now we need to explain how it is genetic. But why don’t we need to do the same explanation for self-control? What reason do we have to treat them differently, other than that self-control intuitively feels more like the sort of thing that would be genetic?
- It might happen after all - May 14, 2023
- Another job market data point - December 17, 2022
- Our place in the fediverse - November 30, 2022