A bit of a controversy erupted on Twitter the other day over an article recently accepted at one of the top economics journal. In their own words, the authors find that “genetic endowments linked to educational attainment strongly and robustly predict wealth at retirement.” (non-paywalled version)

I have a lot of mixed feelings about this paper. In part, it brings to the surface important questions about what responsibilities we have as academic researchers when we study topics that are especially sensitive or fraught in the societies within which we live and work.

Before I go any further, let me dwell a bit on the context surrounding the topic of this paper. As many of my readers will no doubt know, the West has a rather dark history when it comes to theorizing about how human biology and genetics relate to various traits that are considered socially desirable.

Enslavement of Africans.

Forced sterilization.

Nazi Germany.

Jim Crow.

Racist immigration policies.

In every one of these cases, people working under the banner of “science” helped to justify the horrible, dehumanizing treatment of others on a mass scale.

Fast-forward to the 21st century: Charles Murray—one of the most prominent scholars working (now emeritus) at one of the biggest (conservative) think tanks in the U.S.—takes the opportunity in a 2014 interview to double down on the notion that genetic differences help to explain why blacks tend to score lower (on average) than whites on IQ tests. In a widely-read 2016 Washington Post profile, the former self-proclaimed “white nationalist” Derek Black points to studies of IQ disparities as part of the evidence he used in the past to logically justify his racist beliefs.

In sum, there’s a long, twisted tradition of people saying, “you’re poor because you have poor genes.”

But what if it *is* true that genetics are related to wealth?

Well, it probably is. And it’s something that makes me very uncomfortable; I wish it weren’t so. But there’s quite a bit of research (much of it from studies of “identical” versus “fraternal” twins) backing up the claim that genetics do, in fact, affect things like our educational attainment (which is related to wealth). (There’s essentially no evidence, on the other hand, supporting the notion that racial disparities are caused by genetics.)

Given that there’s probably a real relationship here, what are we—as a scientific community—to do?

I don’t know exactly. But I want to explore some possibilities.

Claim: When we’re researching a topic where results similar to the ones we discover have often been used (and continue to be used) to justify atrocities, we should probably take extra care with how we present our work. Specifically, we should…

1. Regularly revisit and educate people about that dark history.
2. Take care to avoid making the mistakes of the past (e.g., underestimating or ignoring the role of social environment).
3. Take extra care to not overstate our findings.
4. Acknowledge that getting it right (having really good methodology) is more important when the social stakes of our research are higher.
5. Highlight (high-quality) research that may provide fuller context to our findings in a way that would undercut those who wish to use our research for evil purposes.
6. Carefully consider whether our efforts would better serve society if we spent more of our time researching some other topic, or some other facet of this topic. (Is there some critical piece of things that has been underexplored and that would help empower society to pursue good?)
7. Always seek the truth. Always tell the truth.

I think some folks worry that point 7 will be compromised. Sometimes, I worry about that too. But I really don’t think we have to compromise point 7 in order to pursue points 1-6.

At the same time, points 1-6 honestly strike me as a bit lofty. And I’m not sure they’re equally important in all contexts. For example, I feel a much greater need to fulfill these points when I’m interacting with students or with the general public than when I’m interacting with other scientists. I study topics like race that can have complex implications for our society.[1] Normally, when I write about race, I am writing for an academic journal that no one outside my small research field will ever read. So I trust that the readers know that when I talk about the effects of race, I do not mean to imply that race is an essentialist biological trait but instead I am describing the results of a whole host of complex social factors tied up with this socially-constructed category we’ve created called race. I don’t feel the need to repeat that disclaimer in every article I write, though I’ve entertained the notion that perhaps I should (after hearing someone at a conference make that suggestion).

I don’t want our communication to get too unwieldy with disclaimers and caveats. I don’t want the standards for doing research in tricky areas to be so high that people avoid researching them.

At the same time, I do think researchers have a responsibility to think about how their research might be used by others. We don’t have to pretend to be fortune-tellers. But if computer scientists are developing technology that can be used to create the sorts of  “deepfake” videos that may bring about an entire new era of misinformation, or if social scientists are creating research that will likely be picked up and used as ammunition for racist arguments, we mustn’t pretend that we bear no responsibility for how our work fits into the landscape of our broader social context.

In light of such concerns, when I read this recent economics article about genetics and wealth (the one I mentioned at the beginning), I couldn’t help but feel that this study somehow missed the mark.

I also want to say that it’s very possible I’m holding the authors to an unreasonable standard (especially considering differences in disciplinary norms… I’m not an economist). Thus, it is with trepidation that I will briefly air my grievances:

• I certainly don’t expect this article to explore all of the causes of wealth inequality. But it seems that the restriction of their sample to only non-Hispanic white households at least deserves mention as a limitation given that racial disparities are a major source of wealth inequality in the U.S. Furthermore, racial disparities are also one of the first types of disparities many think of when they hear about genetically-linked disparities.
• More generally, there’s no mention of eugenics or any of the ways that this sort of research question has been very badly approached in the past.
• I know we all tend to overstate the importance of our own results, but I think this articles gives the false impression in various key places that the genetic association with wealth found in the study is bigger than it actually is. We’re told that genetic endowments “strongly… predict” wealth and that we’ve now gained an explanation for “what has heretofore been a form of unobserved heterogeneity that persists across generations.” It seems to me like we observed most of this heterogeneity before, just in the form of the variables that are considered “mediators” in this study. Even in a model with no control variables, the genetic endowment measure can only account for 5% of the variation in wealth (when 25-48% of the variation in wealth can easily be explained by adding some other predictors).
• The study does highlight that making causal claims is difficult (endogeneity is likely), but the manuscript could be much clearer in stating that the relationships found in the study are very likely *inflated* due to this endogeneity (as the study developing the genetic index used in this paper makes fairly clear). Part of the methodological risk here stems from the fact that we’re examining a measure of genetic endowment that is largely black-box.[2] It strikes me as quite likely that the measure may be picking up ancestry-related factors—region, family reputation, etc.—that affect educational attainment; the authors cannot control for any community-level factors, and this point, in my opinion, deserves more attention.[3] Indeed, the limitations of this genetic index are apparent in the original study that developed it, which indicates its predictive power diminishes considerably when looking at genetic variation within a single family or when applied to people of non-European descent.

Again, part of what I’m suggesting is that the stakes of getting things right are higher here than in other research. Behavioral genetics researcher Eric Turkheimer makes a similar point in his review of the book Blueprint: How DNA Makes Us Who We Are:

It wouldn’t matter if the topic weren’t behavior genetics; it would be just another overstated valedictory by a great social scientist, with little price to be paid. But overstating the science of human behavioral genetics comes with the greatest price imaginable: it encroaches on human freedom and justice.

Fortunately, the authors of the article I’ve discussed here haven’t said that “DNA makes us who we are.” But I do think that this topic still deserves to be handled with a little more care than what’s been demonstrated in this new article.

[Edit: One point the authors of the new economics article do raise prominently is that genes interact with environment, and we can change environment. A colleague just raised the importance of this point on Twitter; I’m glad the authors mention this since it helps to counter the claims of genetic determinists.]

__________________

Footnotes:

1. I once had someone tell me during a presentation that I was arguing for segregation (I was not).

2. Specifically, the index was created using a machine learning method to identify potentially relevant bits of DNA from among 10 million possibilities in order to predict educational attainment. Anyone else worried about overfitting?

3. The article does bring up “population stratification” on page 12, where we read that “the principal components help to control for ethnic background factors that would be absorbed by family fixed effects in research designs that exploit within-family variation.” Yet the study that developed the genetic index employed in this article found that even after controlling for principal components, there was still substantial endogeneous across-family variation, a fact the article we’re discussing eventually alludes to on page 27. Here’s some of the language from the supplemental appendix (page 35) of the study developing the genetic index: “we should usually expect GWAS coefficients to provide exaggerated estimates of the magnitude of causal effects. Such exaggeration implies that one must exercise care when interpreting genetic associations with phenotypes such as EduYears. For example, polygenic scores are sometimes described as measures of genetic endowments.” I’m not accusing the authors of misunderstanding this index (after all, one coauthor of the recent economics study was also a coauthor of the study that developed this index); I only wish they’d brought the same clarity of language regarding the limitations shown in this supplemental appendix to a prominent spot in the economics article.