Jaime Derringer Ph.D.

Why We Vary

What Causes Genes?

A genetic association doesn't necessarily mean a genetic cause.

Posted Apr 16, 2020

 Image by janrye from Pixabay
If tophat-wearers were socially isolated from non-tophat-wearers, in a few generations we would find "genes for tophat-wearing."
Source: Image by janrye from Pixabay

Imagine that it was decreed that, starting tomorrow, all Icelanders are required to wear top hats. Additionally, Icelanders are only allowed to marry each other. How will these new social structures impact what the genetics of Icelanders seem to tell us?

Over generations of these policies being in place, random drift leads genetic variants to shift proportions within the Icelandic population. New (or de novo) mutations emerge, hundreds in every person who is born, most of which have no functional consequence, and may—by random chance alone—spread within the socially (and reproductively) isolated Icelandic population. These genetic shifts among the Icelanders occur independent of similar random shifts occurring in the rest of the human population.

After some time, researchers decide to look for genes associated with tophat-wearing worldwide. Unsurprisingly, they find that the genetic variants that randomly became more prevalent within the Icelandic population compared to the non-Icelandic population, regardless of whether these variants have any biological function, are correlated with tophat-wearing. But it is not because genes cause tophat-wearing. What might have been the presumed direction of cause-and-effect is in fact reversed: the social isolation associated with tophat-wearing has caused the genetic associations to appear.

How social structures cause genetic associations

The above story illustrates a ubiquitous phenomenon that geneticists refer to as population stratification. Population stratification often occurs because organisms (whether human or otherwise) are most likely to produce offspring with others who live in relatively close geographic proximity to them. Although we are undoubtedly a single species, we all carry in our genome the unique signature of the random drifts and mutations that have occurred in our ancestors. Geneticists call this unique signature “ancestry.”

In populations that do not move around frequently or quickly, these ancestry signatures tend to map to specific places, in a continuous pattern of dispersal that spans the globe and tracks the pattern of human movement over the history of our species. Similar ancestry patterns can also be detected in societies where social structures limit the choice of mates (say, by banning interfaith marriages). 

In humans, the solution to resolve this uncertainty has been to examine the predictive utility of genetic associations within families. By comparing siblings raised by the same parents, we can statistically control for many of the parallel genetic and cultural pathways. With this approach, we consistently find that the predictive utility of (or the variance explained by) genes for social, psychological, and behavioral outcomes is substantially reduced. That's true even for the most heritable, most believably "biological" human trait—height. And it is especially true for the measures in which many proponents of applied genetic prediction are particularly interested. The predictive utility of polygenic scores for educational attainment and IQ test scores is slashed in half within families, indicating a substantial influence of social stratification. 

Genes cannot tell the future

Attempts to use genes to make predictions about human outcomes have been based on data from remarkably limited environmental contexts. Participants of European ancestry account for almost 80% of genetic studies, and participants born and educated in the late 20th-century account for nearly all of them. As people become more mobile, as available resources and the social structures used to allocate them change, as our definition of intelligence develops beyond Binet's original statement ("IQ is what the test tests"), we have no way to predict how this will affect the utility of the polygenic scores that researchers construct in the present. Even as statistical and technological methods for genetics develop at a rapid pace, we will remain always unable to know how social structures will change in the future. The unknowable impact those changes might have on human outcomes means that genetic associations, either alone or aggregated into polygenic scores, will never be a perfect, forward-looking index of human potential.

A version of this post was originally drafted a couple of months ago in response to a Wall Street Journal op/ed. It's posting now is inspired by the publication in Science Advances of a paper addressing these issues in-depth (see References below).


Morris, T. T., Davies, N. M., Hemani, G., & Smith, G. D. (2020). Population phenomena inflate genetic associations of complex social traits. Science Advances, 6(16), eaay0328. https://advances.sciencemag.org/content/6/16/eaay0328