A new article in the ICA Journal explores the power of DNA-based scores for predicting cognitive & educational outcomes. The authors found that about half of the predictive power was due to differences between families and half was individual differences in DNA.
This means that when comparing siblings within the same family, the DNA-based scores (called “polygenic scores”) lose some of their predictive power. In contrast, the polygenic scores were less attenuated when used to predict BMI and height (as seen in the image below). Apparently, the polygenic scores for IQ and educational outcomes capture much more between-family sources of variance than polygenic scores for BMI and height do.
To try to understand this between-family influence, the authors examined whether family socioeconomic status (SES) was an important between-family variable. The results (in the graphic below) show that SES is part of this between-family influence, but it is much more important for educational outcomes than IQ/g variables.
Studies like this inform us about how DNA variants relate to life outcomes. Knowing the relative importance of within- and between-family characteristics can give clues about the cause-and-effect relationships between genes and outcomes.
The pessimist may say that because polygenic scores for IQ and educational outcomes are strongly influenced by between-family effects, they are overestimates of the effect of genes on these variables. The authors are more optimistic, though. Most polygenic scores will be used to make predictions about groups of unrelated people–not siblings within the same family. By capturing between- and within-family variance, polygenic scores are going to be more accurate when making these predictions. (On the other hand, predictions within families, such as in embryo selection, should prefer the attenuated predictions based on siblings.)
This is really important for understanding what polygenic scores actually measure. The fact that they lose half their predictive power when comparing siblings shows that a big chunk of the “genetic” signal is actually capturing between-family environmental effects, things like parental education, wealth, neighborhood quality, or cultural capital that correlate with genes but aren’t directly caused by them. For height and BMI, the scores work just as well within families, which suggests those traits are more purely biological. The IQ/education scores are picking up a mix of direct genetic effects AND all the environmental advantages that tend to cluster in families with certain genetic variants.
The SES finding is key here. A lot of what polygenic scores for education capture is basically “comes from a family with resources.” That doesn’t make them useless if you’re trying to predict who’ll get a PhD in the general population, the score works. But it does mean we need to be careful about causal claims. When people say “this gene variant causes higher intelligence,” they’re partly measuring “this gene variant is more common in families that can afford tutors, have educated parents, live in good school districts, etc.” The within-family component is closer to the true direct genetic effect, which is why it’s smaller.
