Skip to main content
KBS_Icon_questionmark link-ico
Two DNA helices. ;

What do polygenic scores really predict? Looking inside families to understand genetic prediction

Yujing Lin

PhD Student, Institute of Psychiatry, Psychology & Neuroscience

27 January 2026

Over the past decade, behavioural genetics has made great advancements in predicting human behaviour using DNA. By combining millions of genetic variants into polygenic scores (PGS), researchers can now predict outcomes such as cognitive abilities, educational achievement, educational attainment, and mental health conditions. For cognitive and educational outcomes in particular, these scores can explain up to 18% of differences between people.

A polygenic score is a genetic predictor for individual differences. For example, a polygenic score for cognitive ability reflects an individual’s genetic propensity for cognitive outcomes compared with others in the population. But most PGS studies compare unrelated individuals. Implicitly, we may assume that genetic differences operate in the same way within families as they do across the population.

Our recent study, conducted at the Institute of Psychiatry, Psychology & Neuroscience (IoPPN) at King’s College London, tries to address this assumption by asking what exactly polygenic scores are capturing.

Looking inside families

twin kids

Using data from the Twins Early Development Study (TEDS), a large longitudinal birth cohort, we examined polygenic score prediction from childhood to early adulthood (ages seven to 26) at three different levels:

  • Within families: genetic differences between siblings.
  • Between families: genetic differences that siblings share, which reflect family-wide influences such as socioeconomic advantage.
  • At the population level: genetic differences among unrelated individuals, which is how genetic prediction is usually studies.

Our study is built on a previous study incorporating more characteristics (called phenotypes) and ages. Understanding the conceptual idea is the most difficult part. Within-family genetic effects are calculated by comparing siblings, in our case twins. The genetic differences among siblings are almost random, coming purely from the random process of inheritance during meiosis and sometimes described as the “Mendelian lottery”. In contrast, differences between families are systematic: they reflect stable advantages or disadvantages that affect all siblings in a household.

By comparing polygenic score prediction within families to prediction in the population, we can estimate how much of genetic prediction reflects random inheritance versus broader family-level processes.

What we found

Our results showed that educational and cognitive outcomes are significantly influenced by between-family genetic effects. On average, about half of the population-level prediction from polygenic scores for these outcomes came from between-family differences.

In other words, genetic prediction of cognitive abilities and educational outcomes reflects both:

  • random genetic differences between siblings, and
  • systematic family-level advantages that are indexed by DNA.

We also found that while the overall predictive power of polygenic scores increases with age, the balance between within-family and between-family effects remains stable over development.

Prediction versus explanation

Since within-family differences are more “random” while between-family differences are more “systematic” operating through social and environmental processes, some researchers may argue that only within-family effects are “valid” genetic effects and that between-family effects are confounding. However, this interpretation can be misleading, as the most “random” genetic effects do not operate in a vacuum. Instead, genes always exert their influence through environmental context.

We argue for a more nuanced view. If our goal is to explain how a trait develops, separating within-family and between-family effects is useful. But if our goal is to predict outcomes, then combining these effects can be both valid and informative. After all, between-family influences, whether biological or not, matter profoundly for people’s lives.

Ironically, the context where within-family prediction would be relevant is embryo selection, a highly controversial application of polygenic scores, particularly for traits like intelligence and educational achievements.

For most real-world uses of polygenic scores, including research, education, and public understanding of genetics, between-family effects are part of the signal, not a nuisance to be ignored.

Public engagement

Yujing Lin The Genetics Behind IQ podcast episode thumbnail

Our study was the topic of an episode of the IQ and Human Intelligence podcast. The podcast, created by Riot IQ (developers of the RIOT IQ test), features in-depth interviews and discussions on cutting-edge research on topics related to intelligence. As I noted in the interview, “We often think of polygenic scores as capturing causal genes. But many of these associations operate through environmental pathways. Genetics doesn’t operate in a vacuum.”

The takeaway

Polygenic scores are powerful predictors, but they do not predict in a simple or purely biological way. By looking inside families, we can better understand what genetic prediction really reflects. Clarifying how genetic influences operate across family and population levels brings us closer to using polygenic scores responsibly and understanding what they truly tell us about human behaviours.

In this story

Yujing Lin

Yujing Lin

PhD Student

Latest news