A DNA Test Can Predict 11% of How Far You Went in School. The Other 89% Is the Interesting Part
Polygenic scores for educational attainment now draw on more than a million people and predict a real, replicable slice of outcomes. Here is exactly how big that slice is, what it does not contain, and why the number falls apart when you compare siblings.
In 2016, a consortium published a genome-wide association study of educational attainment — years of schooling completed — and found 74 loci reaching significance across roughly 293,000 people (Okbay et al., Nature, 2016, PMID: 27225129).
Each variant was almost comically small. The largest effects corresponded to a few weeks of additional schooling.
Two years later the same field ran the study again in 1.1 million individuals. With that much data, the polygenic scores built from the results explained 11–13% of the variance in educational attainment and 7–10% of the variance in cognitive performance (Lee et al., Nature Genetics, 2018, PMID: 30038396).
That is a real number. It is also the beginning of the argument, not the end of it.
What 11% means, and what it doesn't
It means: across a large population of people of European ancestry, a genetic score explains about a ninth of the differences in how long people stayed in school.
It does not mean any individual's education was 11% genetic. It does not mean the remaining 89% is "environment" in a tidy sense. And it certainly does not mean a score can tell you what a particular child is capable of.
A parallel effort focused on measured intelligence rather than schooling, covering 269,867 individuals, found 205 associated genomic loci (Savage et al., Nature Genetics, 2018, PMID: 29942086). Same picture: many variants, each contributing very little.
The sibling problem
Here is the finding that should temper any confidence in these scores.
When you compare siblings — people raised in the same household, by the same parents, in the same neighbourhood — the predictive power of an educational-attainment polygenic score drops substantially compared with comparisons between unrelated strangers.
That gap is informative. A portion of what the score picks up in the general population is not the direct biological effect of those variants at all. It reflects the fact that genes correlate with the environment a child is born into: parental education, income, neighbourhood, school quality. The score is partly measuring a family's circumstances wearing a genome as a disguise.
Educational attainment is not intelligence
Worth stating clearly, because the headlines rarely do. "Years of schooling" is a social outcome shaped by war, wealth, gender norms, migration, illness and luck. It correlates with cognitive test performance, but it is not the same thing, and neither is the same thing as intelligence in any everyday sense.
These studies were also conducted overwhelmingly in people of European ancestry. Polygenic scores transfer poorly across ancestry groups — a score built in one population loses much of its accuracy when applied to another. That is a limitation of the data, not a statement about anyone.
What the app reports
The Intelligence App applies the findings from Okbay and colleagues to your genotype and shows where you fall on the distribution described in that research. Two companion apps use the same literature on narrower outcomes: Math Ability, built on Lee et al., and Educational Attainment.
What it gives you is a position on one measured axis, derived from population-scale statistics, with the source study cited. What it cannot give you is a ceiling, a verdict, or a reason to treat any child differently.
The most honest reading of the last decade of this research is that the genome contributes to cognitive outcomes in ways that are real, measurable, diffuse across thousands of variants — and far weaker at the level of a single person than at the level of a population.
Eleven percent is a genuine scientific achievement. It is also not very much to know about someone.