Ideas & History · Genetics, education and inequality
What DNA can tell us about schooling, and what it cannot
A child’s education can be associated with parental DNA the child never inherited. That finding opens a serious argument about genes, social inheritance and the limits of prediction.
In 2018, a genetic study produced a finding that should interrupt easy talk about educational destiny. Researchers examined parental DNA variants that their children had not inherited. A score built from those variants was nevertheless associated with the children’s educational attainment. The study, by Augustine Kong and colleagues, included 21,637 offspring with at least one genotyped parent. The finding concerns an association with education, not DNA somehow passing to the child by another route. [1]
The authors called this genetic nurture: parental genetic differences can be associated with characteristics of the environment parents provide. Their estimate for the nontransmitted score was 29.9% of the transmitted score’s estimated effect. That is a comparison between score associations, not a claim that parenting causes 29.9% of education. Nor did the study identify every intervening behaviour. Its significance is that an association involving DNA can partly reflect a social pathway. [1]
This makes the politics of inheritance more demanding. A family transmits more than a sequence of DNA. For educational research, the problem is to distinguish pathways that travel together. For public debate, the danger is to turn a useful statistical prediction into an account of what a child inherently is, or a verdict about what society should provide.
Development, inheritance and the meritocratic claim
In his 1982 Tanner lectures, Biological Determinism, Richard Lewontin challenged explanations that made social hierarchy appear biologically inevitable. His target was also political: a society promising equal opportunity could invoke innate ability to justify inequalities it continued to reproduce. His scientific argument centred on development, the history through which genes, environments and contingency produce a trait. [2]
Lewontin used the idea of a reaction norm: the pattern of outcomes a genotype produces across environments. Genotypes can respond differently to altered conditions, and their relative ranking can change. An observation under one set of conditions therefore cannot establish each individual’s maximum possible development. The argument concerns what can be inferred from an observation, while recognising that genes matter. [2]
Later research must be allowed to challenge a classic. The political attraction of Lewontin’s argument cannot settle a modern empirical question, just as a successful genetic prediction cannot settle a political one. We need to ask which result a study actually supplies: a forecast, an explanation of a pathway, or evidence about changing a condition.
What a genetic prediction measures
Heritability is often pulled into this argument as though it were a label attached to a person. As geneticists Peter Visscher, William Hill and Naomi Wray explain in their 2008 review, it is a ratio of variances in a specified population under particular conditions and assumptions. It does not divide an individual into a genetic percentage and an environmental percentage. It also does not, by itself, explain why average outcomes differ between populations. [3]
The measure remains useful, including for studying evolutionary change and predicting responses to selection. High heritability is compatible with environmental change shifting an average: the review uses changes in human height to illustrate the point. Variation among people observed together and a change in the conditions under which people develop are different objects of explanation. [3]
A major 2022 study by Aysu Okbay and colleagues analysed educational attainment in roughly three million people of European genetic ancestries. A polygenic index, combining associations across many DNA variants, predicted about 12–16% of the variation in completed education in the relevant validation samples. That is meaningful predictive information. It is neither a measure of a particular child’s potential nor a percentage of that child’s schooling caused by genes. [4]
Within-family analyses sharpened the result. After controls for parental indices, the score’s association with education was roughly half as large. This refers to an association coefficient, not half the variance explained. Prediction was weaker in the African-ancestry cohorts studied. The index therefore could not be read as a context-free measure of potential. The ancestry comparison measures the transferability of a predictor; it does not explain group differences in educational attainment. [4]
What comparisons within families can establish
A separate 2022 study by Laurence Howe and colleagues compared population estimates with estimates drawn from differences among siblings. It combined roughly 178,000 siblings from 19 cohorts and examined 25 outcomes. Associations were smaller within families for educational attainment and several other traits, while many molecular traits gave broadly consistent results. The comparison reduces influences shared through family background and population structure. It does not assume that every genetic association is equally vulnerable to those influences. [7]
The authors also identified a remaining problem: siblings can affect one another’s environments. A sibling’s genotype may therefore have an indirect effect on the other sibling’s outcome. Comparing relatives improves the design without turning the family into an experiment in which every social pathway has been removed. [7]
This helps locate the disagreement precisely. If the question is who is likely to complete more education under existing arrangements, a predictor can exploit several connected pathways. If the question is what would happen after changing one arrangement, those same connections can become a source of error. A score that carries information about parental resources may predict accurately while giving a misleading account of what the child would achieve with different resources. Better prediction and better explanation can require different comparisons. The distinction is especially consequential when institutions propose to act on the prediction.
These are substantial achievements for genetic research. They also leave the question of intervention open. Even an informative account of inherited influences does not measure how well a school reform would work. Changing an institution changes the opportunities people encounter. To judge that change, we need evidence about their development under the altered conditions.
What changes when schooling changes?
That requires a different research design. Stuart Ritchie and Elliot Tucker-Drob’s 2018 meta-analysis examined 42 datasets covering more than 600,000 people. It combined studies controlling for earlier cognitive scores, changes in compulsory schooling policy, and school-entry age cut-offs. These designs sought to distinguish the effect of additional education from the tendency of people with higher initial scores to remain in education longer. [5]
Depending on the design and adjustments, the estimated increase was roughly one to five points on an IQ scale for an additional year of education. The authors also set limits: test-score changes need not establish a change in general intelligence; a reform’s estimated effect may apply to the people whose schooling it altered; gains cannot simply be added indefinitely. The result supports education’s capacity to affect measured performance. It does not predict a precise return from any particular South African policy. [5]
A curriculum change, an extra year of compulsory attendance and a programme with better resources are distinct interventions. Each changes a different educational experience. A finding about one cannot supply a universal conversion rate between years at school and intellectual development. The political demand for better education still requires specific choices, followed by an honest assessment of their effects.
Prediction in whose service?
An advocate of educational genetics can raise a serious objection: withholding useful information could itself harm children who need support. That possibility deserves examination. But a forecast of attainment is not yet evidence about which assistance will help a particular learner, whether the score adds information beyond observed needs, or whether its use improves outcomes. A proposal to use such information must establish those additional links. A politically troubling result should not be suppressed; an impressive prediction should not acquire authority over school placement simply because it is expressed in DNA.
The purpose of classification also matters. Identifying a barrier in order to remove it and ranking children in order to ration opportunity are different institutional acts, even when they use the same numbers. The second can make its prediction partly self-fulfilling by restricting the experiences through which learners develop. Evaluating that risk requires following the consequences of the decision, including errors and who bears them. The relevant public question is how knowledge is used to enlarge educational possibility and how institutions remain answerable when their classifications fail.
There is also a political judgement that no genome-wide study can make for us. Equal access to education need not be justified by a prediction that everyone will obtain equal marks. It can rest on a commitment that people deserve the means to develop, participate and exercise freedom. Otherwise an argument for equality becomes vulnerable to every measured difference between human beings.
A democratic education system must take both evidence and possibility seriously. Genetic research can improve prediction and explain developmental pathways. Public institutions must decide which barriers to remove and which opportunities to create, then examine the results. Treating a prediction under present conditions as the limit of a child’s future closes that inquiry before the work of education has begun.
Sources and further reading
- Kong et al., The nature of nurture: Effects of parental genotypes (opens in a new tab) Science 359 (2018), pp. 424–428. The reported effect ratio is not a percentage of an individual’s education.
- Richard Lewontin, Biological Determinism, Tanner Lectures (opens in a new tab) Tanner Lectures, University of Utah, 31 March and 1 April 1982. See the argument on development and norms of reaction.
- Visscher, Hill and Wray, Heritability in the genomics era: concepts and misconceptions (opens in a new tab) Nature Reviews Genetics 9 (2008), pp. 255–266, especially Box 2.
- Okbay et al., Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals (opens in a new tab) Nature Genetics 54 (2022), pp. 437–449. See the direct-effect and ancestry-portability analyses.
- Ritchie and Tucker-Drob, How Much Does Education Improve Intelligence? A Meta-Analysis (opens in a new tab) Psychological Science 29 (2018), pp. 1358–1369. Study designs, estimates and limitations.
- Harvard Faculty of Arts and Sciences, Richard Charles Lewontin memorial minute (opens in a new tab) Faculty memorial adopted 6 February 2024. Biographical context and research history.
- Howe et al., Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects (opens in a new tab) Nature Genetics 54 (2022), pp. 581–592. Family comparisons and the limits of isolating genetic effects.
