Social behaviour is often spoken about as though it were one stable personal characteristic.
But the way a child behaves with classmates may not look the same as the way they behave at home, and what a parent notices may differ from what a teacher or the young person themselves reports.
New genetic research suggests that these differences are not merely measurement noise.
A study published in Nature Human Behaviour on 21 September 2026 analysed 491,246 repeated observations from 73,321 people across 25 cohorts, following social behaviour from early childhood into young adulthood.
The researchers found that genetic associations with social behaviour changed depending on the type of social behaviour being measured, the age of the participant and who was reporting the behaviour.
They identified six genome-wide significant loci, estimated single-nucleotide-polymorphism-based heritability at roughly 2% to 7% across contexts, and found that the broader genetic architecture could be organised into four factors reflecting differences in social domain and reporter.
The result does not mean there are six “social behaviour genes”. Instead, it shows that genetic influences on social behaviour appear to be distributed, modest and highly dependent on context.
Social behaviour is not one single trait
The study focused on two broad domains of social behaviour.
The first was low prosocial behaviour, which captures difficulties with behaviours such as helping, sharing, comforting or cooperating with others.
The second was peer and social problems, which captures difficulties in relationships with peers and broader social functioning.
These domains overlap, but they are not identical.
A child may struggle to initiate prosocial behaviour while still having stable friendships, or may behave generously while also experiencing exclusion or conflict with peers.
Treating all of these behaviours as one outcome can therefore hide meaningful differences.
The same person can look different depending on who is asked
The researchers also distinguished between different reporters.
Social behaviour was assessed through parent reports, teacher reports and self-reports.
This matters because each reporter observes the participant in a different environment.
Parents see behaviour at home and in family settings. Teachers observe children in classrooms and peer groups. Self-reports capture experiences that may not be visible to either adults or peers.
The new analysis found that genetic effects varied systematically across these reporting contexts.
That suggests that disagreement between reporters may sometimes reflect genuine context-specific behaviour rather than one reporter simply being wrong.
The study covered development from age 2 to 29
The researchers combined data spanning early childhood, middle childhood, adolescence and young adulthood.
Participants ranged in age from 2 to 29 years across the contributing cohorts.
This allowed the team to ask whether genetic associations remained stable as social expectations and social environments changed with age.
They did not.
Some genetic effects became stronger or weaker depending on developmental stage.
This is important because social behaviour at age 4 is not directly comparable with social behaviour at age 18.
The demands of friendship, independence, cooperation and self-regulation all change substantially across development.
Nearly half a million repeated observations were combined
The study drew on 491,246 repeated observations from 73,321 individuals across 25 independent cohorts.
Because many participants were measured more than once, the researchers were able to model how associations varied across age rather than treating development as a single snapshot.
The data generated 195 sets of genome-wide association summary statistics.
To make those measures comparable, questionnaire items from different studies were aligned using categories from the World Health Organization’s International Classification of Functioning, Disability and Health.
The researchers then used a genome-wide meta-regression framework that explicitly modelled differences in social domain, reporter and age.
Six genomic regions reached genome-wide significance
Across the context-sensitive analyses, the researchers identified six independent loci that reached the conventional threshold for genome-wide significance.
One involved variation within CADM2, a gene that has previously been linked to a range of behavioural and psychological traits.
The important point is not that one locus explains social behaviour.
Each individual genetic variant has a very small effect, and the study found substantial heterogeneity in how the six loci behaved across age, social domain and reporter.
The same variant could therefore be more informative in one context than another.
Heritability was modest, not deterministic
The researchers estimated SNP-based heritability at approximately 2% to 7% across the social-behaviour measures.
This figure is easy to misinterpret.
Heritability does not describe how much of one person’s behaviour is “caused by genes”.
It describes how much variation in a trait across a population can be statistically associated with measured genetic differences under the environmental conditions in which that population was studied.
A heritability estimate can therefore change across populations, ages and environments.
The relatively modest estimates in this study reinforce the point that social behaviour reflects many influences beyond common genetic variants, including family relationships, peers, schools, culture, opportunity and life experience.
The genetic architecture separated into four broader factors
When the researchers examined the broader pattern of genetic correlations, the data did not collapse into one universal social-behaviour factor.
Instead, the genetic architecture was best represented by four factors.
These reflected differences associated with the social domain being assessed and with who was reporting the behaviour.
This supports the central argument of the study: context is not something that should automatically be averaged away.
Two measures may both be labelled “social behaviour” while drawing on partly different genetic and environmental processes.
Polygenic scores worked better when the context matched
The researchers also tested polygenic scores in independent cohorts containing 16,305 individuals.
A polygenic score combines information from many genetic variants into a single statistical estimate.
The scores were better at predicting contextual variation when the discovery measure and the target measure were more closely matched.
For example, a score derived from one type of reporter or social domain may be less informative when applied to a very different social context.
This has methodological consequences for behavioural genetics.
Combining very different measures into one large analysis may increase sample size but can also dilute context-specific information.
The same issue matters when linking social behaviour to mental health
Social difficulties are associated with several neurodevelopmental and psychiatric conditions.
The researchers therefore examined genetic correlations between social behaviour and conditions including ADHD, autism, major depression, schizophrenia and bipolar disorder.
The pattern of shared genetic associations depended on developmental stage and social context.
This does not mean that social difficulties are early forms of any of these conditions.
Genetic correlation indicates statistical overlap in genetic influences across traits. It does not establish that one trait causes the other, nor does it allow a diagnosis to be inferred from a social-behaviour score.
Who reports a child’s behaviour can matter scientifically
Parent and teacher reports are often combined or compared in developmental research.
When those reports disagree, researchers may be tempted to treat the disagreement as measurement error.
The new findings suggest a more nuanced interpretation.
A child may genuinely express different aspects of social behaviour in different settings.
Genetic associations that are stronger in one reporting context may reflect that context-specific expression rather than a universally observable trait.
This does not mean one reporter is more accurate than another in every situation.
It means each reporter may be capturing a different piece of the child’s social world.
The findings do not support genetic labelling of children
Research linking genetics and behaviour can easily be interpreted too strongly.
The study does not provide a genetic test that can determine whether a child will be prosocial, have friendship difficulties or later develop a mental-health condition.
The identified variants have small effects, and the polygenic architecture is spread across many parts of the genome.
More importantly, the genetic associations themselves changed across age, social domain and reporter.
That instability is the opposite of a simple deterministic interpretation.
Social behaviour develops through ongoing interactions between biology and environment.
The ancestry limitation is substantial
The main genome-wide analysis was conducted in cohorts of European ancestry.
The researchers later tested polygenic scores in both European- and African-ancestry samples and reported some cross-ancestry transferability.
However, polygenic scores generally perform less reliably when applied to populations whose ancestry differs from the populations used to generate them.
This means the numerical estimates and predictive performance in the study should not simply be assumed to apply equally across global populations.
For countries such as South Africa, where genetic ancestry is highly diverse, broader representation in discovery datasets is particularly important.
Large samples do not eliminate measurement differences
The study’s enormous number of observations is a major strength.
But the data came from many cohorts using different questionnaires and study designs.
The researchers harmonised these measures using a common classification framework, but harmonisation cannot make every questionnaire identical.
Some apparent context effects may therefore still reflect differences in how behaviour was measured.
The meta-regression approach was designed specifically to model this heterogeneity rather than ignoring it, but no statistical approach can remove every source of measurement variation.
The practical value is mainly in better research design
The immediate value of the findings is not a new clinical test or classroom tool.
It is a warning about how researchers study complex behaviour.
If genetic effects differ by age, social domain and reporter, future studies may lose information when they combine all measures into one broad outcome.
Matching the context of discovery and prediction may produce more meaningful genetic analyses.
The same principle could also apply beyond genetics.
Social behaviour may be better understood when researchers treat home, school, peer relationships and self-perception as related but distinct environments.
The bigger lesson is that context is part of the trait
The study began with a familiar question: which genetic differences are associated with social behaviour?
Its more important answer may be that the question is incomplete without specifying whose behaviour, at what age, in which social domain and according to whom.
The genetic architecture of social behaviour was not fixed across these conditions.
That makes social behaviour harder to reduce to a single score, but it may also make future research more realistic.
Human behaviour does not occur outside a social environment, and the new findings suggest that genetic research works better when that environment is treated as part of the phenomenon rather than background noise.
Source Information
Study Title: Genome-wide analysis of social behaviour across social domains, reporters and developmental stages: a meta-regression approach
Authors: Lucía de Hoyos, Fenja Schlag, Sanjeevan Jahagirdar, Elizabeth C. Corfield, Andrea G. Allegrini, Danielle A. G. Admiraal, Eveline L. de Zeeuw, Ilja M. Nolte and colleagues
Journal: Nature Human Behaviour
Published: 21 September 2026
Dataset: 491,246 repeated observations from 73,321 individuals across 25 cohorts, producing 195 sets of genome-wide association summary statistics. The main discovery analysis used European-ancestry cohorts, with follow-up polygenic-score analyses in independent European- and African-ancestry samples totalling 16,305 individuals.
Method: Genome-wide mixed-effects meta-regression was used to model genetic associations with low prosocial behaviour and peer/social difficulties across parent, teacher and self-reports and across development from age 2 to 29. The researchers also conducted genomic factor analysis, polygenic-score prediction and genetic-correlation analyses with neurodevelopmental and psychiatric conditions.
Main finding: Six independent genome-wide significant loci were identified, while SNP-based heritability ranged from approximately 2% to 7%. Genetic effects varied systematically by social domain, reporter and age, and the broader genetic architecture separated into four context-sensitive factors. Polygenic scores predicted social-behaviour variation more accurately when discovery and target contexts were better matched.
DOI: 10.1038/s41562-026-02551-z
