What Heritability Really Means

Heritability is one of the most misunderstood numbers in genetics. It describes variation among people in a defined population under particular conditions. It does not divide one person's trait into a genetic percentage and an environmental percentage. It does not measure destiny.

If you read that a trait has an estimated heritability of 0.70, the careful interpretation is that genetic differences account statistically for about 70 percent of observed variation in that study population, under that model and environment. Change the population, measurement, age, or environment, and the estimate can change.

Variation is the object being measured#

Heritability concerns why people differ from one another in a sample. It is usually written as a ratio: genetic variance divided by total phenotypic variance. The exact numerator depends on whether the estimate concerns additive genetic effects, all genetic effects, or variation tagged by measured markers.

This is a statement about variance, not average level. Two populations can have different average heights and similar heritability within each population. That does not show that the difference between their averages is genetic.

An analogy helps. Everyone in a class may receive nearly the same high-quality instruction. Differences in test performance could then be more strongly associated with other differences among students. The high share of variation attributed to those differences would not mean instruction was unimportant. Its importance is partly hidden because instruction varied little.

The denominator matters. If environmental conditions become more variable, environmental variance can rise and heritability can fall even when genes do not change, and if environments become more uniform, the genetic share of remaining variation can rise.

Heritability is not “how genetic” a person is#

You do not have 60 percent heritable blood pressure. The statistic belongs to a population distribution, not to any one person in it.

Nor does the unexplained portion map neatly to personal choice or family influence. Environmental variance includes diet, infection, medication, work, social conditions, measurement error, random developmental events, and other differences not assigned to the genetic component by the model.

Genetic and environmental contributions are not always separable. A genetic tendency can influence which environments a person encounters, and an environment can alter how genetic variation is expressed. Statistical partitions depend on assumptions about these relations, and in a clinical conversation, the specific risk factors you carry and the interventions that work on them are usually more actionable than a heritability estimate. The number describes population variation; it does not prescribe care.

Narrow-sense and broad-sense heritability#

Narrow-sense heritability focuses on additive genetic variance, the cumulative contribution of allele effects under a model. It is important in breeding and in predicting resemblance across generations.

Broad-sense heritability includes additive effects plus dominance, interaction among loci, and other genetic components. In human research, estimating all these components cleanly is difficult.

Reports should name which quantity they estimate. A generic “heritability” label can hide important differences in method and interpretation.

On the observed scale for a continuous trait, the variance ratio is relatively direct, and for a binary outcome such as presence or absence of disease, investigators may use a liability-threshold model. Heritability on the liability scale depends on assumptions and population prevalence. It should not be read as the fraction of cases caused by genes.

Twin and family studies#

Twin studies compare resemblance between identical twins, who share nearly all DNA sequence, and fraternal twins, who share on average about half of segregating variants. Family and adoption designs use other degrees of relatedness.

These designs infer genetic and environmental components under assumptions. One common assumption is that relevant environmental similarity is comparable enough across twin types. Assortative mating, gene-environment correlation, diagnostic measurement, and representativeness can affect estimates.

Twins and families also share social and prenatal environments in complex ways. A model can allocate covariance incorrectly if its structure omits an important source. Large and replicated family studies can be informative, but their estimates remain model-dependent. A narrow confidence interval does not cover uncertainty from violated assumptions.

SNP heritability#

Genome-wide data permit estimates of how much trait variation is tagged by measured single-nucleotide polymorphisms, and methods compare genetic similarity among many people with phenotypic similarity or use summary statistics from genome-wide association studies.

SNP heritability often captures common additive variation represented by the genotyping platform and reference data. It may miss rare variants, structural variants, poorly tagged regions, dominance, interactions, or other components included in some family estimates.

Population structure can confound genetic association when ancestry-related marker differences track environmental or social differences. Statistical methods reduce this risk but may not eliminate it. Relatedness, ascertainment, and measurement also matter. An estimate based on one ancestry composition or healthcare system may not transport. Underrepresentation in genomics can make both heritability and downstream prediction less reliable across populations.

Why heritability does not equal genetic prediction#

A trait can be highly heritable while available genetic variants predict it poorly, and heritability may be spread across thousands of variants with tiny effects, rare variants not measured, or components difficult to estimate.

A polygenic score sums variant weights to predict relative propensity. Its performance depends on discovery data, linkage patterns, phenotype definition, target population, and clinical context. A score explains only part of risk and can lose accuracy across ancestry groups.

Heritability sets neither a simple ceiling nor a guarantee for prediction. Different estimators cover different genetic components. Prediction also needs calibration and absolute risk, which depend on age, prevalence, environment, and clinical factors.

The guide to genome-wide association studies explains why association signals are not deterministic. The polygenic-risk-score guide covers additional work needed before individual use.

Missing heritability#

Family studies sometimes estimate more heritability than known associated variants explain. This gap became known as missing heritability.

Possible contributors include many common variants with effects too small for earlier studies, rare variants, structural variation, incomplete tagging, gene-gene interactions, gene-environment interaction, phenotype error, and upward bias in family estimates. Larger genome-wide studies and better methods have explained more variance for some traits, but the gap is not one problem with one solution.

Finding variants associated with more variance does not automatically reveal mechanism. A locus may tag several variants, affect a regulatory region, or act differently across tissues and stages, and getting from variance accounting to biology takes mechanistic work, diverse cohorts, functional experiments, and replication.

High heritability is compatible with effective intervention#

Phenylketonuria provides a classic conceptual example. Its cause involves genetic variants, yet dietary treatment can prevent severe consequences when begun early. Genetic origin does not imply an untreatable outcome.

Likewise, vision can be strongly influenced by inherited biology while eyeglasses change function. A trait can respond to an intervention even if genetic differences account for much of its variation before intervention.

If an intervention reaches everyone similarly and improves the average, it may shift the population without reducing heritability much, and heritability is about differences around the average, not whether the average can move.

The reverse is also possible. A new environmental inequality can increase variance and change the estimate. Heritability therefore cannot answer “would a policy or treatment work?” That requires intervention evidence.

Gene-environment interaction#

Gene-environment interaction means the effect of genetic variation differs across environments, or the effect of an environment differs across genotypes. A variant can matter more under one diet, infection pattern, medication, or developmental condition.

Statistical interaction depends on the scale used. An interaction on an additive scale may not appear on a multiplicative scale. Studies need prespecified hypotheses, adequate sample size, reliable environmental measurement, and replication.

Gene-environment correlation occurs when genetic differences are associated with environmental conditions. This can arise through family context, individual behavior, or responses from others. It complicates a simple variance split. Neither concept means that every observed subgroup difference is genetic. Environmental determinants should be measured directly where possible rather than inferred from population labels.

Metabolic disease as a practical example#

Type 2 diabetes has substantial genetic contribution, but risk also changes with age, adiposity, diet, physical activity, sleep, medicines, pregnancy history, and social conditions. Genetic susceptibility operates within these contexts.

A heritability estimate for diabetes depends on case definition, prevalence, age, ancestry composition, and environment. It does not say what fraction of one person's condition was caused by genes or whether prevention and treatment can help.

Specific monogenic diabetes forms are a different question; a pathogenic variant can have a large effect and may change treatment, but identifying such a condition requires clinical criteria and appropriate testing. Population heritability does not diagnose a monogenic disorder. Risk communication should keep four things apart: your family history, a specific variant you carry, a polygenic score, and a heritability figure. They are related but not equivalent.

Population labels require care#

Race and ethnicity are social and administrative categories with changing meanings. They are not precise genetic groupings. Genetic ancestry is continuous, mixed, and represented imperfectly by reference panels.

Observed group differences can reflect structural conditions, geography, care, discrimination, environment, and measurement as well as allele frequencies. Assigning an unexplained difference to genetics is not a valid default.

The National Academies recommends that researchers justify population descriptors, avoid typological thinking, and measure relevant environmental variables rather than using labels as proxies. The same care should apply when heritability results are communicated publicly, and it means a group estimate should not be used to infer your genotype or your clinical needs. Direct information about you is preferable when it exists.

How to read a heritability claim#

Ask which population, age range, environment, and period were studied. Identify the phenotype definition and whether it is continuous, binary, self-reported, or clinically measured.

Find the estimator: twin, family, SNP-based, sequencing, or another method. Check whether the result is narrow-sense, broad-sense, observed-scale, or liability-scale. Read uncertainty and assumptions.

Look for ancestry diversity, relatedness control, population structure, measurement reliability, and replication. Ask whether the estimate is being transported beyond its study population.

Then inspect the conclusion. A heritability result can point you toward the sources of variation worth investigating, though it cannot by itself establish a causal variant, justify genetic determinism, predict an individual, or determine intervention value.

Replace destiny with a precise question#

The useful questions are narrower than “genes or environment?” Which genetic variants are associated? Through which mechanism? In which population and conditions? How well can risk be predicted? Does an intervention change outcomes? Who can access it?

Heritability offers one population-level answer about variation. Keeping that answer within its boundaries makes genetics more informative and less misleading. The site's research overview connects this distinction to causal inference and evidence appraisal.

References#

1. MedlinePlus Genetics: What is heritability?
2. Heritability in the genomics era: concepts and misconceptions
3. The heritability of human disease: estimation, uses, and abuses
4. Finding the missing heritability of complex diseases
5. Open problems in human trait genetics
6. National Academies guidance on population descriptors in genetics and genomics

For your own health, talk with your clinician or a qualified genetics professional.*