Twin Studies

Why twin studies matter

• They answer "nature or nurture" without genotyping anyone. Long before the human genome was sequenced, the MZ-versus-DZ contrast could reveal that a trait had a genetic basis. That is a rare and cheap kind of leverage: two natural experiments — a genetic clone (MZ) and an ordinary sibling pair (DZ) — that are perfectly age-matched and cohabiting.
• They defined behavioral genetics. Estimates that intelligence is roughly 50 to 80% heritable in adulthood, that the "big five" personality dimensions run around 40 to 50% heritable, and that schizophrenia, bipolar disorder, and autism carry substantial genetic risk all rest heavily on twin data accumulated across the twentieth century.
• They discovered the shrinking shared environment. One of the field's most counterintuitive and robust results is that for many adult psychological traits the common family environment (C) contributes almost nothing to variation, while the non-shared environment (E) — the experiences siblings do not hold in common — dominates. This reframed how we think about parenting and outcome.
• They quantify disease risk. Concordance rates tell a clinician how much of a disorder is inherited. Type 1 diabetes shows roughly 30 to 50% MZ concordance; multiple sclerosis around 25 to 30%; breast cancer far lower. These numbers ground genetic counseling and set expectations for how much genome-wide association studies can eventually explain.
• They power modern designs. The co-twin control design uses MZ pairs discordant for an exposure to control perfectly for genotype and shared upbringing — an elegant way to test causality in observational data. National twin registries in Sweden, Denmark, and the Netherlands, with tens of thousands of pairs, feed epidemiology on cancer, longevity, and psychiatric illness.
• They anchor missing heritability. When GWAS explains only a fraction of the heritability that twin studies predict, the gap — the "missing heritability" problem — becomes a research agenda in itself, pointing to rare variants, non-additive effects, and structural variation that common-SNP arrays miss.

How a twin study works, step by step

The logic begins with a variance equation. In the standard ACE model, the total phenotypic variance of a trait is decomposed into three latent sources: additive genetic variance (A), shared or common environment (C), and non-shared or unique environment plus measurement error (E), scaled so that A + C + E = 1. Narrow-sense heritability (h²) is the proportion of variance due to additive genetic effects, A; broad-sense heritability (H²) additionally includes dominance and epistatic effects.

The design then exploits the known genetic relatedness of the two twin types. Monozygotic twins correlate 1.0 for their additive genes; dizygotic twins correlate 0.5, exactly as full siblings do. Both twin types are assumed to share their common environment fully (the correlation for C is 1.0 in both cases, under the equal-environment assumption), while the non-shared environment is by definition uncorrelated. This yields two simultaneous expectations: the MZ correlation equals A + C, and the DZ correlation equals ½A + C.

Subtracting the two equations isolates the genetics. Because r_MZ − r_DZ = A − ½A = ½A, doubling that difference recovers the additive genetic variance — this is Falconer's formula: h² ≈ 2(r_MZ − r_DZ). The shared environment then falls out as C = 2·r_DZ − r_MZ, and the non-shared environment is whatever is left over: E = 1 − r_MZ. That last term is beautifully direct — anything two genetically identical people raised in the same house do not share must come from unique experience or noise.

For categorical traits — a disease you either have or don't — correlations are replaced by concordance rates. Pairwise concordance is the fraction of affected pairs in which both twins are affected; probandwise concordance counts affected co-twins per ascertained proband and is the figure directly comparable to population risk. A higher MZ than DZ concordance signals heritability, and a liability-threshold model converts these concordances into an underlying heritability on a continuous liability scale.

Finally, modern practice replaces the back-of-envelope Falconer estimate with structural equation modeling. Software such as OpenMx fits the full ACE model (and reduced AE, CE, or ADE variants) to the raw covariance structure across many pairs, produces maximum-likelihood estimates with confidence intervals, and uses likelihood-ratio tests to decide whether C or A can be dropped without significantly worsening fit. This handles sex differences, age moderation, and multivariate traits that a single subtraction cannot.

Identical vs fraternal twins

Feature	Monozygotic (identical)	Dizygotic (fraternal)
Origin	One zygote splits within ~14 days	Two eggs, two sperm, same cycle
Shared nuclear DNA	~100% identical by descent	~50% of segregating genes
Genetic correlation (additive)	1.0	0.5
Sex	Always the same	Same or different
Placenta / chorion	Often monochorionic	Always dichorionic
Fraction of spontaneous twins	~1/3 of pairs (rate ~3–4 per 1000, fairly constant)	~2/3 of pairs (rate varies with age, ancestry, IVF)
Role in the design	Upper bound: A + C	Comparison: ½A + C

Partitioning the variance: A, C, and E

Component	What it captures	MZ correlation	DZ correlation	Typical adult finding
A — additive genetics	Sum of allele effects transmitted across generations	1.0	0.5	Large for height, IQ, personality
C — shared environment	Family, home, neighborhood, parenting shared by both twins	1.0	1.0	Often near zero in adulthood
E — non-shared environment	Unique experiences, illness, error; makes co-twins differ	0.0	0.0	Substantial for most traits

A worked example makes the arithmetic concrete. Suppose a trait shows r_MZ = 0.80 and r_DZ = 0.50. Then h² ≈ 2(0.80 − 0.50) = 0.60, so 60% of the variance is additive-genetic. The shared environment is C = 2(0.50) − 0.80 = 0.20, and the non-shared environment is E = 1 − 0.80 = 0.20. Now contrast a case where r_MZ = 0.50 and r_DZ = 0.25: heritability is again 0.50, C is zero, and E is 0.50 — a trait that is moderately heritable with no shared-family effect at all, which is exactly the profile many adult personality traits display.

Common misconceptions

• "Heritability of 60% means my genes cause 60% of my height." Heritability is about variance across a population, not about a single individual. It says nothing about how much of any one person's trait value is due to genes; it says how much of the differences between people track genetic differences in that population and environment.
• "High heritability means the trait can't change." Height is roughly 80% heritable, yet average height rose by more than 10 centimeters across the twentieth century as nutrition improved. Heritability measures the influence of existing genetic variation given existing environmental variation; change the environment for everyone and the mean can shift even while heritability stays high.
• "Identical twins are genetically identical, so any difference is environmental." Nearly true, but MZ twins accumulate postzygotic somatic mutations and diverging epigenetic marks — DNA methylation drift documented across the lifespan — so even the "100%" is an approximation, and discordant MZ pairs are used precisely to hunt these differences.
• "A shared-environment estimate of zero means parents don't matter." C being near zero means parenting does not make siblings more similar to each other in that trait; it does not mean parenting has no effect. If good parenting raises everyone in a family equally, that effect lives in the mean, not in the twin covariance, and can be invisible to the design.
• "Concordance below 100% in identical twins proves environment dominates." It proves genes are not fully deterministic, which is different. A 50% MZ concordance for schizophrenia against 15% for DZ still implies a strong genetic component; the missing half reflects non-shared environment, stochastic developmental noise, and incomplete penetrance.
• "Twins reared apart were raised in totally different worlds." Separated twins are often placed with relatives or in demographically similar families, and many had contact before study. Good reared-apart studies correct for placement similarity; naive readings of the "Jim twins" coincidences overstate the effect that the aggregate correlations actually support.

The equal-environment assumption and its tests

The entire subtraction rests on one load-bearing premise: that MZ and DZ twins experience equally similar trait-relevant environments. This is the equal-environment assumption (EEA). The obvious objection is that identical twins are dressed alike, share more friends, and are treated more similarly by parents and teachers, so their extra resemblance might be environmental mimicry rather than genetics — which would inflate heritability.

Behavioral geneticists test the EEA rather than assume it. First, the misclassified-twin design: some parents believe fraternal twins are identical or vice versa. If treatment drove resemblance, twins would resemble each other according to perceived zygosity; in fact they resemble each other according to true zygosity confirmed by DNA. Second, researchers show that cosmetic similarity (matching clothes, being confused for each other) predicts trait resemblance far more weakly than actual zygosity. Third, and most powerfully, the twins-reared-apart design removes the shared rearing environment entirely: if MZ twins raised in different homes still correlate highly, no amount of parental "matching" can explain it. Assortative mating — the tendency of people to marry partners like themselves for traits such as intelligence — is a separate complication, because it raises the DZ genetic correlation above 0.5 and can make Falconer's formula underestimate heritability.

A brief history and its landmark studies

• Galton (1875). In "The History of Twins," Francis Galton — Darwin's cousin and the coiner of "nature versus nurture" — first argued that comparing twins who resembled each other closely with those who did not could weigh heredity against upbringing. He lacked the concept of zygosity but grasped the essential experimental logic.
• Fisher (1918). Ronald A. Fisher's paper "The Correlation between Relatives on the Supposition of Mendelian Inheritance" reconciled continuous, biometric traits with Mendelian genetics by partitioning variance into additive, dominance, and environmental components — the mathematical spine of every modern heritability estimate.
• Merriman and Siemens (1920s). Early classical twin studies of intelligence and physical traits established the MZ-versus-DZ comparison as a method, and Hermann Siemens developed the "polysymptomatic similarity" approach for diagnosing zygosity before DNA testing existed.
• Bouchard's Minnesota Study of Twins Reared Apart (1979–). Thomas J. Bouchard Jr. assembled over 100 pairs of twins separated early and reared apart. The landmark 1990 Science report ("Sources of Human Psychological Differences") found reared-apart identical twins about 70% concordant for IQ variation and highly similar in personality — strong evidence for heritability free of the EEA. The reunited "Jim twins" became the study's famous, and frequently over-read, illustration.
• Twinsburg, Sweden, and Danish registries. The annual Twins Days Festival in Twinsburg, Ohio, became a recruiting ground, while the Swedish Twin Registry (tens of thousands of pairs) enabled the 2000 New England Journal of Medicine study by Lichtenstein and colleagues estimating that heritable factors account for a minority of most cancers — a widely cited demonstration of twins answering an epidemiological question at national scale.