Twin Studies vs. Adoption Studies

The investigation into the origins of human individuality—encompassing cognitive capacities, personality configurations, affective dispositions, and vulnerabilities to psychiatric conditions—constitutes one of the most enduring intellectual endeavors in empirical science. For well over a century, the discourse surrounding the relative contributions of biological inheritance and developmental milieu was colloquially categorized as the nature-versus-nurture debate. However, modern behavioral genetics has discarded this simplistic dichotomy in favor of a sophisticated, quantitatively rigorous paradigm centered on etiology, variance decomposition, and complex systemic interaction. Rather than framing heredity and environment as antagonistic forces operating in isolation, contemporary quantitative genetics conceptualizes phenotypic variation as the emergent product of dynamic, polygenic architectures interacting continuously with variegated environmental contexts across the human lifespan.

To unpack this complex architecture, behavioral geneticists rely on natural quasi-experiments that decouple the confounding threads of genetic relatedness and domestic environmental exposure. In standard nuclear families, biological parents provide both the genomic blueprint and the experiential environment in which their offspring develop. This dual transmission creates an intrinsic confound: does a child resemble a parent because of inherited genetic variants, shared domestic routines, modeling of behavioral patterns, socioeconomic opportunities, or an interdependent combination of these vectors? Without specialized research designs that break this confounding cycle, attributing phenotypic resemblance to biological transmission versus environmental socialization remains an intractable methodological problem.

Among the empirical frameworks developed to resolve this fundamental epistemological challenge, two methodologies stand as the foundational pillars of quantitative genetics: twin studies and adoption studies. Twin studies capitalize on the natural biological variation in zygosity, comparing phenotypic concordance rates between monozygotic pairs—who share virtually 100 percent of their genomic sequence—and dizygotic pairs, who share on average 50 percent of their segregating alleles. Adoption studies, conversely, break the biological and social nexus by examining individuals placed away from their biological progenitors at infancy, thus establishing a structural partition between biological lineage and custodial rearing conditions. Together and in contrast, these two methodological paradigms have established the baseline parameters of modern differential psychology, developmental psychopathology, and behavioral medicine, paving the empirical pathway for modern molecular genomics.

1. Introduction to Behavioral Genetic Methodologies

1.1 Defining Twin Studies and Adoption Studies

Twin studies and adoption studies represent the classical biometric designs through which quantitative genetics partitions phenotypic variation across populations into latent genetic and environmental components. At its core, the classical twin design leverages the biological occurrence of two distinct forms of twinning: monozygotic (MZ) twinning, occurring via the cleavage of a single fertilized zygote, and dizygotic (DZ) twinning, resulting from the independent fertilization of two distinct ova by separate spermatozoa. Because monozygotic twins share identical genomic sequences—barring rare post-zygotic somatic mutations—while dizygotic twins share an average of 50 percent of their segregating genetic material identical by descent (IBD), comparing phenotypic resemblance across these two classes allows investigators to infer the magnitude of genetic versus environmental variance. If monozygotic twin pairs systematically demonstrate significantly greater phenotypic concordance or correlation than dizygotic twin pairs across a given trait, and assuming certain core environmental conditions are met, the surplus resemblance is quantitatively attributed to additive genetic influences.

Adoption studies, by contrast, achieve variance partitioning not through variations in genetic relatedness under a uniform household assumption, but via the physical and custodial separation of genetic lineage from domestic rearing. In the classic adoption paradigm, an infant is relinquished by biological progenitors and subsequently reared by non-genetically related adoptive parents. This separation creates a distinct, quasi-experimental bifurcation of transmission vectors. Resemblance between the adopted offspring and their biological parents, with whom they share 50 percent of their segregating genome but zero post-natal domestic environment, serves as an empirical estimate of genetic transmission. Conversely, resemblance between the adopted offspring and their adoptive parents, with whom they share zero segregating alleles above population base rates but 100 percent of the familial rearing environment, serves as a direct indicator of shared environmental transmission.

Both methodologies serve as foundational pillars in differential psychology, psychiatry, and behavioral neuroscience. By evaluating correlations across varying degrees of biological relatedness and household cohabitation, these designs transform vague philosophical inquiries regarding human nature into mathematically verifiable biometric parameters. They offer the necessary empirical architecture to estimate heritability, assess environmental malleability, and establish the structural baselines required to investigate complex developmental trajectories.

1.2 The Epistemological Challenge of Nature versus Nurture

The central epistemological challenge underlying behavioral science is the phenomenon of familial confounding. In unselected, intact nuclear families, parents transmit to their offspring two concurrent developmental streams: a legacy of inherited DNA and a social, material, and educational rearing environment. Consequently, observing a high correlation between parental cognitive ability and offspring academic achievement, or between parental depression and offspring affective dysregulation, yields an intrinsically ambiguous finding. The parent who provides alleles associated with high neurocognitive processing efficiency or psychiatric vulnerability is also the parent structuring the cognitive complexity of the household, navigating socioeconomic strata, modeling emotional coping strategies, and buffering or exacerbating childhood stress. Standard epidemiological family studies, therefore, cannot establish causal attribution; they cannot independently differentiate whether familial transmission is mediated via biological heredity, cultural inheritance, or an intricate feedback loop between both vectors.

Disentangling these intertwined pathways necessitates specialized quasi-experimental designs that isolate one vector while holding the other constant or mathematically parameterizing both simultaneously. Behavioral genetics addresses this problem by conceptualizing familial resemblance through the lens of variance decomposition. Observed phenotypic variance ($V_P$) across a population is formally partitioned into additive genetic variance ($V_A$), non-additive or dominant genetic variance ($V_D$), shared or common environmental variance ($V_C$), and non-shared or unique environmental variance ($V_E$), which typically subsumes stochastic developmental variance and measurement error. Furthermore, complex traits are subject to gene-environment correlations ($r_{GE}$)—wherein an individual’s genetic propensities influence their probability of encountering specific environmental exposures—and gene-environment interactions ($G \times E$), wherein environmental moderating factors modulate the phenotypic penetrance of specific genotypes.

Misinterpretations frequently arise regarding how twin and adoption designs define and partition these variance components. A common conceptual error is reifying heritability as an absolute, deterministic property of an individual’s biology, rather than understanding it as a population-level statistic reflecting the proportion of total phenotypic variance attributable to genetic variance in a specific population at a specific historical juncture. Neither twin studies nor adoption studies measure the direct physiological mechanisms of gene expression. Instead, they quantify the statistical bounds within which genetic and environmental variation manifest phenotypically, providing a critical empirical defense against both biological reductionism and extreme environmental determinism.

1.3 Historical Emergence in Quantitative Genetics

The quantitative conceptualization of human variation can be traced to the late nineteenth century, principally initiated by Francis Galton. In his 1869 treatise Hereditary Genius and his subsequent 1875 paper “The History of Twins,” Galton formally proposed that the study of twins could serve as a natural experiment to disentangle the respective effects of nature and nurture. Although Galton lacked a modern molecular understanding of Mendelian genetics and did not mathematically distinguish between monozygotic and dizygotic embryology with contemporary biological precision, his insight that identical physical appearance and developmental parity could be used to calibrate biological inheritance laid the conceptual groundwork for the emergence of the classical twin design.

The mathematical formalization of these early observations developed during the early twentieth century through the synthesis of Mendelian genetics with Darwinian evolutionary theory and Pearsonian biometrics, achieved most decisively by Ronald A. Fisher in his 1918 landmark paper, “The Correlation between Relatives on the Supposition of Mendelian Inheritance.” Fisher demonstrated that continuous, normally distributed polygenic traits could be understood as the aggregate output of numerous discrete Mendelian loci acting additively and non-additively, alongside environmental variance. During the 1920s and 1930s, researchers such as Curtis Merriman and Edward Thorndike further standardized psychological testing among twin cohorts, paving the way for the algebraic and structural equation formalizations developed by D. W. Fulker, John C. Loehlin, and Kenneth Lange in the mid-to-late twentieth century.

Parallel to the development of the twin paradigm, adoption methodologies emerged in the 1920s as a direct response to the evident shared environmental contamination plaguing standard family studies. Early pioneers, including Barbara Stoddard Burks (1928) and Alice M. Leahy (1935), conducted empirical adoption studies focused primarily on cognitive ability and intellectual quotients (IQ). These investigators recognized that children adopted early in life provided an unprecedented opportunity to observe psychological development absent biological confounding. In the mid-twentieth century, the adoption methodology was further refined by psychiatric geneticists such as Leonard Heston, David Rosenthal, and Seymour Kety, whose landmark Danish Adoption Studies of schizophrenia definitively demonstrated that the risk of developing psychosis remained elevated in the biological offspring of schizophrenic mothers even when reared entirely in non-psychiatric adoptive households. Today, modern developmental psychopathology, behavioral genomics, and macro-epidemiology continue to rely heavily on both twin and adoption frameworks to establish empirical heritability estimates and validate novel molecular findings derived from genome-wide association studies (GWAS).

2. Comprehensive Overview of Twin Studies

2.1 Biological Principles: Monozygotic versus Dizygotic Twins

The classical twin design relies on the natural biological divergence between monozygotic and dizygotic embryogenesis. Monozygotic (MZ) twinning occurs when a single blastocyst, formed by the fertilization of a single ovum by a single spermatozoon, undergoes fission during early post-zygotic development (typically within the first fourteen days post-conception). Consequently, both individuals originate from an identical single-cell zygote and share essentially 100 percent of their genomic DNA sequence, including identical single nucleotide polymorphisms (SNPs), structural variants, and identical initial chromosomal architecture. Dizygotic (DZ) twinning, in contrast, occurs when two separate follicles mature and ovulate simultaneously during a single menstrual cycle and are subsequently fertilized by two distinct spermatozoa. Genetically, dizygotic twins are full siblings who happen to share the same intrauterine gestation and birth date; they share, on average, 50 percent of their segregating genetic material identical by descent (IBD), with a standard genetic correlation coefficient of $r_g = 0.50$.

Given this biological foundation, phenotypic differences observed within monozygotic twin pairs reared together cannot be attributed to inherited structural genomic differences. Instead, any phenotypic divergence observed within an MZ pair must fundamentally reflect non-shared environmental factors ($E$), including idiosyncratic life experiences, peer affiliations, differential treatment by authority figures, accidents, illnesses, systemic developmental noise, and post-zygotic somatic mutations or somatic epigenetic drift. Conversely, phenotypic divergence observed within dizygotic twin pairs arises from both these non-shared environmental factors and the 50 percent segregating genetic variance that differentiates them. Consequently, if a trait is influenced by additive genetic factors, monozygotic twin pairs will consistently display higher phenotypic correlation ($r_{MZ}$) than dizygotic twin pairs ($r_{DZ}$). The surplus correlation observed in MZ twins relative to DZ twins provides the basic mathematical foundation for estimating heritability.

The classical model relies on specific biological assumptions regarding the distribution of genetic variance. Genetic variance itself is typically partitioned into additive genetic variance ($A$), representing the linear, cumulative effects of individual alleles across multiple polygenic loci, and non-additive genetic variance, which comprises genetic dominance ($D$, intra-locus interactions where alleles mask or enhance one another) and epistasis ($I$, inter-locus interactions between distinct genomic regions). Because MZ twins share 100 percent of their genetic configurations, they share all additive, dominance, and epistatic variance components ($r_{g(MZ)} = 1.0$; $r_{d(MZ)} = 1.0$). In contrast, DZ twins share an average of half of the additive genetic variance ($r_{g(DZ)} = 0.50$) and only one-quarter of the dominance genetic variance ($r_{d(DZ)} = 0.25$), with complex epistatic combinations dropping exponentially. This sharp biological contrast allows quantitative geneticists to model both additive and interactive genetic parameters.

2.2 The Classical Twin Design and Structural Equation Modeling

The mathematical engine driving classical twin analysis is the ACE model, which decomposes the total phenotypic variance of a trait ($V_P$) into three latent parameters:

• Additive Genetic Variance ($A$ or $a^2$): The cumulative, linear influence of allelic variants across all segregating loci.
• Shared Environmental Variance ($C$ or $c^2$): The environmental influences that make siblings reared in the same household more similar to one another, independent of genetic relatedness (e.g., family socioeconomic status, parental education, domestic nutrition, local community resources).
• Non-Shared Environmental Variance ($E$ or $e^2$): The environmental influences that make siblings reared together different from one another, encompassing unique personal experiences, differentiated parental treatment, illness, traumatic events, stochastic biological processes during neurodevelopment, and measurement error.

In standard algebraic estimation, known historically as Falconer’s formula, these variance components are calculated directly from the observed intraclass correlation coefficients of monozygotic ($r_{MZ}$) and dizygotic ($r_{DZ}$) pairs reared together:

• $h^2 = A = 2(r_{MZ} – r_{DZ})$
• $C = r_{MZ} – h^2 = 2r_{DZ} – r_{MZ}$
• $E = 1 – r_{MZ}$

While Falconer’s formula provides an intuitive algebraic approximation, modern quantitative genetics utilizes structural equation modeling (SEM) to estimate these parameters via full-information maximum likelihood (FIML) methods. By evaluating covariance and cross-twin cross-trait matrices, structural equation platforms (such as OpenMx) conceptualize the phenotype of each twin as an observed endogenous variable regressed onto latent, unmeasured exogenous variables representing $A$, $C$, and $E$. Because structural equation modeling accommodates missing data, explicitly models measurement error, and permits direct statistical hypothesis testing (such as evaluating whether dropping the $C$ parameter leads to a significant decrement in model fit via a likelihood ratio chi-square test), it has largely superseded raw algebraic formulations. Modern extensions utilize multivariate Cholesky decompositions, direction-of-causation models, and extended twin-family designs that incorporate the parents, spouses, and offspring of twins to isolate assortative mating and non-additive genetic dynamics with heightened precision.

2.3 Core Methodological Assumptions

For the parameters derived from the classical twin design to remain valid, several foundational assumptions must be satisfied:

• The Equal Environments Assumption (EEA): The EEA is the central assumption underlying the twin design. It posits that monozygotic and dizygotic twin pairs are exposed to environmentally comparable rearing conditions with respect to the trait under investigation. While MZ twins may wear identical clothing or share more social circles than DZ twins, the EEA requires that these differences in social treatment do not causally drive their higher phenotypic concordance on the specific trait being measured.
• Absence of Assortative Mating: The classical design assumes that parents of twins mate randomly with respect to the investigated phenotype. If positive assortative mating occurs (i.e., individuals choose partners with similar traits, such as cognitive ability or height), the genetic correlation between dizygotic twins increases beyond the assumed baseline of 0.50. This elevates $r_{DZ}$ relative to $r_{MZ}$, which mathematically deflates estimates of additive genetic variance ($A$) and artificially inflates estimates of shared environmental variance ($C$).
• Absence of Unmodeled Gene-Environment Interplay: The standard linear ACE model assumes that genetic and environmental variance components are additive and orthogonal. It assumes that gene-environment correlation ($r_{GE}$) and gene-environment interaction ($G \times E$) do not significantly account for phenotypic variance unless they are explicitly integrated and parameterized into extended biometric structural equations.
• Generalizability Assumption: The design assumes that twins do not differ systematically from the singleton general population in their biological development, neurocognitive baselines, or psychiatric trajectories. Empirical evaluations have consistently demonstrated that despite experiencing higher rates of pre-term birth and lower average birth weights, twins match singletons across cognitive, behavioral, and somatic distributions by mid-childhood.

2.4 Methodological Constraints and Confounders

Despite its widespread utility, the classical twin design is subject to specific biological and methodological constraints. A major biological confounder relates to variations in prenatal chorionicity and amnionicity. While all dizygotic twins are dichorionic and diamniotic (possessing separate placentas and amniotic sacs), approximately 65 to 70 percent of monozygotic twin pairs are monochorionic, sharing a single placenta. This vascular arrangement can induce severe micro-environmental asymmetries, such as Twin-to-Twin Transfusion Syndrome (TTTS), creating dramatic disparities in fetal blood supply and birth weight. Conversely, monochorionicity can expose MZ twins to an identical intrauterine hormonal and biochemical milieu, creating prenatal environmental parity that does not exist for dichorionic MZ or DZ twin pairs. When unmodeled, chorionicity differences can confound the clean separation of biological inheritance from developmental environments.

Violations of the Equal Environments Assumption represent an ongoing source of empirical debate. If parental expectations, social reinforcement, and environmental feedback loops actively force monozygotic twins into identical psychological trajectories simply because of their identical physical appearance, the classical ACE framework will misattribute this environmental parity to additive genetic variance, artificially inflating heritability ($h^2$) while deflating shared environmental parameters ($C$).

Furthermore, standard classical twin samples frequently suffer from statistical underpowering when attempting to detect shared environmental influences ($C$). Because the statistical power to resolve parameter $C$ depends heavily on the contrast between $r_{MZ}$ and $r_{DZ}$, detecting modest shared environmental effects (accounting for 10 to 15 percent of variance) often requires thousands of twin pairs. Consequently, smaller twin studies frequently drop the $C$ parameter entirely to retain the more parsimonious $AE$ model, potentially overestimating heritability due to statistical limitations rather than biological reality.

3. Comprehensive Overview of Adoption Studies

3.1 Foundational Design: Separating Heredity and Rearing

The adoption design is arguably the most conceptually transparent methodology for separating biological inheritance from environmental socialization. The design operates by studying children who are adopted away at infancy, effectively breaking the correlation between biological lineage and custodial upbringing. In this framework, the developmental pathway is bifurcated into two mutually exclusive relational dyads: the biological parent-offspring dyad and the adoptive parent-offspring dyad.

The biological parents transmit 50 percent of their segregating genome to the offspring identical by descent, but they share no post-natal rearing environment, domestic cohabitation, economic context, or pedagogical scaffolding. Conversely, the adoptive parents provide 100 percent of the post-natal domestic environment, establishing emotional attachments, socioeconomic affordances, and cognitive stimulation, but they share essentially zero segregating genetic variants with the adopted child beyond baseline population frequencies. Therefore, in the absence of confounding variables, any statistically significant resemblance between the adopted child and their biological parents must reflect genetic transmission ($A$). Conversely, any statistically significant resemblance between the adopted child and their adoptive parents must reflect shared environmental transmission ($C$).

This design also extends naturally to adoptive sibling dyads. By examining biologically unrelated children reared together in the same adoptive household (unrelated individuals adopted into the same family), the design establishes a pure test of the shared environment. Because these siblings share zero segregating alleles by descent, any phenotypic correlation observed between them must be attributed entirely to the shared domestic environment ($C$). If their correlation across personality, cognitive, or psychiatric measures approaches zero, it demonstrates that the shared familial environment contributes minimally to individual differences in those traits, independent of genetic relatedness.

3.2 Primary Methodological Paradigms

Over decades of epidemiological application, three distinct methodological paradigms have emerged within adoption research:

• The Adoptees-as-Probands Design: This framework identifies biological parents who manifest a specific disorder or condition (the probands—such as schizophrenia, bipolar disorder, or severe substance dependence) and assesses the prevalence of that condition among their biological offspring who were adopted away at birth into unaffected families. This prevalence is then compared against a control cohort of adopted-away children whose biological parents possessed no such diagnosis. A significantly elevated rate of pathology in the proband-offspring group provides robust evidence for genetic etiology, demonstrating that the transmission of vulnerability occurs despite the complete absence of parental rearing.
• The Adoptees’ Family Method: This retrospective paradigm begins with affected adopted individuals (the proband adoptees) and investigates the prevalence of the condition across two distinct sets of relatives: their biological relatives (parents and siblings) and their adoptive relatives. If the biological relatives exhibit elevated rates of the condition compared to population base rates, while the adoptive relatives exhibit normative baseline rates, the familial clustering of the trait is inferred to be genetic rather than mediated by household transmission.
• The Cross-Fostering Model: This design evaluates outcomes across an integrated two-by-two factorial matrix contrasting high-risk versus low-risk biological backgrounds against high-risk versus low-risk adoptive environments. For example, in studying antisocial personality patterns or criminal behavior, investigators examine four distinct groups:
  • Children with criminal biological parents raised by criminal adoptive parents.
  • Children with criminal biological parents raised by non-criminal adoptive parents.
  • Children with non-criminal biological parents raised by criminal adoptive parents.
  • Children with non-criminal biological parents raised by non-criminal adoptive parents.

  This model is uniquely suited for directly observing gene-environment interaction ($G \times E$), evaluating whether the combination of biological vulnerability and an adverse rearing environment produces an interactive, super-additive escalation of phenotypic risk.

3.3 Core Methodological Assumptions

Similar to twin studies, adoption studies require specific methodological assumptions to guarantee the validity of their biometric estimations:

• The Random Placement Assumption: This assumption posits that adoption agencies do not systematically match biological parents and adoptive parents on specific phenotypic traits. For the design to function cleanly, an infant with a biological background of high cognitive ability or psychiatric risk must have an equal probability of being placed into any adoptive home across the environmental spectrum.
• Environmental Representativeness: The design assumes that the domestic environments provided by adoptive families are representative of the broader population’s environmental variance. If adoptive homes represent an unusually privileged, stable, or homogenous socioeconomic stratum, the observed environmental variance is truncated, potentially biasing estimates of environmental influence.
• Absence of Confounding Prenatal Influences: The design assumes that phenotypic resemblance between the biological mother and the adopted-away child is purely genetic. It assumes that maternal prenatal effects—such as intrauterine nutritional deficits, elevated gestational cortisol, or teratogenic exposures (e.g., alcohol, nicotine, prescribed or illicit pharmaceuticals)—do not systematically alter the child’s developmental trajectory to mimic inherited genetic liability.

3.4 Methodological Constraints and Biases

In practice, adoption studies confront substantial empirical challenges. The most significant of these is selective placement. Adoption agencies have historically engaged in informal or institutional matching, pairing infants with prospective adoptive parents who share similar physical features, religious backgrounds, educational expectations, and socioeconomic standings as the biological parents. When selective placement occurs, it creates a non-zero correlation between the biological parental phenotype and the adoptive home environment. Consequently, the pristine separation of genes and environment is compromised; if unadjusted statistically, selective placement inflates shared environmental estimates and confounds the true magnitude of genetic parameters.

A second major limitation is the restriction of range within adoptive households. Prospective adoptive parents undergo extensive vetting procedures, including rigorous psychological screenings, background checks for criminal histories, financial audits, and home home-environment evaluations. As a direct consequence, adoptive families are systematically more affluent, economically stable, married, and mentally healthy than the general population. Severe parental neglect, acute poverty, and extreme domestic dysfunction are virtually absent from typical adoptive cohorts. By truncating the lower, adverse end of the environmental distribution, adoption studies reduce total observed environmental variance ($V_E$). Statistically, when environmental variance is restricted, the relative proportion of phenotypic variance attributable to genetic factors ($h^2$) is artificially inflated, limiting the ecological validity of the findings when extrapolated to high-adversity populations.

Finally, maternal prenatal effects remain an unmodeled confound in standard adoption models. A biological mother provides not only her nuclear DNA but also the gestational environment spanning nine critical months of neurodevelopment. Maternal stress, poor gestational nutrition, or substance abuse directly influence fetal programming. Because these prenatal variables are provided by the biological mother, postnatal adoption designs that fail to measure and control for gestational parameters risk misclassifying prenatal biological insults as additive genetic variance.

4. Key Differences Between Twin and Adoption Studies

4.1 Mechanism of Environmental and Genetic Partitioning

Twin and adoption designs resolve the nature-nurture confound through fundamentally distinct structural approaches. The classical twin design manipulates the genetic relatedness coefficient while attempting to hold the domestic rearing environment constant. By contrasting individuals who share 100 percent of their genes against those who share 50 percent, all while reared within the same contemporary domestic context, twin models observe the effect of altering genetic similarity against a common background of shared familial socialization. In contrast, adoption studies manipulate the environmental transmission vector while holding the genetic inheritance coefficient fixed. By setting environmental sharing to zero between biological parents and children, or setting genetic relatedness to zero between adoptive parents and children, the adoption design completely separates the custodial household from biological lineage.

This structural difference dictates the statistical frameworks utilized by each design. Twin studies rely fundamentally on intraclass correlation differentials ($r_{MZ}$ versus $r_{DZ}$) between pairs belonging to the same generational cohort. Adoption designs, conversely, operate largely through intergenerational regression coefficients and cross-generational phenotypic associations (e.g., regressing the adoptee’s adult outcome onto the biological mother’s documented phenotype, alongside the adoptive mother’s phenotype). Consequently, while twin studies are ideally calibrated to model cohort-specific and age-specific genetic architectures, adoption studies excel at observing the intergenerational transmission of behaviors, coping mechanisms, and socioeconomic mobility.

4.2 Vulnerability to Specific Confounding Factors

Each design exhibits distinct vulnerabilities to methodological confounding. The classical twin design is structurally sensitive to violations of the Equal Environments Assumption. If monozygotic twins are treated more similarly by their social environment than dizygotic twins, and if this differential treatment directly impacts the phenotype under investigation, the twin design misallocates this environmental variance to the additive genetic parameter ($A$). Additionally, the twin paradigm is vulnerable to prenatal chorionicity disparities, where monochorionic sharing of a single placenta can introduce asymmetric hemodynamics that do not occur in dizygotic pairs, potentially introducing unique prenatal noise or unmeasured biological parity.

Adoption studies, by comparison, are immune to the Equal Environments Assumption, as the unrelated individuals reared together do not share physical identity or social perceptions of twinship. However, adoption designs are uniquely vulnerable to selective placement biases and severe range restriction. The vetting of adoptive parents can truncate environmental variance, leading to an underestimation of shared environmental impacts that occur in extreme environments. Furthermore, adoption studies are uniquely vulnerable to prenatal maternal confounding; intrauterine insults provided by the biological mother are easily conflated with her genetic contribution, a confound that twin studies partially mitigate by comparing MZ and DZ twins who shared the same gestational mother and intrauterine period.

4.3 Sampling Feasibility, Attrition, and Statistical Power

From an epidemiological standpoint, twin studies provide substantial logistical advantages over adoption designs. Twins account for approximately 1.5 to 3.5 percent of live births across global populations, and numerous nations maintain expansive, population-based national twin registries (such as the Swedish Twin Registry or the Netherlands Twin Register) containing tens of thousands of deeply phenotyped, genetically characterized pairs. Because both twins typically mature within the same household at the same developmental stage, longitudinal follow-ups are highly efficient, resulting in superior statistical power for biometric structural equation modeling.

Adoption cohorts, by contrast, are exceedingly difficult to assemble and maintain in contemporary research. In most developed nations, domestic infant adoptions have declined precipitously over recent decades due to shifting cultural norms, reproductive technologies, and changes in social welfare policies. International adoptions introduce severe unmeasured ethnic, nutritional, and early institutionalization confounds. Furthermore, adoption records are protected by stringent legal privacy frameworks, frequently sealing biological parent identities and medical records. Obtaining phenotypic or biological data from biological parents is logistically complex, and attrition rates across longitudinal adoption designs are substantial, routinely resulting in smaller sample sizes that struggle to achieve the statistical power found in twin studies.

4.4 Measurement Sensitivity for Environmental Variance Components

The two methodologies possess contrasting sensitivities when estimating environmental variance components. In the classical twin design, the shared environmental component ($C$) is an indirect mathematical deduction derived from the residual gap between $r_{MZ}$ and twice the difference between $r_{MZ}$ and $r_{DZ}$ ($C = 2r_{DZ} – r_{MZ}$). Because $C$ is estimated as a residual discrepancy, it is susceptible to low statistical power and measurement error. Modest shared environmental influences are frequently lost to statistical insignificance, resulting in many classical twin analyses prematurely dropping $C$ in favor of parsimonious $AE$ models, which risk overestimating heritability.

Adoption studies provide a far more direct, robust empirical window into shared environmental effects ($C$). When unrelated adoptive siblings are reared in the same domestic household, their phenotypic correlation provides an unconfounded, empirical test of shared family environments. If their correlation on a trait is $0.25$, that figure directly estimates the proportion of variance accounted for by the shared home environment, without relying on differential genetic subtractions. Similarly, non-shared environmental variance ($E$) is estimated in twin studies purely as the residual variance unexplained by MZ twin resemblance ($1 – r_{MZ}$), inherently conflating unique experiential environments with random developmental noise and measurement unreliability. In adoption studies, non-shared environmental effects can be measured directly through targeted assessments of differentiated parental practices, discordance in sibling experiences, and distinct peer trajectories.

Dimension / Parameter	Twin Studies	Adoption Studies
Genetic Manipulation	Manipulates genetic relatedness ($r_g = 1.0$ for MZ vs. $0.50$ for DZ) while holding the shared domestic environment relatively constant.	Holds genetic relatedness fixed ($r_g = 0.50$ with biological parent; $r_g = 0$ with adoptive parent) while splitting the domestic rearing environment.
Environmental Manipulation	Assumes equal post-natal environments across zygosity groups within the same familial home.	Directly separates post-natal environmental socialization from inherited genetic transmission.
Core Vulnerability	Violations of the Equal Environments Assumption (EEA) and variations in prenatal chorionicity/placentation.	Selective placement by adoption agencies and restriction of environmental range among vetted adoptive families.
Estimation of Shared Environment ($C$)	Indirectly inferred via mathematical subtraction ($C = 2r_{DZ} – r_{MZ}$); often suffers from low statistical power.	Directly measured via phenotypic correlations between genetically unrelated adoptive siblings or adoptive parent-child pairs.
Prenatal Confounding	Prenatal maternal environment is largely shared across both twins within a pair, though monochorionicity can introduce asymmetry.	Maternal gestational effects (teratogens, nutrition, stress) are confounded with biological mother genetic contributions.
Sample Feasibility & Cohort Size	High; population-based registries provide large, deeply phenotyped cohorts containing tens of thousands of twin pairs.	Low; adoptions are statistically rare, legally protected, and logistically difficult to track across biological and adoptive lines.
Primary Analytical Framework	Structural Equation Modeling (SEM) utilizing cross-twin variance-covariance matrices (ACE/ADE formulations).	Intergenerational parent-offspring regressions, cross-fostering factorial ANOVAs, and unrelated sibling correlations.
Age Concordance	Perfect; twins are precisely the same age and progress through developmental milestones concurrently.	Variable; adoptive siblings or biological parents/offspring are evaluated across different developmental stages or historical eras.

5. Similarities and Methodological Overlap

5.1 Shared Theoretical Foundations

Despite their distinct analytical frameworks, twin and adoption studies share an identical theoretical infrastructure: the classical quantitative genetics paradigm formalized by Ronald Fisher, Sewall Wright, and J. B. S. Haldane. Both methodologies conceptualize complex human traits—such as general cognitive ability, neuroticism, blood pressure, or liability to major depressive disorder—as polygenic phenotypes governed by the aggregate action of hundreds or thousands of discrete genetic loci acting largely additively, alongside diverse environmental pressures.

Both frameworks partition observed phenotypic variance ($V_P$) into underlying latent variables of genetic ($V_G$) and environmental ($V_E$) origins. Both models aim to compute heritability—whether in the narrow sense ($h^2 = V_A / V_P$), reflecting only additive genetic variance, or in the broad sense ($H^2 = V_G / V_P$), which subsumes dominance and epistatic interactions. Furthermore, neither design relies on identifying individual functional nucleotides or specific biological pathways. Instead, both operate as macro-level biometric filters, establishing the empirical bounds of genetic liability and environmental malleability within human populations, providing the statistical framework that guides downstream molecular inquiries.

5.2 The Unified Frontier: Twins Reared Apart

The most elegant synthesis of quantitative genetic methodology occurs when the twin and adoption designs converge into a single empirical framework: the study of monozygotic and dizygotic twins separated at birth and reared in distinct adoptive homes. Conceptualized most prominently in the landmark Minnesota Twin Family Study of Twins Reared Apart (MISTRA), directed by Thomas Bouchard, Jr., David Lykken, Matthew McGue, and Auke Tellegen, this paradigm represents a natural quasi-experiment of profound significance.

In monozygotic twins reared apart (MZA), nature provides identical genomic configurations ($r_g = 1.0$), while adoption provides independent custodial environments, separate geographic regions, and distinct parental socialization. Because the domestic rearing environments are separated, the traditional Equal Environments Assumption is rendered irrelevant; there are no shared domestic routines or parental social expectations to artificially inflate phenotypic resemblance. Consequently, the intraclass correlation between monozygotic twins reared apart ($r_{MZA}$) serves as a direct, unconfounded empirical estimate of broad-sense heritability ($H^2$):

$H^2 \approx r_{MZA}$

The empirical findings generated by the MISTRA and Swedish reared-apart cohorts were transformative. Across hundreds of physiological, cognitive, and personality dimensions, monozygotic twins reared apart demonstrated phenotypic correlations that closely mirrored those observed in monozygotic twins reared together ($r_{MZT}$). For instance, the MZA correlation for adult general cognitive ability was estimated at approximately $0.70$, indicating that around 70 percent of the variance in adult IQ was attributable to genetic variation across the evaluated cohorts. This near-identical concordance across reared-together and reared-apart pairs provided decisive empirical evidence that the high phenotypic correlations observed in classical twin studies were driven primarily by shared biological inheritance rather than by shared domestic socialization or parental treatment.

5.3 Susceptibility to Gene-Environment Interplay

Both classical twin designs and standard adoption paradigms face challenges when accounting for complex gene-environment interplay. In standard linear implementations of both models, gene-environment correlation ($r_{GE}$) and gene-environment interaction ($G \times E$) are assumed to be absent. However, human behavioral development is characterized by extensive gene-environment interplay, which operates through three primary mechanisms:

• Passive Gene-Environment Correlation: Occurs when biological parents transmit both alleles and home environments that align with those genetic propensities (e.g., highly literate parents passing on both reading-supportive genes and a home filled with literature).
• Evocative Gene-Environment Correlation: Occurs when an individual’s genetically influenced traits elicit specific, reinforcing reactions from their social environment (e.g., a child with an easy-going disposition eliciting warm, responsive parenting).
• Active Gene-Environment Correlation: Occurs when an individual actively selects, constructs, and modifies their own ecological niche based on their genetic predispositions (e.g., a sensation-seeking adolescent affiliating with risk-tolerant peer groups).

When gene-environment correlation goes unmeasured, standard biometric models misallocate its variance. In classical twin studies, evocative and active $r_{GE}$ are systematically misclassified as additive genetic variance ($A$), because the child’s genotype is driving the environmental exposure. In standard adoption studies, passive $r_{GE}$ is broken post-natally because the adoptive parents do not share the child’s genome, making adoption studies particularly effective for isolating passive correlation. However, evocative and active $r_{GE}$ persist in adoptive homes, as the child still elicits specific parenting responses and selects their own environments. Furthermore, if a latent gene-environment interaction ($G \times E$) exists—such as genetic risk only manifesting when triggered by an adverse childhood environment—the standard classical twin ACE model misallocates this interactive variance into the additive genetic component ($A$) or non-shared environmental component ($E$), depending on whether the environmental trigger is shared or unshared. Modern biometric modeling circumvents these limitations by incorporating explicit environmental indices into continuous moderator structural equations, allowing researchers to model both $r_{GE}$ and $G \times E$ directly.

6. When to Use Each Design in Research Practice

6.1 Optimal Contexts for Twin Study Deployment

The classical twin design is optimal when investigators require expansive, deeply phenotyped, population-representative samples with substantial statistical power to decompose complex traits across the lifespan. Twin registries are well-suited for high-throughput longitudinal phenotyping across neuroimaging, physiological, cognitive, and psychiatric domains. The matched-age property of twins eliminates age-cohort confounds, permitting developmental geneticists to map how heritability parameters shift from early childhood into late adulthood.

Twin studies are particularly powerful when utilizing the discordant monozygotic twin design to examine non-shared environmental mechanisms ($E$). Because monozygotic twins share 100 percent of their genetic sequence, any pair discordant for a disease phenotype (such as one twin developing schizophrenia, type 1 diabetes, or major depression while the co-twin remains healthy) provides an ideal natural control. By holding the genomic background, gestational timing, family socioeconomic status, and parental characteristics completely constant, researchers can leverage discordant MZ pairs to isolate the specific, non-shared environmental, epigenetic, or stochastic developmental triggers that precipitate clinical disease.

Additionally, modern neuroimaging and multi-omics investigations benefit greatly from the classical twin design. Conducting functional MRI, resting-state connectivity analyses, or epigenome-wide association studies across MZ and DZ twin pairs provides a robust framework to establish whether regional brain volumes, neural activation patterns, or DNA methylation signatures are driven primarily by shared biological inheritance, domestic environments, or unique lifestyle factors.

6.2 Optimal Contexts for Adoption Study Deployment

The adoption methodology is the gold standard when researchers need to assess direct parental transmission pathways and break the passive gene-environment correlations inherent in biological families. When evaluating whether parental psychopathology, divorce, physical punishment, or specific parenting styles causally drive adverse child development, adoption designs provide an unconfounded test. If an association between harsh parental discipline and child externalizing behavior exists in biologically related families, it could simply reflect shared genes for emotional dysregulation. However, if the same association is replicated within adoptive parent-child dyads, it provides compelling evidence for a direct, environmentally mediated causal effect.

Adoption studies are also essential for assessing the efficacy of protective environmental interventions that disrupt intergenerational cycles of disadvantage. By examining the degree to which an enriched, socioeconomically stable adoptive home can buffer against inherited psychiatric vulnerability or elevate cognitive baselines, adoption paradigms offer natural tests of developmental plasticity. For example, educational adoption studies have demonstrated that children adopted from socioeconomically disadvantaged backgrounds into affluent families exhibit significant mean-level increases in IQ and academic attainment (reflecting a substantial environmental intervention effect), even while individual rank-order differences within the group remain heavily influenced by biological parental characteristics (reflecting genetic heritability).

Finally, adoption designs are optimal when researchers wish to quantify shared environmental parameters ($C$) directly. By avoiding the statistical subtraction models of the classical twin design, adoption cohorts provide direct, empirical estimates of how non-biological household sharing shapes individual differences across development.

6.3 Triangulation with Modern Molecular Genomics

The contemporary behavioral genetic landscape has transitioned beyond reliance on quantitative pedigree designs alone. Instead, modern research emphasizes methodological triangulation: cross-validating findings across twin studies, adoption cohorts, and molecular genomic platforms to ensure robust causal inferences.

A prominent application of this triangulation involves the integration of Polygenic Risk Scores (PRS) into twin and adoption frameworks. Genome-wide association studies allow investigators to calculate an individual’s molecular genetic liability for a given trait by summing the risk alleles across their genome, weighted by their effect sizes. However, standard molecular studies in general populations are themselves susceptible to indirect parental effects and population stratification. By calculating polygenic scores within adoption cohorts, researchers can measure what has been termed “genetic nurture.” If the non-transmitted polygenic score of an adoptive parent correlates with the adopted child’s educational outcomes, it demonstrates an environmental effect of parental genetic traits operating via the rearing environment, completely independent of biological inheritance.

Furthermore, triangulation is essential for reconciling the historical “missing heritability” paradox. When Genome-Wide Complex Trait Analysis (GCTA) and Linkage Disequilibrium (LD) Score Regression were first applied to common SNPs, the resulting molecular heritability estimates ($h^2_{SNP}$) were consistently lower than the heritability estimates derived from classical twin studies ($h^2_{twin}$). By applying both molecular methods and pedigree models within adoption samples and extended twin families, behavioral geneticists demonstrated that classical twin studies capture both rare genetic variants and non-additive genetic effects that standard common-variant SNP arrays fail to detect, while also revealing that unmodeled assortative mating can bias these estimations.

Finally, integrating Mendelian randomization (MR) within twin and adoption cohorts provides a powerful methodology to evaluate causal directions in epidemiology. By using genetic variants as natural instrumental variables within families where rearing environments are controlled, researchers can establish whether observed observational risk factors (such as dietary patterns, sleep duration, or physical activity) causally drive somatic and psychiatric health outcomes, free from the confounding of reverse causality and environmental bias.

7. Frequently Asked Questions

7.1 Why are twins reared apart considered the gold standard in behavioral genetics?

Twins reared apart, particularly monozygotic twins reared apart (MZA), are considered the gold standard because they eliminate the need to mathematically model or statistically adjust for shared domestic rearing environments while preserving complete genetic identity. In the classical twin design, monozygotic and dizygotic twins are raised together in the same household, requiring the Equal Environments Assumption to infer that greater MZ resemblance is genetic rather than social. When identical twins are separated early in life and reared by different adoptive parents in distinct households, their shared post-natal environment is reduced to near zero.

Under these conditions, any phenotypic correlation observed between the separated identical twins serves as a direct, empirical measure of broad-sense heritability ($H^2 \approx r_{MZA}$). The resemblance cannot be driven by shared household routines, shared parental discipline, or shared community resources. However, while methodologically powerful, reared-apart twin samples are exceedingly rare, historically limited to small cohorts assembled over decades, such as the Minnesota and Swedish cohorts. Modern child welfare policies and adoption practices place siblings together whenever possible, making the replication of large-scale reared-apart twin studies unfeasible today.

7.2 How does the Equal Environments Assumption affect the validity of twin studies?

The Equal Environments Assumption (EEA) posits that monozygotic and dizygotic twin pairs experience environmentally comparable rearing conditions with respect to the trait under investigation. The primary critique raised against this assumption is that monozygotic twins look alike, have identical biological sexes, and frequently elicit more similar social treatment from parents, teachers, and peers than dizygotic twins. If this environmental similarity causally enhances their phenotypic resemblance, the classical twin design will misallocate this environmental variance to the additive genetic parameter ($A$), leading to an overestimation of heritability and an underestimation of shared environmental effects.

To evaluate this potential bias, behavioral geneticists have subjected the EEA to extensive empirical testing. Researchers have examined “misdiagnosed twins”—cases where parents, teachers, and the twins themselves erroneously believed they were dizygotic due to physical differences, but DNA testing proved they were monozygotic, and vice versa. These studies consistently demonstrate that phenotypic resemblance tracks true biological zygosity rather than perceived zygosity. Even when parents treat their monozygotic twins differently or dress them dissimilarly, their cognitive, psychological, and physiological resemblance remains remarkably robust. While minor violations can occur for specific social phenotypes, empirical evaluations show that the Equal Environments Assumption largely holds across most cognitive, personality, and psychiatric traits.

7.3 What is selective placement, and how does it bias adoption studies?

Selective placement occurs when adoption agencies systematically match biological parents with adoptive parents based on specific social, educational, physical, or religious characteristics. For example, an agency might deliberately place an infant from a biological family with high educational attainment into an adoptive family with high socioeconomic standing, or place an infant from a specific ethnic or rural background into an adoptive home sharing those precise characteristics.

This practice introduces a non-zero correlation between the biological parental phenotype and the adoptive rearing environment, violating the foundational assumption of random placement. When selective placement occurs, the genetic liability transmitted by the biological parents becomes correlated with the environmental affordances provided by the adoptive parents. If researchers do not measure and statistically control for this correlation, the adoption design will misattribute the resulting child resemblance to shared environmental effects ($C$) or overestimate genetic transmission ($A$), confounding the clean separation of nature and nurture. Modern adoption studies address this issue by routinely measuring biological parental characteristics and incorporating cross-family correlation paths directly into their structural equation models.

7.4 Can twin and adoption studies detect epigenetic modifications?

Classical biometric twin and adoption designs cannot directly differentiate stable epigenetic marks (such as DNA methylation or histone modifications) from other underlying biological or environmental components. In a standard ACE structural equation model, any biological process that operates additively and tracks genomic relatedness is partitioned into the $A$ parameter, while any dynamic, environmentally induced epigenetic modification that differs between siblings is absorbed into the non-shared environmental parameter ($E$).

However, modern adaptations of these designs have become foundational platforms for molecular epigenetics. The discordant monozygotic twin design is widely recognized as one of the most powerful templates for epigenome-wide association studies (EWAS). Because monozygotic twins share an identical DNA sequence, any divergence in disease outcome cannot be attributed to structural genetic variation. By comparing DNA methylation profiles across discordant MZ pairs, investigators can identify specific epigenetic alterations associated with the onset of conditions like systemic lupus erythematosus, schizophrenia, or major depression, unconfounded by genetic variation.

Similarly, adoption studies provide a unique opportunity to distinguish between inherited epigenetic patterns and post-natal environmentally induced epigenetic changes. By analyzing whether an adopted child’s epigenetic profile correlates with the biological mother’s historical exposures or with the adoptive home’s current environmental stressors, researchers can directly map how post-natal environments reprogram the human epigenome independent of direct genetic transmission.

8. Conclusion: Integrating Quantitative Genetics Methodologies

8.1 Synthesis of Methodological Strengths and Trade-offs

Twin and adoption studies represent two complementary methodologies designed to resolve the fundamental problem of familial confounding in human behavioral science. Each design possesses unique empirical strengths and distinct methodological limitations. The classical twin design provides substantial statistical power, facilitated by large-scale, population-representative registries containing tens of thousands of twin pairs. It excels in establishing precise biometric parameters, evaluating age-dependent heritability trajectories, and providing an ideal genetically identical control group through discordant monozygotic twin analyses to isolate non-shared environmental factors ($E$). However, it remains dependent on the validity of the Equal Environments Assumption and can struggle with low statistical power when attempting to detect modest shared environmental parameters ($C$).

The adoption design provides an intuitive and direct structural separation of post-natal environmental rearing from inherited biological factors. It serves as an exceptional framework for testing direct parental transmission pathways, evaluating the impact of protective environmental interventions, and directly observing shared environmental influences ($C$) free from twin-specific developmental assumptions. Nonetheless, adoption studies face substantial logistical hurdles, including small, hard-to-access cohorts, legal and privacy restrictions, vulnerability to selective placement biases, restriction of range among vetted adoptive homes, and the confounding influence of unmeasured maternal prenatal factors.

Neither design is universally superior; rather, they serve as mutual methodological correctives. The vulnerabilities of the twin study (such as the Equal Environments Assumption) are absent in the adoption study, while the vulnerabilities of the adoption study (such as selective placement and restricted environmental variance) are largely mitigated in population-based twin registries. When both designs converge on a similar heritability estimate or environmental parameter for a specific human trait, behavioral geneticists can be confident that the observed outcome reflects true biological and developmental reality rather than design-specific methodological artifacts.

8.2 The Future of Quantitative and Molecular Triangulation

As behavioral genetics advances in the post-genomic era, the historical divide between quantitative pedigree modeling and molecular genomics has dissolved. The future of the discipline lies in systematic triangulation: integrating classical twin and adoption cohorts with whole-genome sequencing, polygenic risk scoring, and macro-epidemiological biobanks such as the UK Biobank and All of Us. Pedigree designs are no longer used merely to calculate abstract $A$, $C$, and $E$ variance components; they are increasingly utilized as rigorous frameworks to control for confounding within molecular studies.

By embedding polygenic risk scores within twin and adoption cohorts, researchers can isolate phenomena such as genetic nurture, assortative mating, and indirect genetic effects with unprecedented precision. Furthermore, discordant monozygotic twin designs provide a vital template for functional genomics, metabolomics, and epigenomics, enabling researchers to link environmental exposures directly to cellular mechanisms without the confounding influence of genomic sequence variation. Rather than rendering classical quantitative designs obsolete, the molecular revolution has renewed their relevance, demonstrating that understanding the complex, polygenic architecture of human nature requires both the granular resolution of the nucleotide and the macro-level perspective provided by family, twin, and adoption quasi-experiments.

References

Bouchard, T. J., Jr., Lykken, D. T., McGue, M., Segal, N. L., & Tellegen, A. (1990). Sources of human psychological differences: The Minnesota Study of Twins Reared Apart. Science, 250(4978), 223–228. https://doi.org/10.1126/science.2218526

Fisher, R. A. (1918). The correlation between relatives on the supposition of Mendelian inheritance. Transactions of the Royal Society of Edinburgh, 52(2), 399–433. https://doi.org/10.1017/S008045680001216X

Galton, F. (1875). The history of twins, as a criterion of the relative powers of nature and nurture. Fraser’s Magazine, 12, 566–576.

Heston, L. L. (1966). Psychiatric disorders in foster home reared children of schizophrenic mothers. The British Journal of Psychiatry, 112(489), 819–825. https://doi.org/10.1192/bjp.112.489.819

Kendler, K. S., Neale, M. C., Kessler, R. C., Heath, A. C., & Eaves, L. J. (1993). A test of the equal-environment assumption in twin studies of psychiatric illness. Behavior Genetics, 23(1), 21–27. https://doi.org/10.1007/BF01067551

Kong, A., Thorleifsson, G., Frigge, M. L., Vilhjalmsson, B. J., Young, A. I., Thorgeirsson, T. E., Benonisdottir, S., Oddsson, A., Halldorsson, B. V., Masson, G., Gudbjartsson, D. F., Helgason, A., Bjornsdottir, G., Thorsteinsdottir, U., & Stefansson, K. (2018). The nature of nurture: Effects of parental genotypes. Science, 359(6374), 424–428. https://doi.org/10.1126/science.aan6877

Neale, M. C., & Cardon, L. R. (1992). Methodology for Genetic Studies of Twins and Families. Springer Netherlands. https://doi.org/10.1007/978-94-011-2646-5

Polderman, T. J., Benyamin, B., de Leeuw, C. A., Sullivan, P. F., van Bochoven, A., Visscher, P. M., & Posthuma, D. (2015). Meta-analysis of the heritability of human traits based on fifty years of twin studies. Nature Genetics, 47(7), 702–709. https://doi.org/10.1038/ng.3285

Plomin, R., DeFries, J. C., Knopik, V. S., & Neiderhiser, J. M. (2016). Behavioral Genetics (7th ed.). Worth Publishers.

Stoolmiller, M. (1999). Implications of the restricted range of family environments for estimates of heritability and environmentality in twin and adoption studies. Psychological Bulletin, 125(4), 392–409. https://doi.org/10.1037/0033-2909.125.4.392