Genotype-Environment Correlation In Genetic Architecture

Why Genotype and Environment May Be Correlated

A genetic association with an outcome does not, by itself, show that genetic differences affect the outcome only through a direct biological pathway. People with different genetic propensities may also experience different environments. Some of that pattern can arise because parents transmit both alleles and environments; some can arise when a person’s characteristics elicit responses from others or influence which environments the person enters. These linked pathways are called genotype–environment correlation.

The term describes statistical dependence between genetic variation and environmental exposure. It does not mean that every person with a particular genotype experiences the same environment, nor does it establish which pathway caused the dependence. The distinction matters when interpreting genetic associations: a marginal association can reflect a genetic effect, an environmental pathway correlated with genotype, or both.

Definition: Genotype–environment correlation, often abbreviated \(r_{GE}\), is a statistical association between genetic variation and environmental exposure. It can be described for a measured genotype or genetic score \(G\) and an environmental measure \(E\), for example by their covariance \(\operatorname{Cov}(G,E)\) or correlation. The three standard forms are passive, evocative (also called reactive), and active genotype–environment correlation.

For a sample of \(n\) individuals with measured scores \(G_i\) and environmental values \(E_i\), the sample Pearson correlation is

r_{GE} = \frac{\sum_{i=1}^{n}(G_i-\bar{G})(E_i-\bar{E})} {\sqrt{\sum_{i=1}^{n}(G_i-\bar{G})^2\sum_{i=1}^{n}(E_i-\bar{E})^2}}.

It ranges from \(-1\) to \(1\): positive values indicate that larger genetic scores tend to occur with larger environmental values, while negative values indicate that larger genetic scores tend to occur with smaller environmental values. A correlation summarizes a pattern; it does not identify its cause. In particular, the correlation alone cannot distinguish the three forms below.

Three Pathways to Genotype–Environment Correlation

The labels describe different routes by which genetic variation and environmental exposure become associated. They are not mutually exclusive: more than one route can operate in the same family or population.

Form	How the association arises	Illustrative pathway
Passive	Parents provide both inherited alleles and a rearing environment correlated with their own genetic characteristics.	Parents’ reading-related propensities influence both the alleles their child inherits and the reading materials available at home.
Evocative	A genetically influenced characteristic of the individual elicits a response from other people.	A child’s attention or activity level is associated with how a teacher responds to that child.
Active	An individual’s genetically influenced characteristics contribute to selecting or shaping environments.	A person with a propensity for a particular activity is more likely to seek settings where that activity is practiced.

In passive correlation, the child does not need to seek out or elicit the relevant environment. The association can be present because parents’ characteristics are related to both the alleles they transmit and the home they create. In evocative correlation, the individual’s characteristics help elicit environmental responses. In active correlation, the individual helps select or create the environment. These descriptions concern pathways, not judgments about responsibility or conscious choice.

The environmental measure may be a feature of a family, school, peer group, or activity setting. It need not be a deliberate intervention or a single exposure. Also, genetic propensity is usually probabilistic: a score may be associated with a greater likelihood of a characteristic or environment without determining either one.

Worked Example: Recognizing Passive Correlation

Hypothetical scenario. Four families are described by a parental genetic propensity score \(P\), the child’s inherited score \(G\), and a home reading-resource score \(E\). The pairs \((P,G,E)\) are \((0,0,2)\), \((1,1,4)\), \((1,1,5)\), and \((2,2,7)\). Higher \(E\) means more reading resources at home.

Across these families, the child’s inherited score and home-resource score tend to rise together. The proposed passive pathway is that parental characteristics associated with \(P\) are related both to the alleles passed to the child and to the reading environment the parents provide. The child need not have elicited the resources or selected them.

This small hypothetical pattern illustrates a possible mechanism; it does not establish that the parental score caused the resources, or that the child’s score directly caused reading outcomes. Other family characteristics could also contribute to the observed pattern. The classification is passive because the pathway being described runs through parental transmission and rearing, rather than through the child’s response or environmental choice.

Worked Example: Recognizing Evocative Correlation

Hypothetical scenario. In a school, children with a higher genetically influenced activity propensity are, on average, more physically active during class. Teachers respond by assigning those children different classroom tasks. A researcher records a genetic score \(G\), observed activity \(A\), and a teacher-response measure \(E\), with larger \(E\) indicating more movement-oriented tasks.

The proposed sequence is \(G\)-related propensity, then observed activity, then a teacher response. If more active children tend to receive more movement-oriented tasks, \(G\) and \(E\) may be correlated even if the children did not choose those tasks. This is evocative correlation because the child’s characteristics are associated with a response from another person.

The observed association would not show that the genetic score directly caused the teacher’s decision. The teacher may respond to visible behavior, and other factors, such as classroom needs, may influence both behavior and task assignment. A careful interpretation identifies the proposed pathway without treating a correlation as proof of it.

Worked Example: Recognizing Active Correlation

Hypothetical scenario. A study follows adults who differ in a genetic propensity associated with interest in endurance activities. People with higher scores are more likely to join running clubs and to spend time training with other runners. The researchers measure the score \(G\) and weekly participation in running environments \(E\).

If people with higher scores tend to enter these settings more often, \(G\) and \(E\) will tend to be positively associated. The proposed route is that individual characteristics contribute to seeking or shaping an environment, so this is active correlation. The environment could then affect later behavior or outcomes; the score–environment association alone does not determine whether that happens.

The label does not imply that the choice is entirely conscious or unconstrained. Access, family circumstances, health, and social opportunities may also influence participation. The classification concerns the proposed contribution of the person’s characteristics to the environment they experience.

How Correlation Can Change an Observed Genetic Association

A simple linear model makes one consequence precise. Suppose an outcome \(Y\) is described by a genetic score \(G\), an environmental measure \(E\), and a residual \(\varepsilon\):

Y = \beta_G G + \beta_E E + \varepsilon.

For this derivation, assume the model is linear in \(G\) and \(E\), the same coefficients apply across the observations, and \(\operatorname{Cov}(G,\varepsilon)=0\). The coefficient \(\beta_G\) is the genetic-score coefficient in this specified model holding \(E\) fixed; \(\beta_E\) is the environmental coefficient holding \(G\) fixed. These interpretations require the model to be appropriate, and do not follow from correlation alone.

If \(Y\) is instead associated with \(G\) without including \(E\), the population marginal slope is \(\operatorname{Cov}(G,Y)/\operatorname{Var}(G)\), provided \(\operatorname{Var}(G)>0\). Apply covariance to both sides of the model:

\begin{aligned} \operatorname{Cov}(G,Y) &= \operatorname{Cov}(G,\beta_G G+\beta_E E+\varepsilon) \\ &= \beta_G\operatorname{Var}(G) +\beta_E\operatorname{Cov}(G,E) +\operatorname{Cov}(G,\varepsilon) \\ &= \beta_G\operatorname{Var}(G) +\beta_E\operatorname{Cov}(G,E). \end{aligned}

Dividing by \(\operatorname{Var}(G)\) gives the marginal slope:

b_{\text{marginal}} = \beta_G+ \beta_E\frac{\operatorname{Cov}(G,E)}{\operatorname{Var}(G)}.

Thus, under these assumptions, the marginal association includes the model’s genetic-score coefficient plus a term arising when the environment predicts \(Y\) and is correlated with \(G\). If either \(\beta_E=0\) or \(\operatorname{Cov}(G,E)=0\), this particular additional term is zero. The equation does not say that every observed association can be decomposed this way in practice: omitted variables, measurement error, nonlinearity, or a different causal structure can change the interpretation.

Worked Example: A Marginal Slope With a Correlated Environment

Hypothetical model. Let \(G\) and \(E\) be standardized scores, so that \(\operatorname{Var}(G)=1\). Suppose \(\operatorname{Cov}(G,E)=0.40\), \(\beta_G=0.30\) outcome standard deviations per standard deviation of \(G\), and \(\beta_E=0.50\) outcome standard deviations per standard deviation of \(E\). Assume \(\operatorname{Cov}(G,\varepsilon)=0\) and the stated linear model is appropriate.

The marginal slope for \(G\), without including \(E\), is

\begin{aligned} b_{\text{marginal}} &= \beta_G+\beta_E \frac{\operatorname{Cov}(G,E)}{\operatorname{Var}(G)} \\ &= 0.30+0.50\left(\frac{0.40}{1}\right) \\ &= 0.30+0.20 \\ &= 0.50. \end{aligned}

In this hypothetical model, the marginal slope is \(0.50\) outcome standard deviations per standard deviation of \(G\), while the specified coefficient holding \(E\) fixed is \(0.30\). The additional \(0.20\) arises from the environmental term under the model assumptions. This calculation does not prove that genotype caused the environment, or that either coefficient is causal in an observational setting.

Measuring a Correlation Is Not Identifying Its Pathway

A sample correlation can summarize how a measured genetic score and environmental exposure vary together. For six hypothetical individuals, suppose the observed pairs are \((G_i,E_i)=(0,1),(0,2),(1,2),(1,3),(2,3),(2,4)\). Their means are \(\bar{G}=1\) and \(\bar{E}=2.5\). The deviation cross-products are \(1.5, 0.5, 0, 0, 0.5, 1.5\), which sum to \(4\). The squared-deviation sums are \(4\) for \(G\) and \(5.5\) for \(E\).

Worked Example: Calculating a Sample Genotype–Environment Correlation

Using the six hypothetical pairs above, substitute the cross-product and squared-deviation sums into the sample-correlation formula:

r_{GE} = \frac{4}{\sqrt{4\cdot 5.5}} = \frac{4}{\sqrt{22}} \approx 0.8528.

The sample correlation is approximately \(0.8528\), a positive association in these six observations: larger \(G\) values tend to accompany larger \(E\) values. This is a descriptive result for the hypothetical sample, not evidence about a population p-value, the cause of the pattern, or whether it is passive, evocative, or active. The same numerical correlation could be consistent with different pathways.

Genotype–environment correlation is also distinct from gene–environment interaction. Correlation concerns whether \(G\) and \(E\) are associated. Interaction concerns whether the effect of one varies with the level of the other—for example, whether the outcome model needs a \(G\times E\) term. Correlation can occur without interaction, and interaction can occur even when \(G\) and \(E\) are uncorrelated.

Common Pitfalls and Interpretation

Calling every genetic association a direct genetic effect. A marginal association may include a contribution from an environmental exposure correlated with the genetic score. State what was modeled and which assumptions support the interpretation.
Inferring the pathway from the correlation alone. A positive \(r_{GE}\) does not distinguish passive, evocative, and active correlation. Evidence about family transmission, responses, or environmental selection is needed to assess the proposed route.
Confusing correlation with interaction. Correlated genetic and environmental measures do not automatically imply that the effect of genotype differs across environments.
Treating a genetic propensity as destiny. Genetic scores describe variation in a population and do not determine an individual’s environment or outcome.
Assuming “environmental” means unrelated to family genetics. Parental characteristics can be related to both a child’s inherited alleles and the child’s rearing environment, which is the central passive pathway.

Key takeaway: Passive, evocative, and active genotype–environment correlation describe different routes linking genetic variation to environmental exposure. Such correlation can contribute to an observed genetic association when the environment also relates to the outcome, but the correlation alone does not identify the route or establish a direct causal effect.

Check Your Understanding

Answer each question by identifying the relevant pathway or explaining what the available information does and does not establish.

Parents’ characteristics are associated with both the alleles they transmit and the learning resources they provide at home. Which form of genotype–environment correlation does this illustrate, and why?
A child’s genetically influenced behavior is associated with a change in how a coach responds to the child. Is this passive, evocative, or active correlation?
In the linear model \(Y=\beta_GG+\beta_EE+\varepsilon\), what additional term appears in the marginal slope for \(G\) when \(\operatorname{Cov}(G,\varepsilon)=0\)?
Can a positive sample correlation between a genetic score and an environment measure, by itself, show that the association is active rather than passive? Explain.
In one sentence, distinguish genotype–environment correlation from gene–environment interaction.